tests: new fsck tests for interval overlap

The following sequence could lead to this corruption:
  (1) Append new data to bulkdata
  (2) Update interval file positions in SQL
  (3) Flush (2)
  (4) Crash before flushing (1)
  (5) Reload database without running fsck
  (6) Start writing new data to end of bulkdata and introduce new interval

.coveragerc

@@ -0,0 +1,11 @@
+# -*- conf -*-
+
+[run]
+branch = True
+
+[report]
+exclude_lines =
+	pragma: no cover
+	if 0:
+omit = nilmdb/scripts,nilmdb/_version.py,nilmdb/fsck
+show_missing = True

.gitattributes

@@ -0,0 +1 @@
+nilmdb/_version.py export-subst

.gitignore

@@ -0,0 +1,24 @@
+# Tests
+tests/*testdb/
+.coverage
+db/
+
+# Compiled / cythonized files
+README.html
+docs/*.html
+build/
+*.pyc
+nilmdb/server/interval.c
+nilmdb/server/layout.c
+nilmdb/server/rbtree.c
+*.so
+
+# Setup junk
+dist/
+nilmdb.egg-info/
+venv/
+.eggs/
+
+# Misc
+timeit*out
+

MANIFEST.in

@@ -0,0 +1,29 @@
+# Root
+include README.txt
+include setup.cfg
+include setup.py
+include versioneer.py
+include Makefile
+include .coveragerc
+include .pylintrc
+include requirements.txt
+
+# Cython files -- include .pyx source, but not the generated .c files
+# (Downstream systems must have cython installed in order to build)
+recursive-include nilmdb/server *.pyx *.pyxdep *.pxd
+exclude nilmdb/server/interval.c
+exclude nilmdb/server/rbtree.c
+
+# Version
+include nilmdb/_version.py
+
+# Tests
+recursive-include tests *.py
+recursive-include tests/data *
+include tests/test.order
+
+# Docs
+recursive-include docs Makefile *.md
+
+# Extras
+recursive-include extras *

Makefile

@@ -1,2 +1,50 @@
-all:
+# By default, run the tests.
-	nosetests nilmdb/test_interval.py
+all: test
+
+version:
+	python3 setup.py version
+
+build:
+	python3 setup.py build_ext --inplace
+
+dist: sdist
+sdist:
+	python3 setup.py sdist
+
+install:
+	python3 setup.py install
+
+develop:
+	python3 setup.py develop
+
+docs:
+	make -C docs
+
+ctrl: flake
+flake:
+	flake8 nilmdb
+lint:
+	pylint3 --rcfile=setup.cfg nilmdb
+
+test:
+ifneq ($(INSIDE_EMACS),)
+# Use the slightly more flexible script
+	python3 setup.py build_ext --inplace
+	python3 tests/runtests.py
+else
+# Let setup.py check dependencies, build stuff, and run the test
+	python3 setup.py nosetests
+endif
+
+clean::
+	find . -name '*.pyc' -o -name '__pycache__' -print0 | xargs -0 rm -rf
+	rm -f .coverage
+	rm -rf tests/*testdb*
+	rm -rf nilmdb.egg-info/ build/ nilmdb/server/*.so
+	make -C docs clean
+
+gitclean::
+	git clean -dXf
+
+.PHONY: all version build dist sdist install docs test
+.PHONY: ctrl lint flake clean gitclean

README.md

@@ -0,0 +1,40 @@
+# nilmdb: Non-Intrusive Load Monitor Database
+by Jim Paris <jim@jtan.com>
+
+NilmDB requires Python 3.8 or newer.
+
+## Prerequisites:
+
+    # Runtime and build environments
+    sudo apt install python3 python3-dev python3-venv python3-pip
+
+    # Create a new Python virtual environment to isolate deps.
+    python3 -m venv ../venv
+    source ../venv/bin/activate   # run "deactivate" to leave
+
+    # Install all Python dependencies
+    pip3 install -r requirements.txt
+
+## Test:
+
+    python3 setup.py nosetests
+
+## Install:
+
+Install it into the virtual environment
+
+    python3 setup.py install
+
+If you want to instead install it system-wide, you will also need to
+install the requirements system-wide:
+
+    sudo pip3 install -r requirements.txt
+    sudo python3 setup.py install
+
+## Usage:
+
+    nilmdb-server --help
+    nilmdb-fsck --help
+    nilmtool --help
+
+See docs/wsgi.md for info on setting up a WSGI application in Apache.

README.txt

@@ -1,4 +0,0 @@
-To install, 
-
-   python seutp.py install
-

bin/nilm-test.py

@@ -1,26 +0,0 @@
-#!/usr/bin/python
-
-from nilmdb import Interval
-from optparse import OptionParser
-import sys
-
-version = "1.0"
-
-parser = OptionParser()
-parser.add_option("-d", "--db", dest="database", metavar="DATABASE",
-                  help="location of sqlite database")
-parser.add_option("-V", "--version", dest="version", default=False, action="store_true",
-                  help="print version then exit")
-
-(options, args) = parser.parse_args()
-
-if (options.version):
-    print "This script version: " + version
-    sys.exit(0)
-
-if options.database is None:
-    print "Error: database is mandatory"
-    sys.exit(1)
-
-print "Database is " + options.database
-

docs/Makefile

@@ -0,0 +1,9 @@
+ALL_DOCS = $(wildcard *.md)
+
+all: $(ALL_DOCS:.md=.html)
+
+%.html: %.md
+	pandoc -s $< > $@
+
+clean:
+	rm -f *.html

docs/TODO.md

@@ -0,0 +1,5 @@
+- Documentation
+
+- Machine-readable information in OverflowError, parser errors.
+  Maybe subclass `cherrypy.HTTPError` and override `set_response`
+  to add another JSON field?

docs/design.md

@@ -0,0 +1,469 @@
+Structure
+---------
+nilmdb.nilmdb is the NILM database interface.  A nilmdb.BulkData
+interface stores data in flat files, and a SQL database tracks
+metadata and ranges.
+
+Access to the nilmdb must be single-threaded.  This is handled with
+the nilmdb.serializer class.  In the future this could probably
+be turned into a per-path serialization.
+
+nilmdb.server is a HTTP server that provides an interface to talk,
+thorugh the serialization layer, to the nilmdb object.
+
+nilmdb.client is a HTTP client that connects to this.
+
+Sqlite performance
+------------------
+
+Committing a transaction in the default sync mode (PRAGMA synchronous=FULL)
+takes about 125msec.  sqlite3 will commit transactions at 3 times:
+
+1. explicit con.commit()
+
+2. between a series of DML commands and non-DML commands, e.g.
+   after a series of INSERT, SELECT, but before a CREATE TABLE or
+   PRAGMA.
+
+3. at the end of an explicit transaction, e.g. "with self.con as con:"
+
+To speed up testing, or if this transaction speed becomes an issue,
+the sync=False option to NilmDB will set PRAGMA synchronous=OFF.
+
+
+Inserting streams
+-----------------
+
+We need to send the contents of "data" as POST.  Do we need chunked
+transfer?
+
+- Don't know the size in advance, so we would need to use chunked if
+  we send the entire thing in one request.
+- But we shouldn't send one chunk per line, so we need to buffer some
+  anyway; why not just make new requests?
+- Consider the infinite-streaming case, we might want to send it
+  immediately?  Not really -- server still should do explicit inserts
+  of fixed-size chunks.
+- Even chunked encoding needs the size of each chunk beforehand, so
+  everything still gets buffered.  Just a tradeoff of buffer size.
+
+Before timestamps are added:
+
+- Raw data is about 440 kB/s    (9 channels)
+- Prep data is about 12.5 kB/s  (1 phase)
+- How do we know how much data to send?
+
+    - Remember that we can only do maybe 8-50 transactions per second on
+      the sqlite database.  So if one block of inserted data is one
+      transaction, we'd need the raw case to be around 64kB per request,
+      ideally more.
+    - Maybe use a range, based on how long it's taking to read the data
+        - If no more data, send it
+        - If data > 1 MB, send it
+    - If more than 10 seconds have elapsed, send it
+    - Should those numbers come from the server?
+
+Converting from ASCII to PyTables:
+
+- For each row getting added, we need to set attributes on a PyTables
+  Row object and call table.append().  This means that there isn't a
+  particularly efficient way of converting from ascii.
+- Could create a function like nilmdb.layout.Layout("foo".fillRow(asciiline)
+    - But this means we're doing parsing on the serialized side
+    - Let's keep parsing on the threaded server side so we can detect
+      errors better, and not block the serialized nilmdb for a slow
+      parsing process.
+- Client sends ASCII data
+- Server converts this ACSII data to a list of values
+    - Maybe:
+
+            # threaded side creates this object
+            parser = nilmdb.layout.Parser("layout_name")
+            # threaded side parses and fills it with data
+            parser.parse(textdata)
+            # serialized side pulls out rows
+            for n in xrange(parser.nrows):
+                parser.fill_row(rowinstance, n)
+                table.append()
+
+
+Inserting streams, inside nilmdb
+--------------------------------
+
+- First check that the new stream doesn't overlap.
+    - Get minimum timestamp, maximum timestamp from data parser.
+        - (extend parser to verify monotonicity and track extents)
+    - Get all intervals for this stream in the database
+    - See if new interval overlaps any existing ones
+        - If so, bail
+    - Question: should we cache intervals inside NilmDB?
+        - Assume database is fast for now, and always rebuild fom DB.
+        - Can add a caching layer later if we need to.
+    - `stream_get_ranges(path)` -> return IntervalSet?
+
+Speed
+-----
+
+- First approach was quadratic.  Adding four hours of data:
+
+        $ time zcat /home/jim/bpnilm-data/snapshot-1-20110513-110002.raw.gz | ./nilmtool.py insert -s 20110513-110000 /bpnilm/1/raw
+        real    24m31.093s
+        $ time zcat /home/jim/bpnilm-data/snapshot-1-20110513-110002.raw.gz | ./nilmtool.py insert -s 20110513-120001 /bpnilm/1/raw
+        real    43m44.528s
+        $ time zcat /home/jim/bpnilm-data/snapshot-1-20110513-110002.raw.gz | ./nilmtool.py insert -s 20110513-130002 /bpnilm/1/raw
+        real    93m29.713s
+        $ time zcat /home/jim/bpnilm-data/snapshot-1-20110513-110002.raw.gz | ./nilmtool.py insert -s 20110513-140003 /bpnilm/1/raw
+        real    166m53.007s
+
+- Disabling pytables indexing didn't help:
+
+        real    31m21.492s
+        real    52m51.963s
+        real    102m8.151s
+        real    176m12.469s
+
+- Server RAM usage is constant.
+
+- Speed problems were due to IntervalSet speed, of parsing intervals
+  from the database and adding the new one each time.
+
+    - First optimization is to cache result of `nilmdb:_get_intervals`,
+      which gives the best speedup.
+
+    - Also switched to internally using bxInterval from bx-python package.
+      Speed of `tests/test_interval:TestIntervalSpeed` is pretty decent
+      and seems to be growing logarithmically now.  About 85μs per insertion
+      for inserting 131k entries.
+
+    - Storing the interval data in SQL might be better, with a scheme like:
+      http://www.logarithmic.net/pfh/blog/01235197474
+
+- Next slowdown target is nilmdb.layout.Parser.parse().
+    - Rewrote parsers using cython and sscanf
+    - Stats (rev 10831), with `_add_interval` disabled
+
+        layout.pyx.Parser.parse:128        6303 sec, 262k calls
+         layout.pyx.parse:63               13913 sec, 5.1g calls
+        numpy:records.py.fromrecords:569   7410 sec, 262k calls
+
+- Probably OK for now.
+
+- After all updates, now takes about 8.5 minutes to insert an hour of
+  data, constant after adding 171 hours (4.9 billion data points)
+
+- Data set size: 98 gigs = 20 bytes per data point.
+  6 uint16 data + 1 uint32 timestamp = 16 bytes per point
+  So compression must be off -- will retry with compression forced on.
+
+IntervalSet speed
+-----------------
+- Initial implementation was pretty slow, even with binary search in
+  sorted list
+
+- Replaced with bxInterval; now takes about log n time for an insertion
+    - TestIntervalSpeed with range(17,18) and profiling
+        - 85 μs each
+        - 131072 calls to `__iadd__`
+        - 131072 to bx.insert_interval
+        - 131072 to bx.insert:395
+        - 2355835 to bx.insert:106  (18x as many?)
+
+- Tried blist too, worse than bxinterval.
+
+- Might be algorithmic improvements to be made in Interval.py,
+  like in `__and__`
+
+- Replaced again with rbtree.  Seems decent.  Numbers are time per
+  insert for 2**17 insertions, followed by total wall time and RAM
+  usage for running "make test" with `test_rbtree` and `test_interval`
+  with range(5,20):
+    - old values with bxinterval:
+      20.2 μS, total 20 s, 177 MB RAM
+    - rbtree, plain python:
+      97 μS, total 105 s, 846 MB RAM
+    - rbtree converted to cython:
+      26 μS, total 29 s, 320 MB RAM
+    - rbtree and interval converted to cython:
+      8.4 μS, total 12 s, 134 MB RAM
+
+- Would like to move Interval itself back to Python so other
+  non-cythonized code like client code can use it more easily.
+  Testing speed with just `test_interval` being tested, with
+  `range(5,22)`, using `/usr/bin/time -v python tests/runtests.py`,
+  times recorded for 2097152:
+    - 52ae397 (Interval in cython):
+	  12.6133 μs each, ratio 0.866533, total 47 sec, 399 MB RAM
+	- 9759dcf (Interval in python):
+	  21.2937 μs each, ratio 1.462870, total 83 sec, 1107 MB RAM
+  That's a huge difference!  Instead, will keep Interval and DBInterval
+  cythonized inside nilmdb, and just have an additional copy in
+  nilmdb.utils for clients to use.
+
+Layouts
+-------
+Current/old design has specific layouts: RawData, PrepData, RawNotchedData.
+Let's get rid of this entirely and switch to simpler data types that are
+just collections and counts of a single type.  We'll still use strings
+to describe them, with format:
+
+    type_count
+
+where type is "uint16", "float32", or "float64", and count is an integer.
+
+nilmdb.layout.named() will parse these strings into the appropriate
+handlers.  For compatibility:
+
+    "RawData" == "uint16_6"
+    "RawNotchedData" == "uint16_9"
+    "PrepData" == "float32_8"
+
+
+BulkData design
+---------------
+
+BulkData is a custom bulk data storage system that was written to
+replace PyTables.  The general structure is a `data` subdirectory in
+the main NilmDB directory.  Within `data`, paths are created for each
+created stream.  These locations are called tables.  For example,
+tables might be located at
+
+    nilmdb/data/newton/raw/
+    nilmdb/data/newton/prep/
+    nilmdb/data/cottage/raw/
+
+Each table contains:
+
+- An unchanging `_format` file (Python pickle format) that describes
+  parameters of how the data is broken up, like files per directory,
+  rows per file, and the binary data format
+
+- Hex named subdirectories `("%04x", although more than 65536 can exist)`
+
+- Hex named files within those subdirectories, like:
+
+        /nilmdb/data/newton/raw/000b/010a
+
+    The data format of these files is raw binary, interpreted by the
+    Python `struct` module according to the format string in the
+    `_format` file.
+
+- Same as above, with `.removed` suffix, is an optional file (Python
+  pickle format) containing a list of row numbers that have been
+  logically removed from the file.  If this range covers the entire
+  file, the entire file will be removed.
+
+- Note that the `bulkdata.nrows` variable is calculated once in
+  `BulkData.__init__()`, and only ever incremented during use.  Thus,
+  even if all data is removed, `nrows` can remain high.  However, if
+  the server is restarted, the newly calculated `nrows` may be lower
+  than in a previous run due to deleted data.  To be specific, this
+  sequence of events:
+
+    - insert data
+    - remove all data
+    - insert data
+
+    will result in having different row numbers in the database, and
+    differently numbered files on the filesystem, than the sequence:
+
+    - insert data
+    - remove all data
+    - restart server
+    - insert data
+
+    This is okay!  Everything should remain consistent both in the
+    `BulkData` and `NilmDB`.  Not attempting to readjust `nrows` during
+    deletion makes the code quite a bit simpler.
+
+- Similarly, data files are never truncated shorter.  Removing data
+  from the end of the file will not shorten it; it will only be
+  deleted when it has been fully filled and all of the data has been
+  subsequently removed.
+
+
+Rocket
+------
+
+Original design had the nilmdb.nilmdb thread (through bulkdata)
+convert from on-disk layout to a Python list, and then the
+nilmdb.server thread (from cherrypy) converts to ASCII.  For at least
+the extraction side of things, it's easy to pass the bulkdata a layout
+name instead, and have it convert directly from on-disk to ASCII
+format, because this conversion can then be shoved into a C module.
+This module, which provides a means for converting directly from
+on-disk format to ASCII or Python lists, is the "rocket" interface.
+Python is still used to manage the files and figure out where the
+data should go; rocket just puts binary data directly in or out of
+those files at specified locations.
+
+Before rocket, testing speed with uint16_6 data, with an end-to-end
+test (extracting data with nilmtool):
+
+- insert: 65 klines/sec
+- extract: 120 klines/sec
+
+After switching to the rocket design, but using the Python version
+(pyrocket):
+
+- insert: 57 klines/sec
+- extract: 120 klines/sec
+
+After switching to a C extension module (rocket.c)
+
+- insert: 74 klines/sec through insert.py; 99.6 klines/sec through nilmtool
+- extract: 335 klines/sec
+
+After client block updates (described below):
+
+- insert: 180 klines/sec through nilmtool (pre-timestamped)
+- extract: 390 klines/sec through nilmtool
+
+Using "insert --timestamp" or "extract --bare" cuts the speed in half.
+
+Blocks versus lines
+-------------------
+
+Generally want to avoid parsing the bulk of the data as lines if
+possible, and transfer things in bigger blocks at once.
+
+Current places where we use lines:
+
+- All data returned by `client.stream_extract`, since it comes from
+  `httpclient.get_gen`, which iterates over lines.  Not sure if this
+  should be changed, because a `nilmtool extract` is just about the
+  same speed as `curl -q .../stream/extract`!
+
+- `client.StreamInserter.insert_iter` and
+  `client.StreamInserter.insert_line`, which should probably get
+  replaced with block versions.  There's no real need to keep
+  updating the timestamp every time we get a new line of data.
+
+  - Finished.  Just a single insert() that takes any length string and
+    does very little processing until it's time to send it to the
+	server.
+
+Timestamps
+----------
+
+Timestamps are currently double-precision floats (64 bit).  Since the
+mantissa is 53-bit, this can only represent about 15-17 significant
+figures, and microsecond Unix timestamps like 1222333444.000111 are
+already 16 significant figures.  Rounding is therefore an issue;
+it's hard to sure that converting from ASCII, then back to ASCII,
+will always give the same result.
+
+Also, if the client provides a floating point value like 1.9999999999,
+we need to be careful that we don't store it as 1.9999999999 but later
+print it as 2.000000, because then round-trips change the data.
+
+Possible solutions:
+
+- When the client provides a floating point value to the server,
+  always round to the 6th decimal digit before verifying & storing.
+  Good for compatibility and simplicity.  But still might have rounding
+  issues, and clients will also need to round when doing their own
+  verification.  Having every piece of code need to know which digit
+  to round at is not ideal.
+
+- Always store int64 timestamps on the server, representing
+  microseconds since epoch.  int64 timestamps are used in all HTTP
+  parameters, in insert/extract ASCII strings, client API, commandline
+  raw timestamps, etc.  Pretty big change.
+
+  This is what we'll go with...
+
+  - Client programs that interpret the timestamps as doubles instead
+    of ints will remain accurate until 2^53 microseconds, or year
+    2255.
+
+  - On insert, maybe it's OK to send floating point microsecond values
+    (1234567890123456.0), just to cope with clients that want to print
+    everything as a double.  Server could try parsing as int64, and if
+    that fails, parse as double and truncate to int64.  However, this
+    wouldn't catch imprecise inputs like "1.23456789012e+15".  But
+    maybe that can just be ignored; it's likely to cause a
+    non-monotonic error at the client.
+
+  - Timestamps like 1234567890.123456 never show up anywhere, except
+    for interfacing to datetime_tz etc.  Command line "raw timestamps"
+    are always printed as int64 values, and a new format
+    "@1234567890123456" is added to the parser for specifying them
+    exactly.
+
+Binary interface
+----------------
+
+The ASCII interface is too slow for high-bandwidth processing, like
+sinefits, prep, etc.  A binary interface was added so that you can
+extract the raw binary out of the bulkdata storage.  This binary is
+a little-endian format, e.g. in C a uint16_6 stream would be:
+
+    #include <endian.h>
+    #include <stdint.h>
+    struct {
+        int64_t timestamp_le;
+        uint16_t data_le[6];
+    } __attribute__((packed));
+
+Remember to byteswap (with e.g. `letoh` in C)!
+
+This interface is used by the new `nilmdb.client.numpyclient.NumpyClient`
+class, which is a subclass of the normal `nilmcb.client.client.Client`
+and has all of the same functions.  It adds three new functions:
+
+- `stream_extract_numpy` to extract data as a Numpy array
+
+- `stream_insert_numpy` to insert data as a Numpy array
+
+- `stream_insert_numpy_context` is the context manager for
+  incrementally inserting data
+
+It is significantly faster!  It is about 20 times faster to decimate a
+stream with `nilm-decimate` when the filter code is using the new
+binary/numpy interface.
+
+
+WSGI interface & chunked requests
+---------------------------------
+
+mod_wsgi requires "WSGIChunkedRequest On" to handle
+"Transfer-encoding: Chunked" requests.  However, `/stream/insert`
+doesn't handle this correctly right now, because:
+
+- The `cherrypy.request.body.read()` call needs to be fixed for chunked requests
+
+- We don't want to just buffer endlessly in the server, and it will
+  require some thought on how to handle data in chunks (what to do about
+  interval endpoints).
+
+It is probably better to just keep the endpoint management on the client
+side, so leave "WSGIChunkedRequest off" for now.
+
+
+Unicode & character encoding
+----------------------------
+
+Stream data is passed back and forth as raw `bytes` objects in most
+places, including the `nilmdb.client` and command-line interfaces.
+This is done partially for performance reasons, and partially to
+support the binary insert/extract options, where character-set encoding
+would not apply.
+
+For the HTTP server, the raw bytes transferred over HTTP are interpreted
+as follows:
+- For `/stream/insert`, the client-provided `Content-Type` is ignored,
+  and the data is read as if it were `application/octet-stream`.
+- For `/stream/extract`, the returned data is `application/octet-stream`.
+- All other endpoints communicate via JSON, which is specified to always
+  be encoded as UTF-8.  This includes:
+    - `/version`
+    - `/dbinfo`
+    - `/stream/list`
+    - `/stream/create`
+    - `/stream/destroy`
+    - `/stream/rename`
+    - `/stream/get_metadata`
+    - `/stream/set_metadata`
+    - `/stream/update_metadata`
+    - `/stream/remove`
+    - `/stream/intervals`

docs/wsgi.md

@@ -0,0 +1,32 @@
+WSGI Application in Apache
+--------------------------
+
+Install `apache2` and `libapache2-mod-wsgi`
+
+We'll set up the database server at URL `http://myhost.com/nilmdb`.
+The database will be stored in `/home/nilm/db`, and the process will
+run as user `nilm`, group `nilm`.
+
+First, create a WSGI script `/home/nilm/nilmdb.wsgi` containing:
+
+    import nilmdb.server
+    application = nilmdb.server.wsgi_application("/home/nilm/db", "/nilmdb")
+
+The first parameter is the local filesystem path, and the second
+parameter is the path part of the URL.
+
+Then, set up Apache with a configuration like:
+
+    <VirtualHost>
+        WSGIScriptAlias /nilmdb /home/nilm/nilmdb.wsgi
+        WSGIDaemonProcess nilmdb-procgroup threads=32 user=nilm group=nilm
+        <Location /nilmdb>
+            WSGIProcessGroup nilmdb-procgroup
+            WSGIApplicationGroup nilmdb-appgroup
+
+            # Access control example:
+            Order deny,allow
+            Deny from all
+            Allow from 1.2.3.4
+        </Location>
+    </VirtualHost>

extras/fix-oversize-files.py

@@ -0,0 +1,50 @@
+#!/usr/bin/env python3
+
+import os
+import sys
+import pickle
+import argparse
+import fcntl
+import re
+from nilmdb.client.numpyclient import layout_to_dtype
+
+parser = argparse.ArgumentParser(
+    description = """
+Fix database corruption where binary writes caused too much data to be
+written to the file.  Truncates files to the correct length.  This was
+fixed by b98ff1331a515ad47fd3203615e835b529b039f9.
+""")
+parser.add_argument("path", action="store", help='Database root path')
+parser.add_argument("-y", "--yes", action="store_true", help='Fix them')
+args = parser.parse_args()
+
+lock = os.path.join(args.path, "data.lock")
+with open(lock, "w") as f:
+    fcntl.flock(f.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB)
+
+    fix = {}
+
+    for (path, dirs, files) in os.walk(args.path):
+        if "_format" in files:
+            with open(os.path.join(path, "_format")) as format:
+                fmt = pickle.load(format)
+                rowsize = layout_to_dtype(fmt["layout"]).itemsize
+                maxsize = rowsize * fmt["rows_per_file"]
+                fix[path] = maxsize
+                if maxsize < 128000000: # sanity check
+                    raise Exception("bad maxsize " + str(maxsize))
+
+    for fixpath in fix:
+        for (path, dirs, files) in os.walk(fixpath):
+            for fn in files:
+                if not re.match("^[0-9a-f]{4,}$", fn):
+                    continue
+                fn = os.path.join(path, fn)
+                size = os.path.getsize(fn)
+                maxsize = fix[fixpath]
+                if size > maxsize:
+                    diff = size - maxsize
+                    print(diff, "too big:", fn)
+                    if args.yes:
+                        with open(fn, "a+") as dbfile:
+                            dbfile.truncate(maxsize)

extras/nilmtool-bash-completion.sh

@@ -0,0 +1,20 @@
+# To enable bash completion:
+#
+# 1. Ensure python-argcomplete is installed:
+#       pip install argcomplete
+# 2. Source this file:
+#       . nilmtool-bash-completion.sh
+
+_nilmtool_argcomplete() {
+    local IFS=$(printf "\013")
+    COMPREPLY=( $(IFS="$IFS" \
+                  COMP_LINE="$COMP_LINE" \
+	          COMP_WORDBREAKS="$COMP_WORDBREAKS" \
+                  COMP_POINT="$COMP_POINT" \
+                  _ARGCOMPLETE=1 \
+                  "$1" 8>&1 9>&2 1>/dev/null 2>/dev/null) )
+    if [[ $? != 0 ]]; then
+        unset COMPREPLY
+    fi
+}
+complete -o nospace -F _nilmtool_argcomplete nilmtool

nilmdb/__init__.py

@@ -1,2 +1,5 @@
-from nilmdb.interval import *
+"""Main NilmDB import"""
-from nilmdb.fileinterval import *
+
+from ._version import get_versions
+__version__ = get_versions()['version']
+del get_versions

nilmdb/_version.py

@@ -0,0 +1,520 @@
+
+# This file helps to compute a version number in source trees obtained from
+# git-archive tarball (such as those provided by githubs download-from-tag
+# feature). Distribution tarballs (built by setup.py sdist) and build
+# directories (produced by setup.py build) will contain a much shorter file
+# that just contains the computed version number.
+
+# This file is released into the public domain. Generated by
+# versioneer-0.18 (https://github.com/warner/python-versioneer)
+
+"""Git implementation of _version.py."""
+
+import errno
+import os
+import re
+import subprocess
+import sys
+
+
+def get_keywords():
+    """Get the keywords needed to look up the version information."""
+    # these strings will be replaced by git during git-archive.
+    # setup.py/versioneer.py will grep for the variable names, so they must
+    # each be defined on a line of their own. _version.py will just call
+    # get_keywords().
+    git_refnames = "$Format:%d$"
+    git_full = "$Format:%H$"
+    git_date = "$Format:%ci$"
+    keywords = {"refnames": git_refnames, "full": git_full, "date": git_date}
+    return keywords
+
+
+class VersioneerConfig:
+    """Container for Versioneer configuration parameters."""
+
+
+def get_config():
+    """Create, populate and return the VersioneerConfig() object."""
+    # these strings are filled in when 'setup.py versioneer' creates
+    # _version.py
+    cfg = VersioneerConfig()
+    cfg.VCS = "git"
+    cfg.style = "pep440"
+    cfg.tag_prefix = "nilmdb-"
+    cfg.parentdir_prefix = "nilmdb-"
+    cfg.versionfile_source = "nilmdb/_version.py"
+    cfg.verbose = False
+    return cfg
+
+
+class NotThisMethod(Exception):
+    """Exception raised if a method is not valid for the current scenario."""
+
+
+LONG_VERSION_PY = {}
+HANDLERS = {}
+
+
+def register_vcs_handler(vcs, method):  # decorator
+    """Decorator to mark a method as the handler for a particular VCS."""
+    def decorate(f):
+        """Store f in HANDLERS[vcs][method]."""
+        if vcs not in HANDLERS:
+            HANDLERS[vcs] = {}
+        HANDLERS[vcs][method] = f
+        return f
+    return decorate
+
+
+def run_command(commands, args, cwd=None, verbose=False, hide_stderr=False,
+                env=None):
+    """Call the given command(s)."""
+    assert isinstance(commands, list)
+    p = None
+    for c in commands:
+        try:
+            dispcmd = str([c] + args)
+            # remember shell=False, so use git.cmd on windows, not just git
+            p = subprocess.Popen([c] + args, cwd=cwd, env=env,
+                                 stdout=subprocess.PIPE,
+                                 stderr=(subprocess.PIPE if hide_stderr
+                                         else None))
+            break
+        except EnvironmentError:
+            e = sys.exc_info()[1]
+            if e.errno == errno.ENOENT:
+                continue
+            if verbose:
+                print("unable to run %s" % dispcmd)
+                print(e)
+            return None, None
+    else:
+        if verbose:
+            print("unable to find command, tried %s" % (commands,))
+        return None, None
+    stdout = p.communicate()[0].strip()
+    if sys.version_info[0] >= 3:
+        stdout = stdout.decode()
+    if p.returncode != 0:
+        if verbose:
+            print("unable to run %s (error)" % dispcmd)
+            print("stdout was %s" % stdout)
+        return None, p.returncode
+    return stdout, p.returncode
+
+
+def versions_from_parentdir(parentdir_prefix, root, verbose):
+    """Try to determine the version from the parent directory name.
+
+    Source tarballs conventionally unpack into a directory that includes both
+    the project name and a version string. We will also support searching up
+    two directory levels for an appropriately named parent directory
+    """
+    rootdirs = []
+
+    for i in range(3):
+        dirname = os.path.basename(root)
+        if dirname.startswith(parentdir_prefix):
+            return {"version": dirname[len(parentdir_prefix):],
+                    "full-revisionid": None,
+                    "dirty": False, "error": None, "date": None}
+        else:
+            rootdirs.append(root)
+            root = os.path.dirname(root)  # up a level
+
+    if verbose:
+        print("Tried directories %s but none started with prefix %s" %
+              (str(rootdirs), parentdir_prefix))
+    raise NotThisMethod("rootdir doesn't start with parentdir_prefix")
+
+
+@register_vcs_handler("git", "get_keywords")
+def git_get_keywords(versionfile_abs):
+    """Extract version information from the given file."""
+    # the code embedded in _version.py can just fetch the value of these
+    # keywords. When used from setup.py, we don't want to import _version.py,
+    # so we do it with a regexp instead. This function is not used from
+    # _version.py.
+    keywords = {}
+    try:
+        f = open(versionfile_abs, "r")
+        for line in f.readlines():
+            if line.strip().startswith("git_refnames ="):
+                mo = re.search(r'=\s*"(.*)"', line)
+                if mo:
+                    keywords["refnames"] = mo.group(1)
+            if line.strip().startswith("git_full ="):
+                mo = re.search(r'=\s*"(.*)"', line)
+                if mo:
+                    keywords["full"] = mo.group(1)
+            if line.strip().startswith("git_date ="):
+                mo = re.search(r'=\s*"(.*)"', line)
+                if mo:
+                    keywords["date"] = mo.group(1)
+        f.close()
+    except EnvironmentError:
+        pass
+    return keywords
+
+
+@register_vcs_handler("git", "keywords")
+def git_versions_from_keywords(keywords, tag_prefix, verbose):
+    """Get version information from git keywords."""
+    if not keywords:
+        raise NotThisMethod("no keywords at all, weird")
+    date = keywords.get("date")
+    if date is not None:
+        # git-2.2.0 added "%cI", which expands to an ISO-8601 -compliant
+        # datestamp. However we prefer "%ci" (which expands to an "ISO-8601
+        # -like" string, which we must then edit to make compliant), because
+        # it's been around since git-1.5.3, and it's too difficult to
+        # discover which version we're using, or to work around using an
+        # older one.
+        date = date.strip().replace(" ", "T", 1).replace(" ", "", 1)
+    refnames = keywords["refnames"].strip()
+    if refnames.startswith("$Format"):
+        if verbose:
+            print("keywords are unexpanded, not using")
+        raise NotThisMethod("unexpanded keywords, not a git-archive tarball")
+    refs = set([r.strip() for r in refnames.strip("()").split(",")])
+    # starting in git-1.8.3, tags are listed as "tag: foo-1.0" instead of
+    # just "foo-1.0". If we see a "tag: " prefix, prefer those.
+    TAG = "tag: "
+    tags = set([r[len(TAG):] for r in refs if r.startswith(TAG)])
+    if not tags:
+        # Either we're using git < 1.8.3, or there really are no tags. We use
+        # a heuristic: assume all version tags have a digit. The old git %d
+        # expansion behaves like git log --decorate=short and strips out the
+        # refs/heads/ and refs/tags/ prefixes that would let us distinguish
+        # between branches and tags. By ignoring refnames without digits, we
+        # filter out many common branch names like "release" and
+        # "stabilization", as well as "HEAD" and "master".
+        tags = set([r for r in refs if re.search(r'\d', r)])
+        if verbose:
+            print("discarding '%s', no digits" % ",".join(refs - tags))
+    if verbose:
+        print("likely tags: %s" % ",".join(sorted(tags)))
+    for ref in sorted(tags):
+        # sorting will prefer e.g. "2.0" over "2.0rc1"
+        if ref.startswith(tag_prefix):
+            r = ref[len(tag_prefix):]
+            if verbose:
+                print("picking %s" % r)
+            return {"version": r,
+                    "full-revisionid": keywords["full"].strip(),
+                    "dirty": False, "error": None,
+                    "date": date}
+    # no suitable tags, so version is "0+unknown", but full hex is still there
+    if verbose:
+        print("no suitable tags, using unknown + full revision id")
+    return {"version": "0+unknown",
+            "full-revisionid": keywords["full"].strip(),
+            "dirty": False, "error": "no suitable tags", "date": None}
+
+
+@register_vcs_handler("git", "pieces_from_vcs")
+def git_pieces_from_vcs(tag_prefix, root, verbose, run_command=run_command):
+    """Get version from 'git describe' in the root of the source tree.
+
+    This only gets called if the git-archive 'subst' keywords were *not*
+    expanded, and _version.py hasn't already been rewritten with a short
+    version string, meaning we're inside a checked out source tree.
+    """
+    GITS = ["git"]
+    if sys.platform == "win32":
+        GITS = ["git.cmd", "git.exe"]
+
+    out, rc = run_command(GITS, ["rev-parse", "--git-dir"], cwd=root,
+                          hide_stderr=True)
+    if rc != 0:
+        if verbose:
+            print("Directory %s not under git control" % root)
+        raise NotThisMethod("'git rev-parse --git-dir' returned error")
+
+    # if there is a tag matching tag_prefix, this yields TAG-NUM-gHEX[-dirty]
+    # if there isn't one, this yields HEX[-dirty] (no NUM)
+    describe_out, rc = run_command(GITS, ["describe", "--tags", "--dirty",
+                                          "--always", "--long",
+                                          "--match", "%s*" % tag_prefix],
+                                   cwd=root)
+    # --long was added in git-1.5.5
+    if describe_out is None:
+        raise NotThisMethod("'git describe' failed")
+    describe_out = describe_out.strip()
+    full_out, rc = run_command(GITS, ["rev-parse", "HEAD"], cwd=root)
+    if full_out is None:
+        raise NotThisMethod("'git rev-parse' failed")
+    full_out = full_out.strip()
+
+    pieces = {}
+    pieces["long"] = full_out
+    pieces["short"] = full_out[:7]  # maybe improved later
+    pieces["error"] = None
+
+    # parse describe_out. It will be like TAG-NUM-gHEX[-dirty] or HEX[-dirty]
+    # TAG might have hyphens.
+    git_describe = describe_out
+
+    # look for -dirty suffix
+    dirty = git_describe.endswith("-dirty")
+    pieces["dirty"] = dirty
+    if dirty:
+        git_describe = git_describe[:git_describe.rindex("-dirty")]
+
+    # now we have TAG-NUM-gHEX or HEX
+
+    if "-" in git_describe:
+        # TAG-NUM-gHEX
+        mo = re.search(r'^(.+)-(\d+)-g([0-9a-f]+)$', git_describe)
+        if not mo:
+            # unparseable. Maybe git-describe is misbehaving?
+            pieces["error"] = ("unable to parse git-describe output: '%s'"
+                               % describe_out)
+            return pieces
+
+        # tag
+        full_tag = mo.group(1)
+        if not full_tag.startswith(tag_prefix):
+            if verbose:
+                fmt = "tag '%s' doesn't start with prefix '%s'"
+                print(fmt % (full_tag, tag_prefix))
+            pieces["error"] = ("tag '%s' doesn't start with prefix '%s'"
+                               % (full_tag, tag_prefix))
+            return pieces
+        pieces["closest-tag"] = full_tag[len(tag_prefix):]
+
+        # distance: number of commits since tag
+        pieces["distance"] = int(mo.group(2))
+
+        # commit: short hex revision ID
+        pieces["short"] = mo.group(3)
+
+    else:
+        # HEX: no tags
+        pieces["closest-tag"] = None
+        count_out, rc = run_command(GITS, ["rev-list", "HEAD", "--count"],
+                                    cwd=root)
+        pieces["distance"] = int(count_out)  # total number of commits
+
+    # commit date: see ISO-8601 comment in git_versions_from_keywords()
+    date = run_command(GITS, ["show", "-s", "--format=%ci", "HEAD"],
+                       cwd=root)[0].strip()
+    pieces["date"] = date.strip().replace(" ", "T", 1).replace(" ", "", 1)
+
+    return pieces
+
+
+def plus_or_dot(pieces):
+    """Return a + if we don't already have one, else return a ."""
+    if "+" in pieces.get("closest-tag", ""):
+        return "."
+    return "+"
+
+
+def render_pep440(pieces):
+    """Build up version string, with post-release "local version identifier".
+
+    Our goal: TAG[+DISTANCE.gHEX[.dirty]] . Note that if you
+    get a tagged build and then dirty it, you'll get TAG+0.gHEX.dirty
+
+    Exceptions:
+    1: no tags. git_describe was just HEX. 0+untagged.DISTANCE.gHEX[.dirty]
+    """
+    if pieces["closest-tag"]:
+        rendered = pieces["closest-tag"]
+        if pieces["distance"] or pieces["dirty"]:
+            rendered += plus_or_dot(pieces)
+            rendered += "%d.g%s" % (pieces["distance"], pieces["short"])
+            if pieces["dirty"]:
+                rendered += ".dirty"
+    else:
+        # exception #1
+        rendered = "0+untagged.%d.g%s" % (pieces["distance"],
+                                          pieces["short"])
+        if pieces["dirty"]:
+            rendered += ".dirty"
+    return rendered
+
+
+def render_pep440_pre(pieces):
+    """TAG[.post.devDISTANCE] -- No -dirty.
+
+    Exceptions:
+    1: no tags. 0.post.devDISTANCE
+    """
+    if pieces["closest-tag"]:
+        rendered = pieces["closest-tag"]
+        if pieces["distance"]:
+            rendered += ".post.dev%d" % pieces["distance"]
+    else:
+        # exception #1
+        rendered = "0.post.dev%d" % pieces["distance"]
+    return rendered
+
+
+def render_pep440_post(pieces):
+    """TAG[.postDISTANCE[.dev0]+gHEX] .
+
+    The ".dev0" means dirty. Note that .dev0 sorts backwards
+    (a dirty tree will appear "older" than the corresponding clean one),
+    but you shouldn't be releasing software with -dirty anyways.
+
+    Exceptions:
+    1: no tags. 0.postDISTANCE[.dev0]
+    """
+    if pieces["closest-tag"]:
+        rendered = pieces["closest-tag"]
+        if pieces["distance"] or pieces["dirty"]:
+            rendered += ".post%d" % pieces["distance"]
+            if pieces["dirty"]:
+                rendered += ".dev0"
+            rendered += plus_or_dot(pieces)
+            rendered += "g%s" % pieces["short"]
+    else:
+        # exception #1
+        rendered = "0.post%d" % pieces["distance"]
+        if pieces["dirty"]:
+            rendered += ".dev0"
+        rendered += "+g%s" % pieces["short"]
+    return rendered
+
+
+def render_pep440_old(pieces):
+    """TAG[.postDISTANCE[.dev0]] .
+
+    The ".dev0" means dirty.
+
+    Eexceptions:
+    1: no tags. 0.postDISTANCE[.dev0]
+    """
+    if pieces["closest-tag"]:
+        rendered = pieces["closest-tag"]
+        if pieces["distance"] or pieces["dirty"]:
+            rendered += ".post%d" % pieces["distance"]
+            if pieces["dirty"]:
+                rendered += ".dev0"
+    else:
+        # exception #1
+        rendered = "0.post%d" % pieces["distance"]
+        if pieces["dirty"]:
+            rendered += ".dev0"
+    return rendered
+
+
+def render_git_describe(pieces):
+    """TAG[-DISTANCE-gHEX][-dirty].
+
+    Like 'git describe --tags --dirty --always'.
+
+    Exceptions:
+    1: no tags. HEX[-dirty]  (note: no 'g' prefix)
+    """
+    if pieces["closest-tag"]:
+        rendered = pieces["closest-tag"]
+        if pieces["distance"]:
+            rendered += "-%d-g%s" % (pieces["distance"], pieces["short"])
+    else:
+        # exception #1
+        rendered = pieces["short"]
+    if pieces["dirty"]:
+        rendered += "-dirty"
+    return rendered
+
+
+def render_git_describe_long(pieces):
+    """TAG-DISTANCE-gHEX[-dirty].
+
+    Like 'git describe --tags --dirty --always -long'.
+    The distance/hash is unconditional.
+
+    Exceptions:
+    1: no tags. HEX[-dirty]  (note: no 'g' prefix)
+    """
+    if pieces["closest-tag"]:
+        rendered = pieces["closest-tag"]
+        rendered += "-%d-g%s" % (pieces["distance"], pieces["short"])
+    else:
+        # exception #1
+        rendered = pieces["short"]
+    if pieces["dirty"]:
+        rendered += "-dirty"
+    return rendered
+
+
+def render(pieces, style):
+    """Render the given version pieces into the requested style."""
+    if pieces["error"]:
+        return {"version": "unknown",
+                "full-revisionid": pieces.get("long"),
+                "dirty": None,
+                "error": pieces["error"],
+                "date": None}
+
+    if not style or style == "default":
+        style = "pep440"  # the default
+
+    if style == "pep440":
+        rendered = render_pep440(pieces)
+    elif style == "pep440-pre":
+        rendered = render_pep440_pre(pieces)
+    elif style == "pep440-post":
+        rendered = render_pep440_post(pieces)
+    elif style == "pep440-old":
+        rendered = render_pep440_old(pieces)
+    elif style == "git-describe":
+        rendered = render_git_describe(pieces)
+    elif style == "git-describe-long":
+        rendered = render_git_describe_long(pieces)
+    else:
+        raise ValueError("unknown style '%s'" % style)
+
+    return {"version": rendered, "full-revisionid": pieces["long"],
+            "dirty": pieces["dirty"], "error": None,
+            "date": pieces.get("date")}
+
+
+def get_versions():
+    """Get version information or return default if unable to do so."""
+    # I am in _version.py, which lives at ROOT/VERSIONFILE_SOURCE. If we have
+    # __file__, we can work backwards from there to the root. Some
+    # py2exe/bbfreeze/non-CPython implementations don't do __file__, in which
+    # case we can only use expanded keywords.
+
+    cfg = get_config()
+    verbose = cfg.verbose
+
+    try:
+        return git_versions_from_keywords(get_keywords(), cfg.tag_prefix,
+                                          verbose)
+    except NotThisMethod:
+        pass
+
+    try:
+        root = os.path.realpath(__file__)
+        # versionfile_source is the relative path from the top of the source
+        # tree (where the .git directory might live) to this file. Invert
+        # this to find the root from __file__.
+        for i in cfg.versionfile_source.split('/'):
+            root = os.path.dirname(root)
+    except NameError:
+        return {"version": "0+unknown", "full-revisionid": None,
+                "dirty": None,
+                "error": "unable to find root of source tree",
+                "date": None}
+
+    try:
+        pieces = git_pieces_from_vcs(cfg.tag_prefix, root, verbose)
+        return render(pieces, cfg.style)
+    except NotThisMethod:
+        pass
+
+    try:
+        if cfg.parentdir_prefix:
+            return versions_from_parentdir(cfg.parentdir_prefix, root, verbose)
+    except NotThisMethod:
+        pass
+
+    return {"version": "0+unknown", "full-revisionid": None,
+            "dirty": None,
+            "error": "unable to compute version", "date": None}

nilmdb/client/__init__.py

@@ -0,0 +1,4 @@
+"""nilmdb.client"""
+
+from nilmdb.client.client import Client
+from nilmdb.client.errors import ClientError, ServerError, Error

nilmdb/client/client.py

@@ -0,0 +1,477 @@
+# -*- coding: utf-8 -*-
+
+"""Class for performing HTTP client requests via libcurl"""
+
+import json
+import contextlib
+
+import nilmdb.utils
+import nilmdb.client.httpclient
+from nilmdb.client.errors import ClientError
+from nilmdb.utils.time import timestamp_to_string, string_to_timestamp
+
+
+def extract_timestamp(line):
+    """Extract just the timestamp from a line of data text"""
+    return string_to_timestamp(line.split()[0])
+
+
+class Client():
+    """Main client interface to the Nilm database."""
+
+    def __init__(self, url, post_json=False):
+        """Initialize client with given URL.  If post_json is true,
+        POST requests are sent with Content-Type 'application/json'
+        instead of the default 'x-www-form-urlencoded'."""
+        self.http = nilmdb.client.httpclient.HTTPClient(url, post_json)
+        self.post_json = post_json
+
+    # __enter__/__exit__ allow this class to be a context manager
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        self.close()
+
+    def _json_post_param(self, data):
+        """Return compact json-encoded version of parameter"""
+        if self.post_json:
+            # If we're posting as JSON, we don't need to encode it further here
+            return data
+        return json.dumps(data, separators=(',', ':'))
+
+    def close(self):
+        """Close the connection; safe to call multiple times"""
+        self.http.close()
+
+    def geturl(self):
+        """Return the URL we're using"""
+        return self.http.baseurl
+
+    def version(self):
+        """Return server version"""
+        return self.http.get("version")
+
+    def dbinfo(self):
+        """Return server database info (path, size, free space)
+        as a dictionary."""
+        return self.http.get("dbinfo")
+
+    def stream_list(self, path=None, layout=None, extended=False):
+        """Return a sorted list of [path, layout] lists.  If 'path' or
+        'layout' are specified, only return streams that match those
+        exact values.  If 'extended' is True, the returned lists have
+        extended info, e.g.: [path, layout, extent_min, extent_max,
+        total_rows, total_seconds."""
+        params = {}
+        if path is not None:
+            params["path"] = path
+        if layout is not None:
+            params["layout"] = layout
+        if extended:
+            params["extended"] = 1
+        streams = self.http.get("stream/list", params)
+        return nilmdb.utils.sort.sort_human(streams, key=lambda s: s[0])
+
+    def stream_get_metadata(self, path, keys=None):
+        """Get stream metadata"""
+        params = {"path": path}
+        if keys is not None:
+            params["key"] = keys
+        return self.http.get("stream/get_metadata", params)
+
+    def stream_set_metadata(self, path, data):
+        """Set stream metadata from a dictionary, replacing all existing
+        metadata."""
+        params = {
+            "path": path,
+            "data": self._json_post_param(data)
+        }
+        return self.http.post("stream/set_metadata", params)
+
+    def stream_update_metadata(self, path, data):
+        """Update stream metadata from a dictionary"""
+        params = {
+            "path": path,
+            "data": self._json_post_param(data)
+        }
+        return self.http.post("stream/update_metadata", params)
+
+    def stream_create(self, path, layout):
+        """Create a new stream"""
+        params = {
+            "path": path,
+            "layout": layout
+        }
+        return self.http.post("stream/create", params)
+
+    def stream_destroy(self, path):
+        """Delete stream.  Fails if any data is still present."""
+        params = {
+            "path": path
+        }
+        return self.http.post("stream/destroy", params)
+
+    def stream_rename(self, oldpath, newpath):
+        """Rename a stream."""
+        params = {
+            "oldpath": oldpath,
+            "newpath": newpath
+        }
+        return self.http.post("stream/rename", params)
+
+    def stream_remove(self, path, start=None, end=None):
+        """Remove data from the specified time range"""
+        params = {
+            "path": path
+        }
+        if start is not None:
+            params["start"] = timestamp_to_string(start)
+        if end is not None:
+            params["end"] = timestamp_to_string(end)
+        total = 0
+        for count in self.http.post_gen("stream/remove", params):
+            total += int(count)
+        return total
+
+    @contextlib.contextmanager
+    def stream_insert_context(self, path, start=None, end=None):
+        """Return a context manager that allows data to be efficiently
+        inserted into a stream in a piecewise manner.  Data is
+        provided as ASCII lines, and is aggregated and sent to the
+        server in larger or smaller chunks as necessary.  Data lines
+        must match the database layout for the given path, and end
+        with a newline.
+
+        Example:
+          with client.stream_insert_context('/path', start, end) as ctx:
+            ctx.insert('1234567890000000 1 2 3 4\\n')
+            ctx.insert('1234567891000000 1 2 3 4\\n')
+
+        For more details, see help for nilmdb.client.client.StreamInserter
+
+        This may make multiple requests to the server, if the data is
+        large enough or enough time has passed between insertions.
+        """
+        ctx = StreamInserter(self, path, start, end)
+        yield ctx
+        ctx.finalize()
+        ctx.destroy()
+
+    def stream_insert(self, path, data, start=None, end=None):
+        """Insert rows of data into a stream.  data should be a string
+        or iterable that provides ASCII data that matches the database
+        layout for path.  Data is passed through stream_insert_context,
+        so it will be broken into reasonably-sized chunks and
+        start/end will be deduced if missing."""
+        with self.stream_insert_context(path, start, end) as ctx:
+            if isinstance(data, bytes):
+                ctx.insert(data)
+            else:
+                for chunk in data:
+                    ctx.insert(chunk)
+        return ctx.last_response
+
+    def stream_insert_block(self, path, data, start, end, binary=False):
+        """Insert a single fixed block of data into the stream.  It is
+        sent directly to the server in one block with no further
+        processing.
+
+        If 'binary' is True, provide raw binary data in little-endian
+        format matching the path layout, including an int64 timestamp.
+        Otherwise, provide ASCII data matching the layout."""
+        params = {
+            "path": path,
+            "start": timestamp_to_string(start),
+            "end": timestamp_to_string(end),
+        }
+        if binary:
+            params["binary"] = 1
+        return self.http.put("stream/insert", data, params)
+
+    def stream_intervals(self, path, start=None, end=None, diffpath=None):
+        """
+        Return a generator that yields each stream interval.
+
+        If 'diffpath' is not None, yields only interval ranges that are
+        present in 'path' but not in 'diffpath'.
+        """
+        params = {
+            "path": path
+        }
+        if diffpath is not None:
+            params["diffpath"] = diffpath
+        if start is not None:
+            params["start"] = timestamp_to_string(start)
+        if end is not None:
+            params["end"] = timestamp_to_string(end)
+        return self.http.get_gen("stream/intervals", params)
+
+    def stream_extract(self, path, start=None, end=None,
+                       count=False, markup=False, binary=False):
+        """
+        Extract data from a stream.  Returns a generator that yields
+        lines of ASCII-formatted data that matches the database
+        layout for the given path.
+
+        If 'count' is True, return a count of matching data points
+        rather than the actual data.  The output format is unchanged.
+
+        If 'markup' is True, include comments in the returned data
+        that indicate interval starts and ends.
+
+        If 'binary' is True, return chunks of raw binary data, rather
+        than lines of ASCII-formatted data.  Raw binary data is
+        little-endian and matches the database types (including an
+        int64 timestamp).
+        """
+        params = {
+            "path": path,
+        }
+        if start is not None:
+            params["start"] = timestamp_to_string(start)
+        if end is not None:
+            params["end"] = timestamp_to_string(end)
+        if count:
+            params["count"] = 1
+        if markup:
+            params["markup"] = 1
+        if binary:
+            params["binary"] = 1
+        return self.http.get_gen("stream/extract", params, binary=binary)
+
+    def stream_count(self, path, start=None, end=None):
+        """
+        Return the number of rows of data in the stream that satisfy
+        the given timestamps.
+        """
+        counts = list(self.stream_extract(path, start, end, count=True))
+        return int(counts[0])
+
+
+class StreamInserter():
+    """Object returned by stream_insert_context() that manages
+    the insertion of rows of data into a particular path.
+
+    The basic data flow is that we are filling a contiguous interval
+    on the server, with no gaps, that extends from timestamp 'start'
+    to timestamp 'end'.  Data timestamps satisfy 'start <= t < end'.
+
+    Data is provided to .insert() as ASCII formatted data separated by
+    newlines.  The chunks of data passed to .insert() do not need to
+    match up with the newlines; less or more than one line can be passed.
+
+    1. The first inserted line begins a new interval that starts at
+    'start'.  If 'start' is not given, it is deduced from the first
+    line's timestamp.
+
+    2. Subsequent lines go into the same contiguous interval.  As lines
+    are inserted, this routine may make multiple insertion requests to
+    the server, but will structure the timestamps to leave no gaps.
+
+    3. The current contiguous interval can be completed by manually
+    calling .finalize(), which the context manager will also do
+    automatically.  This will send any remaining data to the server,
+    using the 'end' timestamp to end the interval.  If no 'end'
+    was provided, it is deduced from the last timestamp seen,
+    plus a small delta.
+
+    After a .finalize(), inserting new data goes back to step 1.
+
+    .update_start() can be called before step 1 to change the start
+    time for the interval.  .update_end() can be called before step 3
+    to change the end time for the interval.
+    """
+
+    # See design.md for a discussion of how much data to send.  This
+    # is a soft limit -- we might send up to twice as much or so
+    _max_data = 2 * 1024 * 1024
+    _max_data_after_send = 64 * 1024
+
+    def __init__(self, client, path, start, end):
+        """'client' is the client object.  'path' is the database
+        path to insert to.  'start' and 'end' are used for the first
+        contiguous interval and may be None."""
+        self.last_response = None
+
+        self._client = client
+        self._path = path
+
+        # Start and end for the overall contiguous interval we're
+        # filling
+        self._interval_start = start
+        self._interval_end = end
+
+        # Current data we're building up to send.  Each string
+        # goes into the array, and gets joined all at once.
+        self._block_data = []
+        self._block_len = 0
+
+        self.destroyed = False
+
+    def destroy(self):
+        """Ensure this object can't be used again without raising
+        an error"""
+        def error(*args, **kwargs):
+            raise Exception("don't reuse this context object")
+        self._send_block = self.insert = self.finalize = self.send = error
+
+    def insert(self, data):
+        """Insert a chunk of ASCII formatted data in string form.  The
+        overall data must consist of lines terminated by '\\n'."""
+        length = len(data)
+        maxdata = self._max_data
+
+        if length > maxdata:
+            # This could make our buffer more than twice what we
+            # wanted to send, so split it up.  This is a bit
+            # inefficient, but the user really shouldn't be providing
+            # this much data at once.
+            for cut in range(0, length, maxdata):
+                self.insert(data[cut:(cut + maxdata)])
+            return
+
+        # Append this string to our list
+        self._block_data.append(data)
+        self._block_len += length
+
+        # Send the block once we have enough data
+        if self._block_len >= maxdata:
+            self._send_block(final=False)
+            if self._block_len >= self._max_data_after_send:
+                raise ValueError("too much data left over after trying"
+                                 " to send intermediate block; is it"
+                                 " missing newlines or malformed?")
+
+    def update_start(self, start):
+        """Update the start time for the next contiguous interval.
+        Call this before starting to insert data for a new interval,
+        for example, after .finalize()"""
+        self._interval_start = start
+
+    def update_end(self, end):
+        """Update the end time for the current contiguous interval.
+        Call this before .finalize()"""
+        self._interval_end = end
+
+    def finalize(self):
+        """Stop filling the current contiguous interval.
+        All outstanding data will be sent, and the interval end
+        time of the interval will be taken from the 'end' argument
+        used when initializing this class, or the most recent
+        value passed to update_end(), or the last timestamp plus
+        a small epsilon value if no other endpoint was provided.
+
+        If more data is inserted after a finalize(), it will become
+        part of a new interval and there may be a gap left in-between."""
+        self._send_block(final=True)
+
+    def send(self):
+        """Send any data that we might have buffered up.  Does not affect
+        any other treatment of timestamps or endpoints."""
+        self._send_block(final=False)
+
+    def _get_first_noncomment(self, block):
+        """Return the (start, end) indices of the first full line in
+        block that isn't a comment, or raise IndexError if
+        there isn't one."""
+        start = 0
+        while True:
+            end = block.find(b'\n', start)
+            if end < 0:
+                raise IndexError
+            if block[start] != b'#'[0]:
+                return (start, (end + 1))
+            start = end + 1
+
+    def _get_last_noncomment(self, block):
+        """Return the (start, end) indices of the last full line in
+        block[:length] that isn't a comment, or raise IndexError if
+        there isn't one."""
+        end = block.rfind(b'\n')
+        if end <= 0:
+            raise IndexError
+        while True:
+            start = block.rfind(b'\n', 0, end)
+            if block[start + 1] != b'#'[0]:
+                return ((start + 1), end)
+            if start == -1:
+                raise IndexError
+            end = start
+
+    def _send_block(self, final=False):
+        """Send data currently in the block.  The data sent will
+        consist of full lines only, so some might be left over."""
+        # Build the full string to send
+        block = b"".join(self._block_data)
+
+        start_ts = self._interval_start
+        if start_ts is None:
+            # Pull start from the first line
+            try:
+                (spos, epos) = self._get_first_noncomment(block)
+                start_ts = extract_timestamp(block[spos:epos])
+            except (ValueError, IndexError):
+                pass  # no timestamp is OK, if we have no data
+
+        if final:
+            # For a final block, it must end in a newline, and the
+            # ending timestamp is either the user-provided end,
+            # or the timestamp of the last line plus epsilon.
+            end_ts = self._interval_end
+            try:
+                if block[-1] != b'\n'[0]:
+                    raise ValueError("final block didn't end with a newline")
+                if end_ts is None:
+                    (spos, epos) = self._get_last_noncomment(block)
+                    end_ts = extract_timestamp(block[spos:epos])
+                    end_ts += nilmdb.utils.time.epsilon
+            except (ValueError, IndexError):
+                pass  # no timestamp is OK, if we have no data
+            self._block_data = []
+            self._block_len = 0
+
+            # Next block is completely fresh
+            self._interval_start = None
+            self._interval_end = None
+        else:
+            # An intermediate block, e.g. "line1\nline2\nline3\nline4"
+            # We need to save "line3\nline4" for the next block, and
+            # use the timestamp from "line3" as the ending timestamp
+            # for this one.
+            try:
+                (spos, epos) = self._get_last_noncomment(block)
+                end_ts = extract_timestamp(block[spos:epos])
+            except (ValueError, IndexError):
+                # If we found no timestamp, give up; we could send this
+                # block later when we have more data.
+                return
+            if spos == 0:
+                # Not enough data to send an intermediate block
+                return
+            if self._interval_end is not None and end_ts > self._interval_end:
+                # User gave us bad endpoints; send it anyway, and let
+                # the server complain so that the error is the same
+                # as if we hadn't done this chunking.
+                end_ts = self._interval_end
+            self._block_data = [block[spos:]]
+            self._block_len = (epos - spos)
+            block = block[:spos]
+
+            # Next block continues where this one ended
+            self._interval_start = end_ts
+
+        # Double check endpoints
+        if (start_ts is None or end_ts is None) or (start_ts == end_ts):
+            # If the block has no non-comment lines, it's OK
+            try:
+                self._get_first_noncomment(block)
+            except IndexError:
+                return
+            raise ClientError("have data to send, but no start/end times")
+
+        # Send it
+        self.last_response = self._client.stream_insert_block(
+            self._path, block, start_ts, end_ts, binary=False)
+
+        return

nilmdb/client/errors.py

@@ -0,0 +1,41 @@
+"""HTTP client errors"""
+
+from nilmdb.utils.printf import sprintf
+
+
+class Error(Exception):
+    """Base exception for both ClientError and ServerError responses"""
+    def __init__(self,
+                 status="Unspecified error",
+                 message=None,
+                 url=None,
+                 traceback=None):
+        super().__init__(status)
+        self.status = status     # e.g. "400 Bad Request"
+        self.message = message   # textual message from the server
+        self.url = url           # URL we were requesting
+        self.traceback = traceback  # server traceback, if available
+
+    def _format_error(self, show_url):
+        s = sprintf("[%s]", self.status)
+        if self.message:
+            s += sprintf(" %s", self.message)
+        if show_url and self.url:
+            s += sprintf(" (%s)", self.url)
+        if self.traceback:
+            s += sprintf("\nServer traceback:\n%s", self.traceback)
+        return s
+
+    def __str__(self):
+        return self._format_error(show_url=False)
+
+    def __repr__(self):
+        return self._format_error(show_url=True)
+
+
+class ClientError(Error):
+    pass
+
+
+class ServerError(Error):
+    pass

nilmdb/client/httpclient.py

@@ -0,0 +1,189 @@
+"""HTTP client library"""
+
+import json
+import urllib.parse
+import requests
+
+from nilmdb.client.errors import ClientError, ServerError, Error
+
+
+class HTTPClient():
+    """Class to manage and perform HTTP requests from the client"""
+    def __init__(self, baseurl="", post_json=False, verify_ssl=True):
+        """If baseurl is supplied, all other functions that take
+        a URL can be given a relative URL instead."""
+        # Verify / clean up URL
+        reparsed = urllib.parse.urlparse(baseurl).geturl()
+        if '://' not in reparsed:
+            reparsed = urllib.parse.urlparse("http://" + baseurl).geturl()
+        self.baseurl = reparsed.rstrip('/') + '/'
+
+        # Note whether we want SSL verification
+        self.verify_ssl = verify_ssl
+
+        # Saved response, so that tests can verify a few things.
+        self._last_response = {}
+
+        # Whether to send application/json POST bodies (versus
+        # x-www-form-urlencoded)
+        self.post_json = post_json
+
+    def _handle_error(self, url, code, body):
+        # Default variables for exception.  We use the entire body as
+        # the default message, in case we can't extract it from a JSON
+        # response.
+        args = {
+            "url": url,
+            "status": str(code),
+            "message": body,
+            "traceback": None
+        }
+        try:
+            # Fill with server-provided data if we can
+            jsonerror = json.loads(body)
+            args["status"] = jsonerror["status"]
+            args["message"] = jsonerror["message"]
+            args["traceback"] = jsonerror["traceback"]
+        except Exception:
+            pass
+        if 400 <= code <= 499:
+            raise ClientError(**args)
+        else:
+            if 500 <= code <= 599:
+                if args["message"] is None:
+                    args["message"] = ("(no message; try disabling "
+                                       "response.stream option in "
+                                       "nilmdb.server for better debugging)")
+                raise ServerError(**args)
+            else:
+                raise Error(**args)
+
+    def close(self):
+        pass
+
+    def _do_req(self, method, url, query_data, body_data, stream, headers):
+        url = urllib.parse.urljoin(self.baseurl, url)
+        try:
+            # Create a new session, ensure we send "Connection: close",
+            # and explicitly close connection after the transfer.
+            # This is to avoid HTTP/1.1 persistent connections
+            # (keepalive), because they have fundamental race
+            # conditions when there are delays between requests:
+            # a new request may be sent at the same instant that the
+            # server decides to timeout the connection.
+            session = requests.Session()
+            if headers is None:
+                headers = {}
+            headers["Connection"] = "close"
+            response = session.request(method, url,
+                                       params=query_data,
+                                       data=body_data,
+                                       stream=stream,
+                                       headers=headers,
+                                       verify=self.verify_ssl)
+
+            # Close the connection.  If it's a generator (stream =
+            # True), the requests library shouldn't actually close the
+            # HTTP connection until all data has been read from the
+            # response.
+            session.close()
+        except requests.RequestException as e:
+            raise ServerError(status="502 Error", url=url,
+                              message=str(e))
+        if response.status_code != 200:
+            self._handle_error(url, response.status_code, response.content)
+        self._last_response = response
+        if response.headers["content-type"] in ("application/json",
+                                                "application/x-json-stream"):
+            return (response, True)
+        else:
+            return (response, False)
+
+    # Normal versions that return data directly
+    def _req(self, method, url, query=None, body=None, headers=None):
+        """
+        Make a request and return the body data as a string or parsed
+        JSON object, or raise an error if it contained an error.
+        """
+        (response, isjson) = self._do_req(method, url, query, body,
+                                          stream=False, headers=headers)
+        if isjson:
+            return json.loads(response.content)
+        return response.text
+
+    def get(self, url, params=None):
+        """Simple GET (parameters in URL)"""
+        return self._req("GET", url, params, None)
+
+    def post(self, url, params=None):
+        """Simple POST (parameters in body)"""
+        if self.post_json:
+            return self._req("POST", url, None,
+                             json.dumps(params),
+                             {'Content-type': 'application/json'})
+        else:
+            return self._req("POST", url, None, params)
+
+    def put(self, url, data, params=None,
+            content_type="application/octet-stream"):
+        """Simple PUT (parameters in URL, data in body)"""
+        h = {'Content-type': content_type}
+        return self._req("PUT", url, query=params, body=data, headers=h)
+
+    # Generator versions that return data one line at a time.
+    def _req_gen(self, method, url, query=None, body=None,
+                 headers=None, binary=False):
+        """
+        Make a request and return a generator that gives back strings
+        or JSON decoded lines of the body data, or raise an error if
+        it contained an eror.
+        """
+        (response, isjson) = self._do_req(method, url, query, body,
+                                          stream=True, headers=headers)
+
+        # Like the iter_lines function in Requests, but only splits on
+        # the specified line ending.
+        def lines(source, ending):
+            pending = None
+            for chunk in source:
+                if pending is not None:
+                    chunk = pending + chunk
+                tmp = chunk.split(ending)
+                lines = tmp[:-1]
+                if chunk.endswith(ending):
+                    pending = None
+                else:
+                    pending = tmp[-1]
+                for line in lines:
+                    yield line
+            if pending is not None:
+                yield pending
+
+        # Yield the chunks or lines as requested
+        if binary:
+            for chunk in response.iter_content(chunk_size=65536):
+                yield chunk
+        elif isjson:
+            for line in lines(response.iter_content(chunk_size=1),
+                              ending=b'\r\n'):
+                yield json.loads(line)
+        else:
+            for line in lines(response.iter_content(chunk_size=65536),
+                              ending=b'\n'):
+                yield line
+
+    def get_gen(self, url, params=None, binary=False):
+        """Simple GET (parameters in URL) returning a generator"""
+        return self._req_gen("GET", url, params, binary=binary)
+
+    def post_gen(self, url, params=None):
+        """Simple POST (parameters in body) returning a generator"""
+        if self.post_json:
+            return self._req_gen("POST", url, None,
+                                 json.dumps(params),
+                                 {'Content-type': 'application/json'})
+        else:
+            return self._req_gen("POST", url, None, params)
+
+    # Not much use for a POST or PUT generator, since they don't
+    # return much data.

nilmdb/client/numpyclient.py

@@ -0,0 +1,263 @@
+# -*- coding: utf-8 -*-
+
+"""Provide a NumpyClient class that is based on normal Client, but has
+additional methods for extracting and inserting data via Numpy arrays."""
+
+import contextlib
+
+import numpy
+
+import nilmdb.utils
+import nilmdb.client.client
+import nilmdb.client.httpclient
+from nilmdb.client.errors import ClientError
+
+
+def layout_to_dtype(layout):
+    ltype = layout.split('_')[0]
+    lcount = int(layout.split('_')[1])
+    if ltype.startswith('int'):
+        atype = '<i' + str(int(ltype[3:]) // 8)
+    elif ltype.startswith('uint'):
+        atype = '<u' + str(int(ltype[4:]) // 8)
+    elif ltype.startswith('float'):
+        atype = '<f' + str(int(ltype[5:]) // 8)
+    else:
+        raise ValueError("bad layout")
+    if lcount == 1:
+        dtype = [('timestamp', '<i8'), ('data', atype)]
+    else:
+        dtype = [('timestamp', '<i8'), ('data', atype, lcount)]
+    return numpy.dtype(dtype)
+
+
+class NumpyClient(nilmdb.client.client.Client):
+    """Subclass of nilmdb.client.Client that adds additional methods for
+    extracting and inserting data via Numpy arrays."""
+
+    def _get_dtype(self, path, layout):
+        if layout is None:
+            streams = self.stream_list(path)
+            if len(streams) != 1:
+                raise ClientError("can't get layout for path: " + path)
+            layout = streams[0][1]
+        return layout_to_dtype(layout)
+
+    def stream_extract_numpy(self, path, start=None, end=None,
+                             layout=None, maxrows=100000,
+                             structured=False):
+        """
+        Extract data from a stream.  Returns a generator that yields
+        Numpy arrays of up to 'maxrows' of data each.
+
+        If 'layout' is None, it is read using stream_info.
+
+        If 'structured' is False, all data is converted to float64
+        and returned in a flat 2D array.  Otherwise, data is returned
+        as a structured dtype in a 1D array.
+        """
+        dtype = self._get_dtype(path, layout)
+
+        def to_numpy(data):
+            a = numpy.frombuffer(data, dtype)
+            if structured:
+                return a
+            return numpy.c_[a['timestamp'], a['data']]
+
+        chunks = []
+        total_len = 0
+        maxsize = dtype.itemsize * maxrows
+        for data in self.stream_extract(path, start, end, binary=True):
+            # Add this block of binary data
+            chunks.append(data)
+            total_len += len(data)
+
+            # See if we have enough to make the requested Numpy array
+            while total_len >= maxsize:
+                assembled = b"".join(chunks)
+                total_len -= maxsize
+                chunks = [assembled[maxsize:]]
+                block = assembled[:maxsize]
+                yield to_numpy(block)
+
+        if total_len:
+            yield to_numpy(b"".join(chunks))
+
+    @contextlib.contextmanager
+    def stream_insert_numpy_context(self, path, start=None, end=None,
+                                    layout=None):
+        """Return a context manager that allows data to be efficiently
+        inserted into a stream in a piecewise manner.  Data is
+        provided as Numpy arrays, and is aggregated and sent to the
+        server in larger or smaller chunks as necessary.  Data format
+        must match the database layout for the given path.
+
+        For more details, see help for
+        nilmdb.client.numpyclient.StreamInserterNumpy
+
+        If 'layout' is not None, use it as the layout rather than
+        querying the database.
+        """
+        dtype = self._get_dtype(path, layout)
+        ctx = StreamInserterNumpy(self, path, start, end, dtype)
+        yield ctx
+        ctx.finalize()
+        ctx.destroy()
+
+    def stream_insert_numpy(self, path, data, start=None, end=None,
+                            layout=None):
+        """Insert data into a stream.  data should be a Numpy array
+        which will be passed through stream_insert_numpy_context to
+        break it into chunks etc.  See the help for that function
+        for details."""
+        with self.stream_insert_numpy_context(path, start, end, layout) as ctx:
+            if isinstance(data, numpy.ndarray):
+                ctx.insert(data)
+            else:
+                for chunk in data:
+                    ctx.insert(chunk)
+        return ctx.last_response
+
+
+class StreamInserterNumpy(nilmdb.client.client.StreamInserter):
+    """Object returned by stream_insert_numpy_context() that manages
+    the insertion of rows of data into a particular path.
+
+    See help for nilmdb.client.client.StreamInserter for details.
+    The only difference is that, instead of ASCII formatted data,
+    this context manager can take Numpy arrays, which are either
+    structured (1D with complex dtype) or flat (2D with simple dtype).
+    """
+
+    # Soft limit of how many bytes to send per HTTP request.
+    _max_data = 2 * 1024 * 1024
+
+    def __init__(self, client, path, start, end, dtype):
+        """
+        'client' is the client object.  'path' is the database path
+        to insert to.  'start' and 'end' are used for the first
+        contiguous interval and may be None.  'dtype' is the Numpy
+        dtype for this stream.
+        """
+        super(StreamInserterNumpy, self).__init__(client, path, start, end)
+        self._dtype = dtype
+
+        # Max rows to send at once
+        self._max_rows = self._max_data // self._dtype.itemsize
+
+        # List of the current arrays we're building up to send
+        self._block_arrays = []
+        self._block_rows = 0
+
+    def insert(self, array):
+        """Insert Numpy data, which must match the layout type."""
+        if not isinstance(array, numpy.ndarray):
+            array = numpy.array(array)
+        if array.ndim == 1:
+            # Already a structured array; just verify the type
+            if array.dtype != self._dtype:
+                raise ValueError("wrong dtype for 1D (structured) array")
+        elif array.ndim == 2:
+            # Convert to structured array
+            sarray = numpy.zeros(array.shape[0], dtype=self._dtype)
+            try:
+                sarray['timestamp'] = array[:, 0]
+                # Need the squeeze in case sarray['data'] is 1 dimensional
+                sarray['data'] = numpy.squeeze(array[:, 1:])
+            except (IndexError, ValueError):
+                raise ValueError("wrong number of fields for this data type")
+            array = sarray
+        else:
+            raise ValueError("wrong number of dimensions in array")
+
+        length = len(array)
+        maxrows = self._max_rows
+
+        if length == 0:
+            return
+        if length > maxrows:
+            # This is more than twice what we wanted to send, so split
+            # it up.  This is a bit inefficient, but the user really
+            # shouldn't be providing this much data at once.
+            for cut in range(0, length, maxrows):
+                self.insert(array[cut:(cut + maxrows)])
+            return
+
+        # Add this array to our list
+        self._block_arrays.append(array)
+        self._block_rows += length
+
+        # Send if it's too long
+        if self._block_rows >= maxrows:
+            self._send_block(final=False)
+
+    def _send_block(self, final=False):
+        """Send the data current stored up.  One row might be left
+        over if we need its timestamp saved."""
+
+        # Build the full array to send
+        if self._block_rows == 0:
+            array = numpy.zeros(0, dtype=self._dtype)
+        else:
+            array = numpy.hstack(self._block_arrays)
+
+        # Get starting timestamp
+        start_ts = self._interval_start
+        if start_ts is None:
+            # Pull start from the first row
+            try:
+                start_ts = array['timestamp'][0]
+            except IndexError:
+                pass  # no timestamp is OK, if we have no data
+
+        # Get ending timestamp
+        if final:
+            # For a final block, the timestamp is either the
+            # user-provided end, or the timestamp of the last line
+            # plus epsilon.
+            end_ts = self._interval_end
+            if end_ts is None:
+                try:
+                    end_ts = array['timestamp'][-1]
+                    end_ts += nilmdb.utils.time.epsilon
+                except IndexError:
+                    pass  # no timestamp is OK, if we have no data
+            self._block_arrays = []
+            self._block_rows = 0
+
+            # Next block is completely fresh
+            self._interval_start = None
+            self._interval_end = None
+        else:
+            # An intermediate block.  We need to save the last row
+            # for the next block, and use its timestamp as the ending
+            # timestamp for this one.
+            if len(array) < 2:
+                # Not enough data to send an intermediate block
+                return
+            end_ts = array['timestamp'][-1]
+            if self._interval_end is not None and end_ts > self._interval_end:
+                # User gave us bad endpoints; send it anyway, and let
+                # the server complain so that the error is the same
+                # as if we hadn't done this chunking.
+                end_ts = self._interval_end
+            self._block_arrays = [array[-1:]]
+            self._block_rows = 1
+            array = array[:-1]
+
+            # Next block continues where this one ended
+            self._interval_start = end_ts
+
+        # If we have no endpoints, or equal endpoints, it's OK as long
+        # as there's no data to send
+        if (start_ts is None or end_ts is None) or (start_ts == end_ts):
+            if not array:
+                return
+            raise ClientError("have data to send, but invalid start/end times")
+
+        # Send it
+        data = array.tostring()
+        self.last_response = self._client.stream_insert_block(
+            self._path, data, start_ts, end_ts, binary=True)
+
+        return

nilmdb/cmdline/__init__.py

@@ -0,0 +1,3 @@
+"""nilmdb.cmdline"""
+
+from nilmdb.cmdline.cmdline import Cmdline

nilmdb/cmdline/cmdline.py

@@ -0,0 +1,173 @@
+"""Command line client functionality"""
+
+import os
+import sys
+import signal
+import argparse
+from argparse import ArgumentDefaultsHelpFormatter as def_form
+
+import nilmdb.client
+from nilmdb.utils.printf import fprintf, sprintf
+import nilmdb.utils.time
+
+import argcomplete
+import datetime_tz
+
+# Valid subcommands.  Defined in separate files just to break
+# things up -- they're still called with Cmdline as self.
+subcommands = ["help", "info", "create", "rename", "list", "intervals",
+               "metadata", "insert", "extract", "remove", "destroy"]
+
+# Import the subcommand modules
+subcmd_mods = {}
+for cmd in subcommands:
+    subcmd_mods[cmd] = __import__("nilmdb.cmdline." + cmd, fromlist=[cmd])
+
+
+class JimArgumentParser(argparse.ArgumentParser):
+    def parse_args(self, args=None, namespace=None):
+        # Look for --version anywhere and change it to just "nilmtool
+        # --version".  This makes "nilmtool cmd --version" work, which
+        # is needed by help2man.
+        if "--version" in (args or sys.argv[1:]):
+            args = ["--version"]
+        return argparse.ArgumentParser.parse_args(self, args, namespace)
+
+    def error(self, message):
+        self.print_usage(sys.stderr)
+        self.exit(2, sprintf("error: %s\n", message))
+
+
+class Complete():
+    # Completion helpers, for using argcomplete (see
+    # extras/nilmtool-bash-completion.sh)
+    def escape(self, s):
+        quote_chars = ["\\", "\"", "'", " "]
+        for char in quote_chars:
+            s = s.replace(char, "\\" + char)
+        return s
+
+    def none(self, prefix, parsed_args, **kwargs):
+        return []
+    rate = none
+    time = none
+    url = none
+
+    def path(self, prefix, parsed_args, **kwargs):
+        client = nilmdb.client.Client(parsed_args.url)
+        return (self.escape(s[0])
+                for s in client.stream_list()
+                if s[0].startswith(prefix))
+
+    def layout(self, prefix, parsed_args, **kwargs):
+        types = ["int8", "int16", "int32", "int64",
+                 "uint8", "uint16", "uint32", "uint64",
+                 "float32", "float64"]
+        layouts = []
+        for i in range(1, 10):
+            layouts.extend([(t + "_" + str(i)) for t in types])
+        return (lay for lay in layouts if lay.startswith(prefix))
+
+    def meta_key(self, prefix, parsed_args, **kwargs):
+        return (kv.split('=')[0] for kv
+                in self.meta_keyval(prefix, parsed_args, **kwargs))
+
+    def meta_keyval(self, prefix, parsed_args, **kwargs):
+        client = nilmdb.client.Client(parsed_args.url)
+        path = parsed_args.path
+        if not path:
+            return []
+        results = []
+        for (k, v) in client.stream_get_metadata(path).items():
+            kv = self.escape(k + '=' + v)
+            if kv.startswith(prefix):
+                results.append(kv)
+        return results
+
+
+class Cmdline():
+
+    def __init__(self, argv=None):
+        self.argv = argv or sys.argv[1:]
+        self.client = None
+        self.def_url = os.environ.get("NILMDB_URL", "http://localhost/nilmdb/")
+        self.subcmd = {}
+        self.complete = Complete()
+        self.complete_output_stream = None  # overridden by test suite
+
+    def arg_time(self, toparse):
+        """Parse a time string argument"""
+        try:
+            return nilmdb.utils.time.parse_time(toparse)
+        except ValueError as e:
+            raise argparse.ArgumentTypeError(sprintf("%s \"%s\"",
+                                                     str(e), toparse))
+
+    # Set up the parser
+    def parser_setup(self):
+        self.parser = JimArgumentParser(add_help=False,
+                                        formatter_class=def_form)
+
+        group = self.parser.add_argument_group("General options")
+        group.add_argument("-h", "--help", action='help',
+                           help='show this help message and exit')
+        group.add_argument("-v", "--version", action="version",
+                           version=nilmdb.__version__)
+
+        group = self.parser.add_argument_group("Server")
+        group.add_argument("-u", "--url", action="store",
+                           default=self.def_url,
+                           help="NilmDB server URL (default: %(default)s)"
+                           ).completer = self.complete.url
+
+        sub = self.parser.add_subparsers(
+            title="Commands", dest="command", required=True,
+            description="Use 'help command' or 'command --help' for more "
+            "details on a particular command.")
+
+        # Set up subcommands (defined in separate files)
+        for cmd in subcommands:
+            self.subcmd[cmd] = subcmd_mods[cmd].setup(self, sub)
+
+    def die(self, formatstr, *args):
+        fprintf(sys.stderr, formatstr + "\n", *args)
+        if self.client:
+            self.client.close()
+        sys.exit(-1)
+
+    def run(self):
+        # Set SIGPIPE to its default handler -- we don't need Python
+        # to catch it for us.
+        signal.signal(signal.SIGPIPE, signal.SIG_DFL)
+
+        # Clear cached timezone, so that we can pick up timezone changes
+        # while running this from the test suite.
+        datetime_tz._localtz = None
+
+        # Run parser
+        self.parser_setup()
+        argcomplete.autocomplete(self.parser, exit_method=sys.exit,
+                                 output_stream=self.complete_output_stream)
+        self.args = self.parser.parse_args(self.argv)
+
+        # Run arg verify handler if there is one
+        if "verify" in self.args:
+            self.args.verify(self)
+
+        self.client = nilmdb.client.Client(self.args.url)
+
+        # Make a test connection to make sure things work,
+        # unless the particular command requests that we don't.
+        if "no_test_connect" not in self.args:
+            try:
+                self.client.version()
+            except nilmdb.client.Error as e:
+                self.die("error connecting to server: %s", str(e))
+
+        # Now dispatch client request to appropriate function.  Parser
+        # should have ensured that we don't have any unknown commands
+        # here.
+        retval = self.args.handler(self) or 0
+
+        self.client.close()
+        sys.exit(retval)

nilmdb/cmdline/create.py

@@ -0,0 +1,38 @@
+from argparse import RawDescriptionHelpFormatter as raw_form
+
+import nilmdb.client
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("create", help="Create a new stream",
+                         formatter_class=raw_form,
+                         description="""
+Create a new empty stream at the specified path and with the specified
+layout type.
+
+Layout types are of the format: type_count
+
+  'type' is a data type like 'float32', 'float64', 'uint16', 'int32', etc.
+
+  'count' is the number of columns of this type.
+
+  For example, 'float32_8' means the data for this stream has 8 columns of
+  32-bit floating point values.
+""")
+    cmd.set_defaults(handler=cmd_create)
+    group = cmd.add_argument_group("Required arguments")
+    group.add_argument("path",
+                       help="Path (in database) of new stream, e.g. /foo/bar",
+                       ).completer = self.complete.path
+    group.add_argument("layout",
+                       help="Layout type for new stream, e.g. float32_8",
+                       ).completer = self.complete.layout
+    return cmd
+
+
+def cmd_create(self):
+    """Create new stream"""
+    try:
+        self.client.stream_create(self.args.path, self.args.layout)
+    except nilmdb.client.ClientError as e:
+        self.die("error creating stream: %s", str(e))

nilmdb/cmdline/destroy.py

@@ -0,0 +1,52 @@
+import fnmatch
+
+from argparse import ArgumentDefaultsHelpFormatter as def_form
+
+from nilmdb.utils.printf import printf
+import nilmdb.client
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("destroy", help="Delete a stream and all data",
+                         formatter_class=def_form,
+                         description="""
+                         Destroy the stream at the specified path.
+                         The stream must be empty.  All metadata
+                         related to the stream is permanently deleted.
+
+                         Wildcards and multiple paths are supported.
+                         """)
+    cmd.set_defaults(handler=cmd_destroy)
+    group = cmd.add_argument_group("Options")
+    group.add_argument("-R", "--remove", action="store_true",
+                       help="Remove all data before destroying stream")
+    group.add_argument("-q", "--quiet", action="store_true",
+                       help="Don't display names when destroying "
+                       "multiple paths")
+    group = cmd.add_argument_group("Required arguments")
+    group.add_argument("path", nargs='+',
+                       help="Path of the stream to delete, e.g. /foo/bar/*",
+                       ).completer = self.complete.path
+    return cmd
+
+
+def cmd_destroy(self):
+    """Destroy stream"""
+    streams = [s[0] for s in self.client.stream_list()]
+    paths = []
+    for path in self.args.path:
+        new = fnmatch.filter(streams, path)
+        if not new:
+            self.die("error: no stream matched path: %s", path)
+        paths.extend(new)
+
+    for path in paths:
+        if not self.args.quiet and len(paths) > 1:
+            printf("Destroying %s\n", path)
+
+        try:
+            if self.args.remove:
+                self.client.stream_remove(path)
+            self.client.stream_destroy(path)
+        except nilmdb.client.ClientError as e:
+            self.die("error destroying stream: %s", str(e))

nilmdb/cmdline/extract.py

@@ -0,0 +1,96 @@
+import sys
+
+from nilmdb.utils.printf import printf
+import nilmdb.client
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("extract", help="Extract data",
+                         description="""
+                         Extract data from a stream.
+                         """)
+    cmd.set_defaults(verify=cmd_extract_verify,
+                     handler=cmd_extract)
+
+    group = cmd.add_argument_group("Data selection")
+    group.add_argument("path",
+                       help="Path of stream, e.g. /foo/bar",
+                       ).completer = self.complete.path
+    group.add_argument("-s", "--start", required=True,
+                       metavar="TIME", type=self.arg_time,
+                       help="Starting timestamp (free-form, inclusive)",
+                       ).completer = self.complete.time
+    group.add_argument("-e", "--end", required=True,
+                       metavar="TIME", type=self.arg_time,
+                       help="Ending timestamp (free-form, noninclusive)",
+                       ).completer = self.complete.time
+
+    group = cmd.add_argument_group("Output format")
+    group.add_argument("-B", "--binary", action="store_true",
+                       help="Raw binary output")
+    group.add_argument("-b", "--bare", action="store_true",
+                       help="Exclude timestamps from output lines")
+    group.add_argument("-a", "--annotate", action="store_true",
+                       help="Include comments with some information "
+                       "about the stream")
+    group.add_argument("-m", "--markup", action="store_true",
+                       help="Include comments with interval starts and ends")
+    group.add_argument("-T", "--timestamp-raw", action="store_true",
+                       help="Show raw timestamps in annotated information")
+    group.add_argument("-c", "--count", action="store_true",
+                       help="Just output a count of matched data points")
+    return cmd
+
+
+def cmd_extract_verify(self):
+    if self.args.start > self.args.end:
+        self.parser.error("start is after end")
+
+    if self.args.binary:
+        if (self.args.bare or self.args.annotate or self.args.markup or
+                self.args.timestamp_raw or self.args.count):
+            self.parser.error("--binary cannot be combined with other options")
+
+
+def cmd_extract(self):
+    streams = self.client.stream_list(self.args.path)
+    if len(streams) != 1:
+        self.die("error getting stream info for path %s", self.args.path)
+    layout = streams[0][1]
+
+    if self.args.timestamp_raw:
+        time_string = nilmdb.utils.time.timestamp_to_string
+    else:
+        time_string = nilmdb.utils.time.timestamp_to_human
+
+    if self.args.annotate:
+        printf("# path: %s\n", self.args.path)
+        printf("# layout: %s\n", layout)
+        printf("# start: %s\n", time_string(self.args.start))
+        printf("# end: %s\n", time_string(self.args.end))
+
+    printed = False
+    if self.args.binary:
+        printer = sys.stdout.buffer.write
+    else:
+        printer = lambda x: print(x.decode('utf-8'))
+    bare = self.args.bare
+    count = self.args.count
+    for dataline in self.client.stream_extract(self.args.path,
+                                               self.args.start,
+                                               self.args.end,
+                                               self.args.count,
+                                               self.args.markup,
+                                               self.args.binary):
+        if bare and not count:
+            # Strip timestamp (first element).  Doesn't make sense
+            # if we are only returning a count.
+            dataline = b' '.join(dataline.split(b' ')[1:])
+        printer(dataline)
+        printed = True
+    if not printed:
+        if self.args.annotate:
+            printf("# no data\n")
+        return 2
+
+    return 0

nilmdb/cmdline/help.py

@@ -0,0 +1,25 @@
+import argparse
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("help", help="Show detailed help for a command",
+                         description="""
+                         Show help for a command. 'help command' is
+                         the same as 'command --help'.
+                         """)
+    cmd.set_defaults(handler=cmd_help)
+    cmd.set_defaults(no_test_connect=True)
+    cmd.add_argument("command", nargs="?",
+                     help="Command to get help about")
+    cmd.add_argument("rest", nargs=argparse.REMAINDER,
+                     help=argparse.SUPPRESS)
+    return cmd
+
+
+def cmd_help(self):
+    if self.args.command in self.subcmd:
+        self.subcmd[self.args.command].print_help()
+    else:
+        self.parser.print_help()
+
+    return

nilmdb/cmdline/info.py

@@ -0,0 +1,30 @@
+from argparse import ArgumentDefaultsHelpFormatter as def_form
+
+import nilmdb.client
+from nilmdb.utils.printf import printf
+from nilmdb.utils import human_size
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("info", help="Server information",
+                         formatter_class=def_form,
+                         description="""
+                         List information about the server, like
+                         version.
+                         """)
+    cmd.set_defaults(handler=cmd_info)
+    return cmd
+
+
+def cmd_info(self):
+    """Print info about the server"""
+    printf("Client version: %s\n", nilmdb.__version__)
+    printf("Server version: %s\n", self.client.version())
+    printf("Server URL: %s\n", self.client.geturl())
+    dbinfo = self.client.dbinfo()
+    printf("Server database path: %s\n", dbinfo["path"])
+    for (desc, field) in [("used by NilmDB", "size"),
+                          ("used by other", "other"),
+                          ("reserved", "reserved"),
+                          ("free", "free")]:
+        printf("Server disk space %s: %s\n", desc, human_size(dbinfo[field]))

nilmdb/cmdline/insert.py

@@ -0,0 +1,135 @@
+import sys
+
+from nilmdb.utils.printf import printf
+import nilmdb.client
+import nilmdb.utils.timestamper as timestamper
+import nilmdb.utils.time
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("insert", help="Insert data",
+                         description="""
+                         Insert data into a stream.
+                         """)
+    cmd.set_defaults(verify=cmd_insert_verify,
+                     handler=cmd_insert)
+    cmd.add_argument("-q", "--quiet", action='store_true',
+                     help='suppress unnecessary messages')
+
+    group = cmd.add_argument_group("Timestamping",
+                                   description="""
+                                   To add timestamps, specify the
+                                   arguments --timestamp and --rate,
+                                   and provide a starting time.
+                                   """)
+
+    group.add_argument("-t", "--timestamp", action="store_true",
+                       help="Add timestamps to each line")
+    group.add_argument("-r", "--rate", type=float,
+                       help="Data rate, in Hz",
+                       ).completer = self.complete.rate
+
+    group = cmd.add_argument_group("Start time",
+                                   description="""
+                                   Start time may be manually
+                                   specified with --start, or guessed
+                                   from the filenames using
+                                   --filename.  Set the TZ environment
+                                   variable to change the default
+                                   timezone.""")
+
+    exc = group.add_mutually_exclusive_group()
+    exc.add_argument("-s", "--start",
+                     metavar="TIME", type=self.arg_time,
+                     help="Starting timestamp (free-form)",
+                     ).completer = self.complete.time
+    exc.add_argument("-f", "--filename", action="store_true",
+                     help="Use filename to determine start time")
+
+    group = cmd.add_argument_group("End time",
+                                   description="""
+                                   End time for the overall stream.
+                                   (required when not using --timestamp).
+                                   Set the TZ environment
+                                   variable to change the default
+                                   timezone.""")
+    group.add_argument("-e", "--end",
+                       metavar="TIME", type=self.arg_time,
+                       help="Ending timestamp (free-form)",
+                       ).completer = self.complete.time
+
+    group = cmd.add_argument_group("Required parameters")
+    group.add_argument("path",
+                       help="Path of stream, e.g. /foo/bar",
+                       ).completer = self.complete.path
+    group.add_argument("file", nargs='?', default='-',
+                       help="File to insert (default: - (stdin))")
+    return cmd
+
+
+def cmd_insert_verify(self):
+    if self.args.timestamp:
+        if not self.args.rate:
+            self.die("error: --rate is needed, but was not specified")
+        if not self.args.filename and self.args.start is None:
+            self.die("error: need --start or --filename "
+                     "when adding timestamps")
+    else:
+        if self.args.start is None or self.args.end is None:
+            self.die("error: when not adding timestamps, --start and "
+                     "--end are required")
+
+
+def cmd_insert(self):
+    # Find requested stream
+    streams = self.client.stream_list(self.args.path)
+    if len(streams) != 1:
+        self.die("error getting stream info for path %s", self.args.path)
+
+    arg = self.args
+
+    try:
+        filename = arg.file
+        if filename == '-':
+            infile = sys.stdin.buffer
+        else:
+            try:
+                infile = open(filename, "rb")
+            except IOError:
+                self.die("error opening input file %s", filename)
+
+        if arg.start is None:
+            try:
+                arg.start = nilmdb.utils.time.parse_time(filename)
+            except ValueError:
+                self.die("error extracting start time from filename '%s'",
+                         filename)
+
+        if arg.timestamp:
+            data = timestamper.TimestamperRate(infile, arg.start, arg.rate)
+        else:
+            data = iter(lambda: infile.read(1048576), b'')
+
+        # Print info
+        if not arg.quiet:
+            printf(" Input file: %s\n", filename)
+            printf(" Start time: %s\n",
+                   nilmdb.utils.time.timestamp_to_human(arg.start))
+            if arg.end:
+                printf("   End time: %s\n",
+                       nilmdb.utils.time.timestamp_to_human(arg.end))
+            if arg.timestamp:
+                printf("Timestamper: %s\n", str(data))
+
+        # Insert the data
+        self.client.stream_insert(arg.path, data, arg.start, arg.end)
+
+    except nilmdb.client.Error as e:
+        # TODO: It would be nice to be able to offer better errors
+        # here, particularly in the case of overlap, which just shows
+        # ugly bracketed ranges of 16-digit numbers and a mangled URL.
+        # Need to consider adding something like e.prettyprint()
+        # that is smarter about the contents of the error.
+        self.die("error inserting data: %s", str(e))
+
+    return

nilmdb/cmdline/intervals.py

@@ -0,0 +1,76 @@
+from argparse import ArgumentDefaultsHelpFormatter as def_form
+
+from nilmdb.utils.printf import printf
+import nilmdb.utils.time
+from nilmdb.utils.interval import Interval
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("intervals", help="List intervals",
+                         formatter_class=def_form,
+                         description="""
+                         List intervals in a stream, similar to
+                         'list --detail path'.
+
+                         If '--diff diffpath' is provided, only
+                         interval ranges that are present in 'path'
+                         and not present in 'diffpath' are printed.
+                         """)
+    cmd.set_defaults(verify=cmd_intervals_verify,
+                     handler=cmd_intervals)
+
+    group = cmd.add_argument_group("Stream selection")
+    group.add_argument("path", metavar="PATH",
+                       help="List intervals for this path",
+                       ).completer = self.complete.path
+    group.add_argument("-d", "--diff", metavar="PATH",
+                       help="Subtract intervals from this path",
+                       ).completer = self.complete.path
+
+    group = cmd.add_argument_group("Interval details")
+    group.add_argument("-s", "--start",
+                       metavar="TIME", type=self.arg_time,
+                       help="Starting timestamp for intervals "
+                       "(free-form, inclusive)",
+                       ).completer = self.complete.time
+    group.add_argument("-e", "--end",
+                       metavar="TIME", type=self.arg_time,
+                       help="Ending timestamp for intervals "
+                       "(free-form, noninclusive)",
+                       ).completer = self.complete.time
+
+    group = cmd.add_argument_group("Misc options")
+    group.add_argument("-T", "--timestamp-raw", action="store_true",
+                       help="Show raw timestamps when printing times")
+    group.add_argument("-o", "--optimize", action="store_true",
+                       help="Optimize (merge adjacent) intervals")
+
+    return cmd
+
+
+def cmd_intervals_verify(self):
+    if self.args.start is not None and self.args.end is not None:
+        if self.args.start >= self.args.end:
+            self.parser.error("start must precede end")
+
+
+def cmd_intervals(self):
+    """List intervals in a stream"""
+    if self.args.timestamp_raw:
+        time_string = nilmdb.utils.time.timestamp_to_string
+    else:
+        time_string = nilmdb.utils.time.timestamp_to_human
+
+    try:
+        intervals = (Interval(start, end) for (start, end) in
+                     self.client.stream_intervals(self.args.path,
+                                                  self.args.start,
+                                                  self.args.end,
+                                                  self.args.diff))
+        if self.args.optimize:
+            intervals = nilmdb.utils.interval.optimize(intervals)
+        for i in intervals:
+            printf("[ %s -> %s ]\n", time_string(i.start), time_string(i.end))
+
+    except nilmdb.client.ClientError as e:
+        self.die("error listing intervals: %s", str(e))

nilmdb/cmdline/list.py

@@ -0,0 +1,105 @@
+import fnmatch
+from argparse import ArgumentDefaultsHelpFormatter as def_form
+
+from nilmdb.utils.printf import printf
+import nilmdb.utils.time
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("list", help="List streams",
+                         formatter_class=def_form,
+                         description="""
+                         List streams available in the database,
+                         optionally filtering by path.  Wildcards
+                         are accepted; non-matching paths or wildcards
+                         are ignored.
+                         """)
+    cmd.set_defaults(verify=cmd_list_verify,
+                     handler=cmd_list)
+
+    group = cmd.add_argument_group("Stream filtering")
+    group.add_argument("path", metavar="PATH", default=["*"], nargs='*',
+                       ).completer = self.complete.path
+
+    group = cmd.add_argument_group("Interval info")
+    group.add_argument("-E", "--ext", action="store_true",
+                       help="Show extended stream info, like interval "
+                       "extents and row count")
+
+    group = cmd.add_argument_group("Interval details")
+    group.add_argument("-d", "--detail", action="store_true",
+                       help="Show available data time intervals")
+    group.add_argument("-s", "--start",
+                       metavar="TIME", type=self.arg_time,
+                       help="Starting timestamp for intervals "
+                       "(free-form, inclusive)",
+                       ).completer = self.complete.time
+    group.add_argument("-e", "--end",
+                       metavar="TIME", type=self.arg_time,
+                       help="Ending timestamp for intervals "
+                       "(free-form, noninclusive)",
+                       ).completer = self.complete.time
+
+    group = cmd.add_argument_group("Misc options")
+    group.add_argument("-T", "--timestamp-raw", action="store_true",
+                       help="Show raw timestamps when printing times")
+    group.add_argument("-l", "--layout", action="store_true",
+                       help="Show layout type next to path name")
+    group.add_argument("-n", "--no-decim", action="store_true",
+                       help="Skip paths containing \"~decim-\"")
+
+    return cmd
+
+
+def cmd_list_verify(self):
+    if self.args.start is not None and self.args.end is not None:
+        if self.args.start >= self.args.end:
+            self.parser.error("start must precede end")
+
+    if self.args.start is not None or self.args.end is not None:
+        if not self.args.detail:
+            self.parser.error("--start and --end only make sense "
+                              "with --detail")
+
+
+def cmd_list(self):
+    """List available streams"""
+    streams = self.client.stream_list(extended=True)
+
+    if self.args.timestamp_raw:
+        time_string = nilmdb.utils.time.timestamp_to_string
+    else:
+        time_string = nilmdb.utils.time.timestamp_to_human
+
+    for argpath in self.args.path:
+        for stream in streams:
+            (path, layout, int_min, int_max, rows, time) = stream[:6]
+            if not fnmatch.fnmatch(path, argpath):
+                continue
+            if self.args.no_decim and "~decim-" in path:
+                continue
+
+            if self.args.layout:
+                printf("%s %s\n", path, layout)
+            else:
+                printf("%s\n", path)
+
+            if self.args.ext:
+                if int_min is None or int_max is None:
+                    printf("  interval extents: (no data)\n")
+                else:
+                    printf("  interval extents: %s -> %s\n",
+                           time_string(int_min), time_string(int_max))
+                printf("        total data: %d rows, %.6f seconds\n",
+                       rows or 0,
+                       nilmdb.utils.time.timestamp_to_seconds(time or 0))
+
+            if self.args.detail:
+                printed = False
+                for (start, end) in self.client.stream_intervals(
+                        path, self.args.start, self.args.end):
+                    printf("  [ %s -> %s ]\n",
+                           time_string(start), time_string(end))
+                    printed = True
+                if not printed:
+                    printf("  (no intervals)\n")

nilmdb/cmdline/metadata.py

@@ -0,0 +1,90 @@
+from nilmdb.utils.printf import printf
+import nilmdb
+import nilmdb.client
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("metadata", help="Get or set stream metadata",
+                         description="""
+                         Get or set key=value metadata associated with
+                         a stream.
+                         """,
+                         usage="%(prog)s path [-g [key ...] | "
+                         "-s key=value [...] | -u key=value [...]] | "
+                         "-d [key ...]")
+    cmd.set_defaults(handler=cmd_metadata)
+
+    group = cmd.add_argument_group("Required arguments")
+    group.add_argument("path",
+                       help="Path of stream, e.g. /foo/bar",
+                       ).completer = self.complete.path
+
+    group = cmd.add_argument_group("Actions")
+    exc = group.add_mutually_exclusive_group()
+    exc.add_argument("-g", "--get", nargs="*", metavar="key",
+                     help="Get metadata for specified keys (default all)",
+                     ).completer = self.complete.meta_key
+    exc.add_argument("-s", "--set", nargs="+", metavar="key=value",
+                     help="Replace all metadata with provided "
+                     "key=value pairs",
+                     ).completer = self.complete.meta_keyval
+    exc.add_argument("-u", "--update", nargs="+", metavar="key=value",
+                     help="Update metadata using provided "
+                     "key=value pairs",
+                     ).completer = self.complete.meta_keyval
+    exc.add_argument("-d", "--delete", nargs="*", metavar="key",
+                     help="Delete metadata for specified keys (default all)",
+                     ).completer = self.complete.meta_key
+    return cmd
+
+
+def cmd_metadata(self):
+    """Manipulate metadata"""
+    if self.args.set is not None or self.args.update is not None:
+        # Either set, or update
+        if self.args.set is not None:
+            keyvals = self.args.set
+            handler = self.client.stream_set_metadata
+        else:
+            keyvals = self.args.update
+            handler = self.client.stream_update_metadata
+
+        # Extract key=value pairs
+        data = {}
+        for keyval in keyvals:
+            kv = keyval.split('=')
+            if len(kv) != 2 or kv[0] == "":
+                self.die("error parsing key=value argument '%s'", keyval)
+            data[kv[0]] = kv[1]
+
+        # Make the call
+        try:
+            handler(self.args.path, data)
+        except nilmdb.client.ClientError as e:
+            self.die("error setting/updating metadata: %s", str(e))
+    elif self.args.delete is not None:
+        # Delete (by setting values to empty strings)
+        keys = None
+        if self.args.delete:
+            keys = list(self.args.delete)
+        try:
+            data = self.client.stream_get_metadata(self.args.path, keys)
+            for key in data:
+                data[key] = ""
+            self.client.stream_update_metadata(self.args.path, data)
+        except nilmdb.client.ClientError as e:
+            self.die("error deleting metadata: %s", str(e))
+    else:
+        # Get (or unspecified)
+        keys = None
+        if self.args.get:
+            keys = list(self.args.get)
+        try:
+            data = self.client.stream_get_metadata(self.args.path, keys)
+        except nilmdb.client.ClientError as e:
+            self.die("error getting metadata: %s", str(e))
+        for key, value in sorted(data.items()):
+            # Print nonexistant keys as having empty value
+            if value is None:
+                value = ""
+            printf("%s=%s\n", key, value)

nilmdb/cmdline/remove.py

@@ -0,0 +1,59 @@
+import fnmatch
+
+from nilmdb.utils.printf import printf
+import nilmdb.client
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("remove", help="Remove data",
+                         description="""
+                         Remove all data from a specified time range within a
+                         stream.  If multiple streams or wildcards are
+                         provided, the same time range is removed from all
+                         streams.
+                         """)
+    cmd.set_defaults(handler=cmd_remove)
+
+    group = cmd.add_argument_group("Data selection")
+    group.add_argument("path", nargs='+',
+                       help="Path of stream, e.g. /foo/bar/*",
+                       ).completer = self.complete.path
+    group.add_argument("-s", "--start", required=True,
+                       metavar="TIME", type=self.arg_time,
+                       help="Starting timestamp (free-form, inclusive)",
+                       ).completer = self.complete.time
+    group.add_argument("-e", "--end", required=True,
+                       metavar="TIME", type=self.arg_time,
+                       help="Ending timestamp (free-form, noninclusive)",
+                       ).completer = self.complete.time
+
+    group = cmd.add_argument_group("Output format")
+    group.add_argument("-q", "--quiet", action="store_true",
+                       help="Don't display names when removing "
+                       "from multiple paths")
+    group.add_argument("-c", "--count", action="store_true",
+                       help="Output number of data points removed")
+    return cmd
+
+
+def cmd_remove(self):
+    streams = [s[0] for s in self.client.stream_list()]
+    paths = []
+    for path in self.args.path:
+        new = fnmatch.filter(streams, path)
+        if not new:
+            self.die("error: no stream matched path: %s", path)
+        paths.extend(new)
+
+    try:
+        for path in paths:
+            if not self.args.quiet and len(paths) > 1:
+                printf("Removing from %s\n", path)
+            count = self.client.stream_remove(path,
+                                              self.args.start, self.args.end)
+            if self.args.count:
+                printf("%d\n", count)
+    except nilmdb.client.ClientError as e:
+        self.die("error removing data: %s", str(e))
+
+    return 0

nilmdb/cmdline/rename.py

@@ -0,0 +1,32 @@
+from argparse import ArgumentDefaultsHelpFormatter as def_form
+
+import nilmdb.client
+
+
+def setup(self, sub):
+    cmd = sub.add_parser("rename", help="Rename a stream",
+                         formatter_class=def_form,
+                         description="""
+                         Rename a stream.
+
+                         Only the stream's path is renamed; no
+                         metadata is changed.
+                         """)
+    cmd.set_defaults(handler=cmd_rename)
+    group = cmd.add_argument_group("Required arguments")
+    group.add_argument("oldpath",
+                       help="Old path, e.g. /foo/old",
+                       ).completer = self.complete.path
+    group.add_argument("newpath",
+                       help="New path, e.g. /foo/bar/new",
+                       ).completer = self.complete.path
+
+    return cmd
+
+
+def cmd_rename(self):
+    """Rename a stream"""
+    try:
+        self.client.stream_rename(self.args.oldpath, self.args.newpath)
+    except nilmdb.client.ClientError as e:
+        self.die("error renaming stream: %s", str(e))

nilmdb/fileinterval.py

@@ -1,37 +0,0 @@
-"""FileInterval
-
-An Interval that is backed with file data storage"""
-
-from nilmdb.interval import Interval, IntervalSet, IntervalError
-from datetime import datetime
-import bisect
-
-class FileInterval(Interval):
-    """Represents an interval of time and its corresponding data"""
-
-    def __init__(self, start, end, 
-                 filename, 
-                 start_offset = None, end_offset = None):
-        self.start = start
-        self.end = end
-        self.filename = filename
-        if start_offset is None:
-            start_offset = 0
-        self.start_offset = start_offset
-        if end_offset is None:
-            f = open(filename, 'rb')
-            f.seek(0, os.SEEK_END)
-            end_offset = f.tell()
-        self.end_offset = end_offset
-
-    def __setattr__(self, name, value):
-        pass
-
-    def subset(self, start, end):
-        """Return a new Interval that is a subset of this one"""
-        # TODO: Any magic regarding file/offset/length mapping for subsets
-        if (start < self.start or end > self.end):
-            raise IntervalError("not a subset")
-        return FileInterval(start, end)            
-
-    

nilmdb/fsck/__init__.py

@@ -0,0 +1,3 @@
+"""nilmdb.fsck"""
+
+from nilmdb.fsck.fsck import Fsck

nilmdb/fsck/fsck.py

@@ -0,0 +1,610 @@
+# -*- coding: utf-8 -*-
+
+"""Check database consistency, with some ability to fix problems.
+This should be able to fix cases where a database gets corrupted due
+to unexpected system shutdown, and detect other cases that may cause
+NilmDB to return errors when trying to manipulate the database."""
+
+import nilmdb.utils
+import nilmdb.server
+import nilmdb.client.numpyclient
+from nilmdb.utils.interval import IntervalError
+from nilmdb.server.interval import Interval, IntervalSet
+from nilmdb.utils.printf import printf, fprintf, sprintf
+
+from collections import defaultdict
+import sqlite3
+import os
+import sys
+import progressbar
+import re
+import shutil
+import pickle
+import numpy
+
+
+class FsckError(Exception):
+    def __init__(self, msg="", *args):
+        if args:
+            msg = sprintf(msg, *args)
+        Exception.__init__(self, msg)
+
+
+class FixableFsckError(FsckError):
+    def __init__(self, msg=""):
+        FsckError.__init__(self, f'{msg}\nThis may be fixable with "--fix".')
+
+
+class RetryFsck(FsckError):
+    pass
+
+
+class FsckFormatError(FsckError):
+    pass
+
+
+def log(format, *args):
+    printf(format, *args)
+
+
+def err(format, *args):
+    fprintf(sys.stderr, format, *args)
+
+
+# Decorator that retries a function if it returns a specific value
+def retry_if_raised(exc, message=None, max_retries=1000):
+    def f1(func):
+        def f2(*args, **kwargs):
+            for n in range(max_retries):
+                try:
+                    return func(*args, **kwargs)
+                except exc:
+                    if message:
+                        log(f"{message} ({n+1})\n\n")
+            raise Exception("Max number of retries (%d) exceeded; giving up" %
+                            max_retries)
+        return f2
+    return f1
+
+
+class Progress(object):
+    def __init__(self, maxval):
+        if maxval == 0:
+            maxval = 1
+        self.bar = progressbar.ProgressBar(
+            maxval=maxval,
+            widgets=[progressbar.Percentage(), ' ',
+                     progressbar.Bar(), ' ',
+                     progressbar.ETA()])
+        self.bar.term_width = self.bar.term_width or 75
+
+    def __enter__(self):
+        self.bar.start()
+        self.last_update = 0
+        return self
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        if exc_type is None:
+            self.bar.finish()
+        else:
+            printf("\n")
+
+    def update(self, val):
+        self.bar.update(val)
+
+
+class Fsck(object):
+    def __init__(self, path, fix=False):
+        self.basepath = path
+        self.sqlpath = os.path.join(path, "data.sql")
+        self.bulkpath = os.path.join(path, "data")
+        self.bulklock = os.path.join(path, "data.lock")
+        self.fix = fix
+
+    ### Main checks
+
+    @retry_if_raised(RetryFsck, "Something was fixed: restarting fsck")
+    def check(self, skip_data=False):
+        self.bulk = None
+        self.sql = None
+        try:
+            self.check_paths()
+            self.check_sql()
+            self.check_streams()
+            self.check_intervals()
+            if skip_data:
+                log("skipped data check\n")
+            else:
+                self.check_data()
+        finally:
+            if self.bulk:
+                self.bulk.close()
+            if self.sql:  # pragma: no cover
+                # (coverage doesn't handle finally clauses correctly;
+                # both branches here are tested)
+                self.sql.commit()
+                self.sql.close()
+        log("ok\n")
+
+    ### Check basic path structure
+
+    def check_paths(self):
+        log("checking paths\n")
+        if self.bulk:
+            self.bulk.close()
+        if not os.path.isfile(self.sqlpath):
+            raise FsckError("SQL database missing (%s)", self.sqlpath)
+        if not os.path.isdir(self.bulkpath):
+            raise FsckError("Bulk data directory missing (%s)", self.bulkpath)
+        with open(self.bulklock, "w") as lockfile:
+            if not nilmdb.utils.lock.exclusive_lock(lockfile):
+                raise FsckError('Database already locked by another process\n'
+                                'Make sure all other processes that might be '
+                                'using the database are stopped.\n'
+                                'Restarting apache will cause it to unlock '
+                                'the db until a request is received.')
+            # unlocked immediately
+        self.bulk = nilmdb.server.bulkdata.BulkData(self.basepath)
+
+    ### Check SQL database health
+
+    def check_sql(self):
+        log("checking sqlite database\n")
+
+        self.sql = sqlite3.connect(self.sqlpath)
+        with self.sql:
+            cur = self.sql.cursor()
+            ver = cur.execute("PRAGMA user_version").fetchone()[0]
+            good = max(nilmdb.server.nilmdb._sql_schema_updates.keys())
+            if ver != good:
+                raise FsckError("database version %d too old, should be %d",
+                                ver, good)
+            self.stream_path = {}
+            self.stream_layout = {}
+            log("  loading paths\n")
+            result = cur.execute("SELECT id, path, layout FROM streams")
+            for r in result:
+                if r[0] in self.stream_path:
+                    raise FsckError("duplicated ID %d in stream IDs", r[0])
+                self.stream_path[r[0]] = r[1]
+                self.stream_layout[r[0]] = r[2]
+
+            log("  loading intervals\n")
+            self.stream_interval = defaultdict(list)
+            result = cur.execute("SELECT stream_id, start_time, end_time, "
+                                 "start_pos, end_pos FROM ranges "
+                                 "ORDER BY start_time")
+            for r in result:
+                if r[0] not in self.stream_path:
+                    raise FsckError("interval ID %d not in streams", r[0])
+                self.stream_interval[r[0]].append((r[1], r[2], r[3], r[4]))
+
+            log("  loading metadata\n")
+            self.stream_meta = defaultdict(dict)
+            result = cur.execute("SELECT stream_id, key, value FROM metadata")
+            for r in result:
+                if r[0] not in self.stream_path:
+                    raise FsckError("metadata ID %d not in streams", r[0])
+                if r[1] in self.stream_meta[r[0]]:
+                    raise FsckError(
+                        "duplicate metadata key '%s' for stream %d",
+                        r[1], r[0])
+                self.stream_meta[r[0]][r[1]] = r[2]
+
+    ### Check streams and basic interval overlap
+
+    def check_streams(self):
+        ids = list(self.stream_path.keys())
+        log("checking %s streams\n", "{:,d}".format(len(ids)))
+        with Progress(len(ids)) as pbar:
+            for i, sid in enumerate(ids):
+                pbar.update(i)
+                path = self.stream_path[sid]
+
+                # unique path, valid layout
+                if list(self.stream_path.values()).count(path) != 1:
+                    raise FsckError("duplicated path %s", path)
+                layout = self.stream_layout[sid].split('_')[0]
+                if layout not in ('int8', 'int16', 'int32', 'int64',
+                                  'uint8', 'uint16', 'uint32', 'uint64',
+                                  'float32', 'float64'):
+                    raise FsckError("bad layout %s for %s", layout, path)
+                count = int(self.stream_layout[sid].split('_')[1])
+                if count < 1 or count > 1024:
+                    raise FsckError("bad count %d for %s", count, path)
+
+                # must exist in bulkdata
+                bulk = self.bulkpath + path
+                bulk = bulk.encode('utf-8')
+                if not os.path.isdir(bulk):
+                    raise FsckError("%s: missing bulkdata dir", path)
+                if not nilmdb.server.bulkdata.Table.exists(bulk):
+                    raise FsckError("%s: bad bulkdata table", path)
+
+                # intervals don't overlap.  Abuse IntervalSet to check
+                # for intervals in file positions, too.
+                timeiset = IntervalSet()
+                posiset = IntervalSet()
+                for (stime, etime, spos, epos) in self.stream_interval[sid]:
+                    new = Interval(stime, etime)
+                    try:
+                        timeiset += new
+                    except IntervalError:
+                        raise FsckError("%s: overlap in intervals:\n"
+                                        "set: %s\nnew: %s",
+                                        path, str(timeiset), str(new))
+                    if spos != epos:
+                        new = Interval(spos, epos)
+                        try:
+                            posiset += new
+                        except IntervalError:
+                            self.fix_row_overlap(sid, path, posiset, new)
+
+                try:
+                    # Check bulkdata
+                    self.check_bulkdata(sid, path, bulk)
+
+                    # Check that we can open bulkdata
+                    tab = nilmdb.server.bulkdata.Table(bulk)
+                except FsckFormatError as e:
+                    # If there are no files except _format, try deleting
+                    # the entire stream; this may remove metadata, but
+                    # it's probably unimportant.
+                    files = list(os.listdir(bulk))
+                    if len(files) > 1:
+                        raise FsckFormatError(f"{path}: can't load _format, "
+                                              f"but data is also present")
+
+                    # Since the stream was empty, just remove it
+                    self.fix_remove_stream(sid, path, bulk,
+                                           "empty, with corrupted format file")
+                except FsckError as e:
+                    raise e
+                except Exception as e: # pragma: no cover
+                    # No coverage because this is an unknown/unexpected error
+                    raise FsckError("%s: can't open bulkdata: %s",
+                                    path, str(e))
+                tab.close()
+
+
+    def fix_row_overlap(self, sid, path, existing, new):
+        # If the file rows (spos, epos) overlap in the interval table,
+        # and the overlapping ranges look like this:
+        #    A --------- C
+        #           B -------- D
+        # Then we can try changing the first interval to go from
+        # A to B instead.
+        msg = (f"{path}: overlap in file offsets:\n"
+               f"existing ranges: {existing}\n"
+               f"overlapping interval: {new}")
+        if not self.fix:
+            raise FixableFsckError(msg)
+        err(f"\n{msg}\nSeeing if we can truncate one of them...\n")
+
+        # See if there'e exactly one interval that overlaps the
+        # conflicting one in the right way
+        match = None
+        for intv in self.stream_interval[sid]:
+            (stime, etime, spos, epos) = intv
+            if spos < new.start and epos > new.start:
+                if match:
+                    err(f"no, more than one interval matched:\n"
+                        f"{intv}\n{match}\n")
+                    raise FsckError(f"{path}: unfixable overlap")
+                match = intv
+        if match is None:
+            err(f"no intervals overlapped in the right way\n")
+            raise FsckError(f"{path}: unfixable overlap")
+
+        # Truncate the file position
+        err(f"truncating {match}\n")
+        with self.sql:
+            cur = self.sql.cursor()
+            cur.execute("UPDATE ranges SET end_pos=? "
+                        "WHERE stream_id=? AND start_time=? AND "
+                        "end_time=? AND start_pos=? AND end_pos=?",
+                        (new.start, sid, *match))
+            if cur.rowcount != 1:  # pragma: no cover (shouldn't fail)
+                raise FsckError("failed to fix SQL database")
+        raise RetryFsck
+
+    ### Check that bulkdata is good enough to be opened
+
+    @retry_if_raised(RetryFsck)
+    def check_bulkdata(self, sid, path, bulk):
+        try:
+            with open(os.path.join(bulk, b"_format"), "rb") as f:
+                fmt = pickle.load(f)
+        except Exception as e:
+            raise FsckFormatError(f"{path}: can't load _format file ({e})")
+
+        if fmt["version"] != 3:
+            raise FsckFormatError("%s: bad or unsupported bulkdata version %d",
+                                  path, fmt["version"])
+        rows_per_file = int(fmt["rows_per_file"])
+        if rows_per_file < 1:
+            raise FsckFormatError(f"{path}: bad rows_per_file {rows_per_file}")
+        files_per_dir = int(fmt["files_per_dir"])
+        if files_per_dir < 1:
+            raise FsckFormatError(f"{path}: bad files_per_dir {files_per_dir}")
+        layout = fmt["layout"]
+        if layout != self.stream_layout[sid]:
+            raise FsckFormatError("%s: layout mismatch %s != %s", path,
+                                  layout, self.stream_layout[sid])
+
+        # Every file should have a size that's the multiple of the row size
+        rkt = nilmdb.server.rocket.Rocket(layout, None)
+        row_size = rkt.binary_size
+        rkt.close()
+
+        # Find all directories
+        regex = re.compile(b"^[0-9a-f]{4,}$")
+        subdirs = sorted(filter(regex.search, os.listdir(bulk)),
+                         key=lambda x: int(x, 16), reverse=True)
+        for subdir in subdirs:
+            # Find all files in that dir
+            subpath = os.path.join(bulk, subdir)
+            files = list(filter(regex.search, os.listdir(subpath)))
+            if not files:
+                self.fix_empty_subdir(subpath)
+
+            # Verify that their size is a multiple of the row size
+            for filename in files:
+                filepath = os.path.join(subpath, filename)
+                offset = os.path.getsize(filepath)
+                if offset % row_size:
+                    self.fix_bad_filesize(path, filepath, offset, row_size)
+
+    def fix_empty_subdir(self, subpath):
+        msg = sprintf("bulkdata path %s is missing data files", subpath)
+        if not self.fix:
+            raise FixableFsckError(msg)
+        # Try to fix it by just deleting whatever is present,
+        # as long as it's only ".removed" files.
+        err("\n%s\n", msg)
+        for fn in os.listdir(subpath):
+            if not fn.endswith(b".removed"):
+                raise FsckError("can't fix automatically: please manually "
+                                "remove the file '%s' and try again",
+                                os.path.join(subpath, fn).decode(
+                                    'utf-8', errors='backslashreplace'))
+        # Remove the whole thing
+        err("Removing empty subpath\n")
+        shutil.rmtree(subpath)
+        raise RetryFsck
+
+    def fix_bad_filesize(self, path, filepath, offset, row_size):
+        extra = offset % row_size
+        msg = sprintf("%s: size of file %s (%d) is not a multiple" +
+                      " of row size (%d): %d extra bytes present",
+                      path, filepath, offset, row_size, extra)
+        if not self.fix:
+            raise FixableFsckError(msg)
+        # Try to fix it by just truncating the file
+        err("\n%s\n", msg)
+        newsize = offset - extra
+        err("Truncating file to %d bytes and retrying\n", newsize)
+        with open(filepath, "r+b") as f:
+            f.truncate(newsize)
+            raise RetryFsck
+
+    def fix_remove_stream(self, sid, path, bulk, reason):
+        msg = f"stream {path} is corrupted: {reason}"
+        if not self.fix:
+            raise FixableFsckError(msg)
+        # Remove the stream from disk and the database
+        err(f"\n{msg}\n")
+        err(f"Removing stream {path} from disk and database\n")
+        shutil.rmtree(bulk)
+        with self.sql:
+            cur = self.sql.cursor()
+            cur.execute("DELETE FROM streams WHERE id=?",
+                        (sid,))
+            if cur.rowcount != 1:  # pragma: no cover (shouldn't fail)
+                raise FsckError("failed to remove stream")
+            cur.execute("DELETE FROM ranges WHERE stream_id=?", (sid,))
+            cur.execute("DELETE FROM metadata WHERE stream_id=?", (sid,))
+        raise RetryFsck
+
+    ### Check interval endpoints
+
+    def check_intervals(self):
+        total_ints = sum(len(x) for x in list(self.stream_interval.values()))
+        log("checking %s intervals\n", "{:,d}".format(total_ints))
+        done = 0
+        with Progress(total_ints) as pbar:
+            for sid in self.stream_interval:
+                try:
+                    bulk = self.bulkpath + self.stream_path[sid]
+                    bulk = bulk.encode('utf-8')
+                    tab = nilmdb.server.bulkdata.Table(bulk)
+
+                    def update(x):
+                        pbar.update(done + x)
+
+                    ints = self.stream_interval[sid]
+                    done += self.check_table_intervals(sid, ints, tab, update)
+                finally:
+                    tab.close()
+
+    def check_table_intervals(self, sid, ints, tab, update):
+        # look in the table to make sure we can pick out the interval's
+        # endpoints
+        path = self.stream_path[sid]  # noqa: F841 unused
+        tab.file_open.cache_remove_all()
+        for (i, intv) in enumerate(ints):
+            update(i)
+            (stime, etime, spos, epos) = intv
+            if spos == epos and spos >= 0 and spos <= tab.nrows:
+                continue
+            try:
+                srow = tab[spos]    # noqa: F841 unused
+                erow = tab[epos-1]  # noqa: F841 unused
+            except Exception as e:
+                self.fix_bad_interval(sid, intv, tab, str(e))
+
+        return len(ints)
+
+    def fix_bad_interval(self, sid, intv, tab, msg):
+        path = self.stream_path[sid]
+        msg = sprintf("%s: interval %s error accessing rows: %s",
+                      path, str(intv), str(msg))
+        if not self.fix:
+            raise FixableFsckError(msg)
+        err("\n%s\n", msg)
+
+        (stime, etime, spos, epos) = intv
+        # If it's just that the end pos is more than the number of rows
+        # in the table, lower end pos and truncate interval time too.
+        if spos < tab.nrows and epos >= tab.nrows:
+            err("end position is past endrows, but it can be truncated\n")
+            err("old end: time %d, pos %d\n", etime, epos)
+            new_epos = tab.nrows
+            new_etime = tab[new_epos-1] + 1
+            err("new end: time %d, pos %d\n", new_etime, new_epos)
+            if stime < new_etime:
+                # Change it in SQL
+                with self.sql:
+                    cur = self.sql.cursor()
+                    cur.execute("UPDATE ranges SET end_time=?, end_pos=? "
+                                "WHERE stream_id=? AND start_time=? AND "
+                                "end_time=? AND start_pos=? AND end_pos=?",
+                                (new_etime, new_epos, sid, stime, etime,
+                                 spos, epos))
+                    if cur.rowcount != 1:  # pragma: no cover (shouldn't fail)
+                        raise FsckError("failed to fix SQL database")
+                raise RetryFsck
+            err("actually it can't be truncated; times are bad too\n")
+
+        # Otherwise, the only hope is to delete the interval entirely.
+        err("*** Deleting the entire interval from SQL.\n")
+        err("This may leave stale data on disk.  To fix that, copy all "
+            "data from this stream to a new stream using nilm-copy, then\n")
+        err("remove all data from and destroy %s.\n", path)
+        with self.sql:
+            cur = self.sql.cursor()
+            cur.execute("DELETE FROM ranges WHERE "
+                        "stream_id=? AND start_time=? AND "
+                        "end_time=? AND start_pos=? AND end_pos=?",
+                        (sid, stime, etime, spos, epos))
+            if cur.rowcount != 1:  # pragma: no cover (shouldn't fail)
+                raise FsckError("failed to remove interval")
+        raise RetryFsck
+
+    ### Check data in each interval
+
+    def check_data(self):
+        total_rows = sum(sum((y[3] - y[2]) for y in x)
+                         for x in list(self.stream_interval.values()))
+        log("checking %s rows of data\n", "{:,d}".format(total_rows))
+        done = 0
+        with Progress(total_rows) as pbar:
+            for sid in self.stream_interval:
+                try:
+                    bulk = self.bulkpath + self.stream_path[sid]
+                    bulk = bulk.encode('utf-8')
+                    tab = nilmdb.server.bulkdata.Table(bulk)
+
+                    def update(x):
+                        pbar.update(done + x)
+
+                    ints = self.stream_interval[sid]
+                    done += self.check_table_data(sid, ints, tab, update)
+                finally:
+                    tab.close()
+
+    def check_table_data(self, sid, ints, tab, update):
+        # Pull out all of the interval's data and verify that it's
+        # monotonic.
+        maxrows = getattr(self, 'maxrows_override', 100000)
+        path = self.stream_path[sid]
+        layout = self.stream_layout[sid]
+        dtype = nilmdb.client.numpyclient.layout_to_dtype(layout)
+        tab.file_open.cache_remove_all()
+        done = 0
+        for intv in ints:
+            last_ts = None
+            (stime, etime, spos, epos) = intv
+
+            # Break interval into maxrows-sized chunks
+            next_start = spos
+            while next_start < epos:
+                start = next_start
+                stop = min(start + maxrows, epos)
+                count = stop - start
+                next_start = stop
+
+                # Get raw data, convert to NumPy arary
+                try:
+                    raw = tab.get_data(start, stop, binary=True)
+                    data = numpy.frombuffer(raw, dtype)
+                except Exception as e:  # pragma: no cover
+                    # No coverage because it's hard to trigger this -- earlier
+                    # checks check the ranges, so this would probably be a real
+                    # disk error, malloc failure, etc.
+                    raise FsckError(
+                        "%s: failed to grab rows %d through %d: %s",
+                        path, start, stop, repr(e))
+
+                ts = data['timestamp']
+
+                # Verify that all timestamps are in range.
+                match = (ts < stime) | (ts >= etime)
+                if match.any():
+                    row = numpy.argmax(match)
+                    if ts[row] != 0:
+                        raise FsckError("%s: data timestamp %d at row %d "
+                                        "outside interval range [%d,%d)",
+                                        path, ts[row], row + start,
+                                        stime, etime)
+
+                    # Timestamp is zero and out of the expected range;
+                    # assume file ends with zeroed data and just truncate it.
+                    self.fix_table_by_truncating(
+                        path, tab, row + start,
+                        "data timestamp is out of range, and zero")
+
+                # Verify that timestamps are monotonic
+                match = numpy.diff(ts) <= 0
+                if match.any():
+                    row = numpy.argmax(match)
+                    if ts[row+1] != 0:
+                        raise FsckError("%s: non-monotonic timestamp (%d -> %d)"
+                                        " at row %d", path, ts[row], ts[row+1],
+                                        row + start)
+
+                    # Timestamp is zero and non-monotonic;
+                    # assume file ends with zeroed data and just truncate it.
+                    self.fix_table_by_truncating(
+                        path, tab, row + start + 1,
+                        "data timestamp is non-monotonic, and zero")
+
+                first_ts = ts[0]
+                if last_ts is not None and first_ts <= last_ts:
+                    raise FsckError("%s: first interval timestamp %d is not "
+                                    "greater than the previous last interval "
+                                    "timestamp %d, at row %d",
+                                    path, first_ts, last_ts, start)
+                last_ts = ts[-1]
+
+                # The previous errors are fixable, by removing the
+                # offending intervals, or changing the data
+                # timestamps.  But these are probably unlikely errors,
+                # so it's not worth implementing that yet.
+
+                # Done
+                done += count
+                update(done)
+        return done
+
+    def fix_table_by_truncating(self, path, tab, row, reason):
+        # Simple fix for bad data: truncate the table at the given row.
+        # On retry, fix_bad_interval will correct the database and timestamps
+        # to account for this truncation.
+        msg = f"{path}: bad data in table, starting at row {row}: {reason}"
+        if not self.fix:
+            raise FixableFsckError(msg)
+        err(f"\n{msg}\nWill try truncating table\n")
+        (subdir, fname, offs, count) = tab._offset_from_row(row)
+        tab._remove_or_truncate_file(subdir, fname, offs)
+        raise RetryFsck

nilmdb/interval.py

@@ -1,205 +0,0 @@
-"""Interval and IntervalSet
-
-Represents an interval of time, and a sorted set of such intervals"""
-
-from datetime import datetime
-import bisect
-
-class IntervalError(Exception):
-    """Error due to interval overlap, etc"""
-    pass
-
-class Interval(object):
-    """Represents an interval of time"""
-
-    start = None
-    end = None
-
-    def __init__(self, start, end):
-        self.start = start
-        self.end = end
-
-    def __repr__(self):
-        return "Interval(" + repr(self.start) + ", " + repr(self.end) + ")"
-
-    def __str__(self):
-        return "[" + str(self.start) + " -> " + str(self.end) + "]"
-
-    def __setattr__(self, name, value):
-        """Set attribute"""
-        # TODO: If we need to manipulate file names, offsets, lengths, etc,
-        # based on start and end time changing, maybe this is the right spot?
-        # Or we could just disallow changing it here.
-        if not isinstance(value, datetime):
-            raise IntervalError("Must set datetime values")
-        self.__dict__[name] = value
-        if (type(self.start) is type(self.end)):
-            if (self.start > self.end):
-                raise IntervalError("Interval start must precede interval end")
-
-    def __cmp__(self, other):
-        """Compare two intervals.  If non-equal, order by start then end"""
-        if not isinstance(other, Interval):
-            raise TypeError("Can't compare to non-interval")
-        if (self.start == other.start):
-            if (self.end < other.end):
-                return -1
-            if (self.end > other.end):
-                return 1
-            return 0
-        if (self.start < other.start):
-            return -1
-        return 1
-                
-    def intersects(self, other):
-        """Return True if two Interval objects intersect"""
-        if (not isinstance(other, Interval)):
-            raise TypeError("need Interval for intersection test")
-        if (self.end <= other.start or
-            self.start >= other.end):
-            return False
-        else:
-            return True
-
-    def is_adjacent(self, other):
-        """Return True if two Intervals are adjacent (same end or start)"""
-        if (not isinstance(other, Interval)):
-            raise TypeError("need Interval for adjacency test")
-        if (self.end == other.start or
-            self.start == other.end):
-            return True
-        else:
-            return False
-
-    def subset(self, start, end):
-        """Return a new Interval that is a subset of this one"""
-        # TODO: Any magic regarding file/offset/length mapping for subsets
-        if (start < self.start or end > self.end):
-            raise IntervalError("not a subset")
-        return Interval(start, end)            
-
-class IntervalSet(object):
-    """A non-intersecting set of intervals
-
-    Kept sorted internally"""
-
-    def __init__(self, iterable=None):
-        self.data = []
-        if iterable is not None:
-            if isinstance(iterable, Interval):
-                iterable = [iterable]
-            self._add_intervals(iterable)
-
-    def __iter__(self):
-        return self.data.__iter__()
-
-    def __repr__(self):
-        return "IntervalSet(" + repr(list(self.data)) + ")"
-
-    def __cmp__(self, other):
-        # compare isn't supported, they don't really have an ordering
-        raise TypeError("can't compare IntervalSets with cmp()")
-
-    def __eq__(self, other):
-        """Test equality of two IntervalSets.
-
-        Treats adjacent Intervals as equivalent to one long interval,
-        so this function really tests whether the IntervalSets cover
-        the same spans of time."""
-        if not isinstance(other, IntervalSet):
-            return False
-        i = 0
-        j = 0
-        outside = True
-        try:
-            while True:
-                if (outside):
-                    # To match, we need to be finished this set
-                    if (i >= len(self) and j >= len(other)):
-                        return True
-                    # Or the starts need to match
-                    if (self[i].start != other[j].start):
-                        return False
-                    outside = False
-                else:
-                    # We can move on if the two interval ends match
-                    if (self[i].end == other[j].end):
-                        i += 1
-                        j += 1
-                        outside = True
-                    else:
-                        # Whichever ends first needs to be adjacent to the next
-                        if (self[i].end < other[j].end):
-                            if (not self[i].is_adjacent(self[i+1])):
-                                return False
-                            i += 1
-                        else:
-                            if (not other[j].is_adjacent(other[j+1])):
-                                return False
-                            j += 1
-        except IndexError:
-            return False
-
-    def __ne__(self, other):
-        return not self.__eq__(other)
-
-    def __len__(self):
-        return len(self.data)
-
-    def __getitem__(self, key):
-        return self.data.__getitem__(key)
-
-    def __iadd__(self, other):
-        """Inplace add -- modifies self
-
-        This throws an exception if the regions being added intersect."""
-        if isinstance(other, Interval):
-            other = [other]
-        self._add_intervals(other)
-        return self
-        
-    def __add__(self, other):
-        """Add -- returns a new object
-
-        This throws an exception if the regions being added intersect."""
-        new = IntervalSet(self)
-        new += IntervalSet(other)
-        return new
-
-    def __and__(self, other):
-        """Compute a new IntervalSet from the intersection of two others
-
-        Output intervals are built as subsets of the intervals in the
-        first argument (self)."""
-        # If we were given a set, intersect with each interval in that set
-        if isinstance(other, IntervalSet):
-            out = IntervalSet()
-            for interval in other.data:
-                out += self & interval
-            return out
-
-        if not isinstance(other, Interval):
-            raise TypeError("can't intersect with that type")
-
-        out = IntervalSet()
-        for this in self.data:
-            # If there's any overlap, add the overlapping region
-            if (this.end > other.start and this.start < other.end):
-                out += this.subset(max(this.start, other.start),
-                                   min(this.end, other.end))
-        return out
-    
-    def _add_intervals(self, iterable):
-        """Add each Interval from an interable to this set"""
-        for element in iter(iterable):
-            self._add_single_interval(element)
-
-    def _add_single_interval(self, interval):
-        """Add one Interval to this set"""
-        if (not isinstance(interval, Interval)):
-            raise TypeError("can only add Intervals")
-        for existing in self.data:
-            if existing.intersects(interval):
-                raise IntervalError("Tried to add overlapping interval "
-                                    "to this set")
-        bisect.insort(self.data, interval)

nilmdb/scripts/__init__.py

@@ -0,0 +1 @@
+# Command line scripts

nilmdb/scripts/nilmdb_fsck.py

@@ -0,0 +1,27 @@
+#!/usr/bin/env python3
+
+import nilmdb.fsck
+import argparse
+
+
+def main():
+    """Main entry point for the 'nilmdb-fsck' command line script"""
+
+    parser = argparse.ArgumentParser(
+        description='Check database consistency',
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+    parser.add_argument("-v", "--version", action="version",
+                        version=nilmdb.__version__)
+    parser.add_argument("-f", "--fix", action="store_true",
+                        default=False, help='Fix errors when possible '
+                        '(which may involve removing data)')
+    parser.add_argument("-n", "--no-data", action="store_true",
+                        default=False, help='Skip the slow full-data check')
+    parser.add_argument('database', help='Database directory')
+    args = parser.parse_args()
+
+    nilmdb.fsck.Fsck(args.database, args.fix).check(skip_data=args.no_data)
+
+
+if __name__ == "__main__":
+    main()

nilmdb/scripts/nilmdb_server.py

@@ -0,0 +1,99 @@
+#!/usr/bin/env python3
+
+import os
+import sys
+import socket
+import argparse
+
+import cherrypy
+
+import nilmdb.server
+
+
+def main():
+    """Main entry point for the 'nilmdb-server' command line script"""
+
+    parser = argparse.ArgumentParser(
+        description='Run the NilmDB server',
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+
+    parser.add_argument("-v", "--version", action="version",
+                        version=nilmdb.__version__)
+
+    group = parser.add_argument_group("Standard options")
+    group.add_argument('-a', '--address',
+                       help='Only listen on the given address',
+                       default='0.0.0.0')
+    group.add_argument('-p', '--port', help='Listen on the given port',
+                       type=int, default=12380)
+    group.add_argument('-d', '--database', help='Database directory',
+                       default="./db")
+    group.add_argument('-q', '--quiet', help='Silence output',
+                       action='store_true')
+    group.add_argument('-t', '--traceback',
+                       help='Provide tracebacks in client errors',
+                       action='store_true', default=False)
+
+    group = parser.add_argument_group("Debug options")
+    group.add_argument('-y', '--yappi', help='Run under yappi profiler and '
+                       'invoke interactive shell afterwards',
+                       action='store_true')
+
+    args = parser.parse_args()
+
+    # Create database object.  Needs to be serialized before passing
+    # to the Server.
+    db = nilmdb.utils.serializer_proxy(nilmdb.server.NilmDB)(args.database)
+
+    # Configure the server
+    if not args.quiet:
+        cherrypy._cpconfig.environments['embedded']['log.screen'] = True
+
+    server = nilmdb.server.Server(db,
+                                  host=args.address,
+                                  port=args.port,
+                                  force_traceback=args.traceback)
+
+    # Print info
+    if not args.quiet:
+        print("Version: %s" % nilmdb.__version__)
+        print("Database: %s" % (os.path.realpath(args.database)))
+        if args.address == '0.0.0.0' or args.address == '::':
+            host = socket.getfqdn()
+        else:
+            host = args.address
+        print("Server URL: http://%s:%d/" % (host, args.port))
+        print("----")
+
+    # Run it
+    try:
+        if args.yappi:
+            print("Running in yappi")
+            try:
+                import yappi
+                yappi.start()
+                server.start(blocking=True)
+            finally:
+                yappi.stop()
+                stats = yappi.get_func_stats()
+                stats.sort("ttot")
+                stats.print_all()
+                try:
+                    from IPython import embed
+                    embed(header="Use the `yappi` or `stats` object to "
+                          "explore further, `quit` to exit")
+                except ModuleNotFoundError:
+                    print("\nInstall ipython to explore further")
+        else:
+            server.start(blocking=True)
+    except nilmdb.server.serverutil.CherryPyExit:
+        print("Exiting due to CherryPy error", file=sys.stderr)
+        raise
+    finally:
+        if not args.quiet:
+            print("Closing database")
+            db.close()
+
+
+if __name__ == "__main__":
+    main()

nilmdb/scripts/nilmtool.py

@@ -0,0 +1,12 @@
+#!/usr/bin/env python3
+
+import nilmdb.cmdline
+
+
+def main():
+    """Main entry point for the 'nilmtool' command line script"""
+    nilmdb.cmdline.Cmdline().run()
+
+
+if __name__ == "__main__":
+    main()

nilmdb/server/__init__.py

@@ -0,0 +1,9 @@
+"""nilmdb.server"""
+
+# Set up pyximport to automatically rebuild Cython modules if needed.
+import pyximport
+pyximport.install(inplace=True, build_in_temp=False)
+
+from nilmdb.server.nilmdb import NilmDB
+from nilmdb.server.server import Server, wsgi_application
+from nilmdb.server.errors import NilmDBError, StreamError, OverlapError

nilmdb/server/bulkdata.py

@@ -0,0 +1,635 @@
+# Fixed record size bulk data storage
+
+import os
+import re
+import sys
+import pickle
+import tempfile
+
+from nilmdb.utils.printf import sprintf
+from nilmdb.utils.time import timestamp_to_string
+import nilmdb.utils
+
+import nilmdb.utils.lock
+from . import rocket
+
+# Up to 256 open file descriptors at any given time.
+# These variables are global so they can be used in the decorator arguments.
+table_cache_size = 32
+fd_cache_size = 8
+
+
+@nilmdb.utils.must_close(wrap_verify=False)
+class BulkData():
+    def __init__(self, basepath, **kwargs):
+        if isinstance(basepath, str):
+            self.basepath = self._encode_filename(basepath)
+        else:
+            self.basepath = basepath
+        self.root = os.path.join(self.basepath, b"data")
+        self.lock = self.root + b".lock"
+        self.lockfile = None
+
+        # Tuneables
+        if "file_size" in kwargs and kwargs["file_size"] is not None:
+            self.file_size = kwargs["file_size"]
+        else:
+            # Default to approximately 128 MiB per file
+            self.file_size = 128 * 1024 * 1024
+
+        if "files_per_dir" in kwargs and kwargs["files_per_dir"] is not None:
+            self.files_per_dir = kwargs["files_per_dir"]
+        else:
+            # 32768 files per dir should work even on FAT32
+            self.files_per_dir = 32768
+
+        if "initial_nrows" in kwargs and kwargs["initial_nrows"] is not None:
+            self.initial_nrows = kwargs["initial_nrows"]
+        else:
+            # First row is 0
+            self.initial_nrows = 0
+
+        # Make root path
+        if not os.path.isdir(self.root):
+            os.mkdir(self.root)
+
+        # Create the lock
+        self.lockfile = open(self.lock, "w")
+        if not nilmdb.utils.lock.exclusive_lock(self.lockfile):
+            raise IOError('database at "' +
+                          self._decode_filename(self.basepath) +
+                          '" is already locked by another process')
+
+    def close(self):
+        self.getnode.cache_remove_all()
+        if self.lockfile:
+            nilmdb.utils.lock.exclusive_unlock(self.lockfile)
+            self.lockfile.close()
+            try:
+                os.unlink(self.lock)
+            except OSError:
+                pass
+            self.lockfile = None
+
+    def _encode_filename(self, path):
+        # Translate unicode strings to raw bytes, if needed.  We
+        # always manipulate paths internally as bytes.
+        return path.encode('utf-8')
+
+    def _decode_filename(self, path):
+        # Translate raw bytes to unicode strings, escaping if needed
+        return path.decode('utf-8', errors='backslashreplace')
+
+    def _create_check_ospath(self, ospath):
+        if ospath[-1:] == b'/':
+            raise ValueError("invalid path; should not end with a /")
+        if Table.exists(ospath):
+            raise ValueError("stream already exists at this path")
+        if os.path.isdir(ospath):
+            # Look for any files in subdirectories.  Fully empty subdirectories
+            # are OK; they might be there during a rename
+            for (root, dirs, files) in os.walk(ospath):
+                if files:
+                    raise ValueError(
+                        "non-empty subdirs of this path already exist")
+
+    def _create_parents(self, unicodepath):
+        """Verify the path name, and create parent directories if they
+        don't exist.  Returns a list of elements that got created."""
+        path = self._encode_filename(unicodepath)
+
+        if path[0:1] != b'/':
+            raise ValueError("paths must start with / ")
+        [group, node] = path.rsplit(b"/", 1)
+        if group == b'':
+            raise ValueError("invalid path; path must contain at least one "
+                             "folder")
+        if node == b'':
+            raise ValueError("invalid path; should not end with a /")
+        if not Table.valid_path(path):
+            raise ValueError("path name is invalid or contains reserved words")
+
+        # Create the table's base dir.  Note that we make a
+        # distinction here between NilmDB paths (always Unix style,
+        # split apart manually) and OS paths (built up with
+        # os.path.join)
+
+        # Make directories leading up to this one
+        elements = path.lstrip(b'/').split(b'/')
+        made_dirs = []
+        try:
+            # Make parent elements
+            for i in range(len(elements)):
+                ospath = os.path.join(self.root, *elements[0:i])
+                if Table.exists(ospath):
+                    raise ValueError("path is subdir of existing node")
+                if not os.path.isdir(ospath):
+                    os.mkdir(ospath)
+                    made_dirs.append(ospath)
+        except Exception:
+            # Remove paths that we created
+            for ospath in reversed(made_dirs):
+                os.rmdir(ospath)
+            raise
+
+        return elements
+
+    def create(self, unicodepath, layout_name):
+        """
+        unicodepath: path to the data (e.g. u'/newton/prep').
+        Paths must contain at least two elements, e.g.:
+           /newton/prep
+           /newton/raw
+           /newton/upstairs/prep
+           /newton/upstairs/raw
+
+        layout_name: string for nilmdb.layout.get_named(), e.g. 'float32_8'
+        """
+        elements = self._create_parents(unicodepath)
+
+        # Make the final dir
+        ospath = os.path.join(self.root, *elements)
+        self._create_check_ospath(ospath)
+        os.mkdir(ospath)
+
+        try:
+            # Write format string to file
+            Table.create(ospath, layout_name, self.file_size,
+                         self.files_per_dir)
+
+            # Open and cache it
+            self.getnode(unicodepath)
+        except Exception:
+            exc_info = sys.exc_info()
+            try:
+                os.rmdir(ospath)
+            except OSError:
+                pass
+            raise exc_info[1].with_traceback(exc_info[2])
+
+        # Success
+        return
+
+    def _remove_leaves(self, unicodepath):
+        """Remove empty directories starting at the leaves of unicodepath"""
+        path = self._encode_filename(unicodepath)
+        elements = path.lstrip(b'/').split(b'/')
+        for i in reversed(list(range(len(elements)))):
+            ospath = os.path.join(self.root, *elements[0:i+1])
+            try:
+                os.rmdir(ospath)
+            except OSError:
+                pass
+
+    def rename(self, oldunicodepath, newunicodepath):
+        """Move entire tree from 'oldunicodepath' to
+        'newunicodepath'"""
+        oldpath = self._encode_filename(oldunicodepath)
+        newpath = self._encode_filename(newunicodepath)
+
+        # Get OS paths
+        oldelements = oldpath.lstrip(b'/').split(b'/')
+        oldospath = os.path.join(self.root, *oldelements)
+        newelements = newpath.lstrip(b'/').split(b'/')
+        newospath = os.path.join(self.root, *newelements)
+
+        # Basic checks
+        if oldospath == newospath:
+            raise ValueError("old and new paths are the same")
+
+        # Remove Table object at old path from cache
+        self.getnode.cache_remove(self, oldunicodepath)
+
+        # Move the table to a temporary location
+        tmpdir = tempfile.mkdtemp(prefix=b"rename-", dir=self.root)
+        tmppath = os.path.join(tmpdir, b"table")
+        os.rename(oldospath, tmppath)
+
+        try:
+            # Check destination path
+            self._create_check_ospath(newospath)
+
+            # Create parent dirs for new location
+            self._create_parents(newunicodepath)
+
+            # Move table into new location
+            os.rename(tmppath, newospath)
+        except Exception:
+            # On failure, move the table back to original path
+            os.rename(tmppath, oldospath)
+            os.rmdir(tmpdir)
+            raise
+
+        # Prune old dirs
+        self._remove_leaves(oldunicodepath)
+        os.rmdir(tmpdir)
+
+    def destroy(self, unicodepath):
+        """Fully remove all data at a particular path.  No way to undo
+        it!  The group/path structure is removed, too."""
+        path = self._encode_filename(unicodepath)
+
+        # Get OS path
+        elements = path.lstrip(b'/').split(b'/')
+        ospath = os.path.join(self.root, *elements)
+
+        # Remove Table object from cache
+        self.getnode.cache_remove(self, unicodepath)
+
+        # Remove the contents of the target directory
+        if not Table.exists(ospath):
+            raise ValueError("nothing at that path")
+        for (root, dirs, files) in os.walk(ospath, topdown=False):
+            for name in files:
+                os.remove(os.path.join(root, name))
+            for name in dirs:
+                os.rmdir(os.path.join(root, name))
+
+        # Remove leftover empty directories
+        self._remove_leaves(unicodepath)
+
+    # Cache open tables
+    @nilmdb.utils.lru_cache(size=table_cache_size,
+                            onremove=lambda x: x.close())
+    def getnode(self, unicodepath):
+        """Return a Table object corresponding to the given database
+        path, which must exist."""
+        path = self._encode_filename(unicodepath)
+        elements = path.lstrip(b'/').split(b'/')
+        ospath = os.path.join(self.root, *elements)
+        return Table(ospath, self.initial_nrows)
+
+
+@nilmdb.utils.must_close(wrap_verify=False)
+class Table():
+    """Tools to help access a single table (data at a specific OS path)."""
+    # See design.md for design details
+
+    # Class methods, to help keep format details in this class.
+    @classmethod
+    def valid_path(cls, root):
+        """Return True if a root path is a valid name"""
+        return b"_format" not in root.split(b"/")
+
+    @classmethod
+    def exists(cls, root):
+        """Return True if a table appears to exist at this OS path"""
+        return os.path.isfile(os.path.join(root, b"_format"))
+
+    @classmethod
+    def create(cls, root, layout, file_size, files_per_dir):
+        """Initialize a table at the given OS path with the
+        given layout string"""
+
+        # Calculate rows per file so that each file is approximately
+        # file_size bytes.
+        rkt = rocket.Rocket(layout, None)
+        rows_per_file = max(file_size // rkt.binary_size, 1)
+        rkt.close()
+
+        fmt = {
+            "rows_per_file": rows_per_file,
+            "files_per_dir": files_per_dir,
+            "layout": layout,
+            "version": 3
+        }
+        nilmdb.utils.atomic.replace_file(
+            os.path.join(root, b"_format"), pickle.dumps(fmt, 2))
+
+    # Normal methods
+    def __init__(self, root, initial_nrows=0):
+        """'root' is the full OS path to the directory of this table"""
+        self.root = root
+        self.initial_nrows = initial_nrows
+
+        # Load the format
+        with open(os.path.join(self.root, b"_format"), "rb") as f:
+            fmt = pickle.load(f)
+
+        if fmt["version"] != 3:
+            # Old versions used floating point timestamps, which aren't
+            # valid anymore.
+            raise NotImplementedError("old version " + str(fmt["version"]) +
+                                      " bulk data store is not supported")
+
+        self.rows_per_file = fmt["rows_per_file"]
+        self.files_per_dir = fmt["files_per_dir"]
+        self.layout = fmt["layout"]
+
+        # Use rocket to get row size and file size
+        rkt = rocket.Rocket(self.layout, None)
+        self.row_size = rkt.binary_size
+        self.file_size = rkt.binary_size * self.rows_per_file
+        rkt.close()
+
+        # Find nrows
+        self.nrows = self._get_nrows()
+
+    def close(self):
+        self.file_open.cache_remove_all()
+
+    # Internal helpers
+    def _get_nrows(self):
+        """Find nrows by locating the lexicographically last filename
+        and using its size"""
+        # Note that this just finds a 'nrows' that is guaranteed to be
+        # greater than the row number of any piece of data that
+        # currently exists, not necessarily all data that _ever_
+        # existed.
+        regex = re.compile(b"^[0-9a-f]{4,}$")
+
+        # Find the last directory.  We sort and loop through all of them,
+        # starting with the numerically greatest, because the dirs could be
+        # empty if something was deleted but the directory was unexpectedly
+        # not deleted.
+        subdirs = sorted(filter(regex.search, os.listdir(self.root)),
+                         key=lambda x: int(x, 16), reverse=True)
+
+        for subdir in subdirs:
+            # Now find the last file in that dir
+            path = os.path.join(self.root, subdir)
+            files = list(filter(regex.search, os.listdir(path)))
+            if not files:
+                # Empty dir: try the next one
+                continue
+
+            # Find the numerical max
+            filename = max(files, key=lambda x: int(x, 16))
+            offset = os.path.getsize(os.path.join(self.root, subdir, filename))
+
+            # Convert to row number
+            return self._row_from_offset(subdir, filename, offset)
+
+        # No files, so no data.  We typically start at row 0 in this
+        # case, although initial_nrows is specified during some tests
+        # to exercise other parts of the code better.  Since we have
+        # no files yet, round initial_nrows up so it points to a row
+        # that would begin a new file.
+        nrows = ((self.initial_nrows + (self.rows_per_file - 1)) //
+                 self.rows_per_file) * self.rows_per_file
+        return nrows
+
+    def _offset_from_row(self, row):
+        """Return a (subdir, filename, offset, count) tuple:
+
+          subdir: subdirectory for the file
+        filename: the filename that contains the specified row
+          offset: byte offset of the specified row within the file
+           count: number of rows (starting at offset) that fit in the file
+        """
+        filenum = row // self.rows_per_file
+        # It's OK if these format specifiers are too short; the filenames
+        # will just get longer but will still sort correctly.
+        dirname = sprintf(b"%04x", filenum // self.files_per_dir)
+        filename = sprintf(b"%04x", filenum % self.files_per_dir)
+        offset = (row % self.rows_per_file) * self.row_size
+        count = self.rows_per_file - (row % self.rows_per_file)
+        return (dirname, filename, offset, count)
+
+    def _row_from_offset(self, subdir, filename, offset):
+        """Return the row number that corresponds to the given
+        'subdir/filename' and byte-offset within that file."""
+        if (offset % self.row_size) != 0:
+            # this shouldn't occur, unless there is some corruption somewhere
+            raise ValueError("file offset is not a multiple of data size")
+        filenum = int(subdir, 16) * self.files_per_dir + int(filename, 16)
+        row = (filenum * self.rows_per_file) + (offset // self.row_size)
+        return row
+
+    def _remove_or_truncate_file(self, subdir, filename, offset=0):
+        """Remove the given file, and remove the subdirectory too
+        if it's empty.  If offset is nonzero, truncate the file
+        to that size instead."""
+        # Close potentially open file in file_open LRU cache
+        self.file_open.cache_remove(self, subdir, filename)
+        if offset:
+            # Truncate it
+            with open(os.path.join(self.root, subdir, filename), "r+b") as f:
+                f.truncate(offset)
+        else:
+            # Remove file
+            os.remove(os.path.join(self.root, subdir, filename))
+            # Try deleting subdir, too
+            try:
+                os.rmdir(os.path.join(self.root, subdir))
+            except Exception:
+                pass
+
+    # Cache open files
+    @nilmdb.utils.lru_cache(size=fd_cache_size,
+                            onremove=lambda f: f.close())
+    def file_open(self, subdir, filename):
+        """Open and map a given 'subdir/filename' (relative to self.root).
+        Will be automatically closed when evicted from the cache."""
+        # Create path if it doesn't exist
+        try:
+            os.mkdir(os.path.join(self.root, subdir))
+        except OSError:
+            pass
+        # Return a rocket.Rocket object, which contains the open file
+        return rocket.Rocket(self.layout,
+                             os.path.join(self.root, subdir, filename))
+
+    def append_data(self, data, start, end, binary=False):
+        """Parse the formatted string in 'data', according to the
+        current layout, and append it to the table.  If any timestamps
+        are non-monotonic, or don't fall between 'start' and 'end',
+        a ValueError is raised.
+
+        Note that data is always of 'bytes' type.
+
+        If 'binary' is True, the data should be in raw binary format
+        instead: little-endian, matching the current table's layout,
+        including the int64 timestamp.
+
+        If this function succeeds, it returns normally.  Otherwise,
+        the table is reverted back to its original state by truncating
+        or deleting files as necessary."""
+        data_offset = 0
+        last_timestamp = nilmdb.utils.time.min_timestamp
+        tot_rows = self.nrows
+        count = 0
+        linenum = 0
+        try:
+            while data_offset < len(data):
+                # See how many rows we can fit into the current file,
+                # and open it
+                (subdir, fname, offs, count) = self._offset_from_row(tot_rows)
+                f = self.file_open(subdir, fname)
+
+                # Ask the rocket object to parse and append up to "count"
+                # rows of data, verifying things along the way.
+                try:
+                    if binary:
+                        appender = f.append_binary
+                    else:
+                        appender = f.append_string
+                    (added_rows, data_offset, last_timestamp, linenum
+                     ) = appender(count, data, data_offset, linenum,
+                                  start, end, last_timestamp)
+                except rocket.ParseError as e:
+                    (linenum, colnum, errtype, obj) = e.args
+                    if binary:
+                        where = "byte %d: " % (linenum)
+                    else:
+                        where = "line %d, column %d: " % (linenum, colnum)
+                    # Extract out the error line, add column marker
+                    try:
+                        if binary:
+                            raise IndexError
+                        bad = data.splitlines()[linenum-1]
+                        bad += b'\n' + b' ' * (colnum - 1) + b'^'
+                    except IndexError:
+                        bad = b""
+                    if errtype == rocket.ERR_NON_MONOTONIC:
+                        err = "timestamp is not monotonically increasing"
+                    elif errtype == rocket.ERR_OUT_OF_INTERVAL:
+                        if obj < start:
+                            err = sprintf("Data timestamp %s < start time %s",
+                                          timestamp_to_string(obj),
+                                          timestamp_to_string(start))
+                        else:
+                            err = sprintf("Data timestamp %s >= end time %s",
+                                          timestamp_to_string(obj),
+                                          timestamp_to_string(end))
+                    else:
+                        err = str(obj)
+                    bad_str = bad.decode('utf-8', errors='backslashreplace')
+                    raise ValueError("error parsing input data: " +
+                                     where + err + "\n" + bad_str)
+                tot_rows += added_rows
+        except Exception:
+            # Some failure, so try to roll things back by truncating or
+            # deleting files that we may have appended data to.
+            cleanpos = self.nrows
+            while cleanpos <= tot_rows:
+                (subdir, fname, offs, count) = self._offset_from_row(cleanpos)
+                self._remove_or_truncate_file(subdir, fname, offs)
+                cleanpos += count
+            # Re-raise original exception
+            raise
+        else:
+            # Success, so update self.nrows accordingly
+            self.nrows = tot_rows
+
+    def get_data(self, start, stop, binary=False):
+        """Extract data corresponding to Python range [n:m],
+        and returns a formatted string"""
+        if (start is None or stop is None or
+                start > stop or start < 0 or stop > self.nrows):
+            raise IndexError("Index out of range")
+
+        ret = []
+        row = start
+        remaining = stop - start
+        while remaining > 0:
+            (subdir, filename, offset, count) = self._offset_from_row(row)
+            if count > remaining:
+                count = remaining
+            f = self.file_open(subdir, filename)
+            if binary:
+                ret.append(f.extract_binary(offset, count))
+            else:
+                ret.append(f.extract_string(offset, count))
+            remaining -= count
+            row += count
+        return b"".join(ret)
+
+    def __getitem__(self, row):
+        """Extract timestamps from a row, with table[n] notation."""
+        if row < 0 or row >= self.nrows:
+            raise IndexError("Index out of range")
+        (subdir, filename, offset, count) = self._offset_from_row(row)
+        f = self.file_open(subdir, filename)
+        return f.extract_timestamp(offset)
+
+    def _remove_rows(self, subdir, filename, start, stop):
+        """Helper to mark specific rows as being removed from a
+        file, and potentially remove or truncate the file itself."""
+        # Close potentially open file in file_open LRU cache
+        self.file_open.cache_remove(self, subdir, filename)
+
+        # We keep a file like 0000.removed that contains a list of
+        # which rows have been "removed".  Note that we never have to
+        # remove entries from this list, because we never decrease
+        # self.nrows, and so we will never overwrite those locations in the
+        # file.  Only when the list covers the entire extent of the
+        # file will that file be removed.
+        datafile = os.path.join(self.root, subdir, filename)
+        cachefile = datafile + b".removed"
+        try:
+            with open(cachefile, "rb") as f:
+                ranges = pickle.load(f)
+            cachefile_present = True
+        except Exception:
+            ranges = []
+            cachefile_present = False
+
+        # Append our new range and sort
+        ranges.append((start, stop))
+        ranges.sort()
+
+        # Merge adjacent ranges into "out"
+        merged = []
+        prev = None
+        for new in ranges:
+            if prev is None:
+                # No previous range, so remember this one
+                prev = new
+            elif prev[1] == new[0]:
+                # Previous range connected to this new one; extend prev
+                prev = (prev[0], new[1])
+            else:
+                # Not connected; append previous and start again
+                merged.append(prev)
+                prev = new
+        # Last range we were looking at goes into the file.  We know
+        # there was at least one (the one we just removed).
+        merged.append(prev)
+
+        # If the range covered the whole file, we can delete it now.
+        # Note that the last file in a table may be only partially
+        # full (smaller than self.rows_per_file).  We purposely leave
+        # those files around rather than deleting them, because the
+        # remainder will be filled on a subsequent append(), and things
+        # are generally easier if we don't have to special-case that.
+        if (len(merged) == 1 and
+                merged[0][0] == 0 and merged[0][1] == self.rows_per_file):
+            # Delete files
+            if cachefile_present:
+                os.remove(cachefile)
+            self._remove_or_truncate_file(subdir, filename, 0)
+        else:
+            # File needs to stick around.  This means we can get
+            # degenerate cases where we have large files containing as
+            # little as one row.  Try to punch a hole in the file,
+            # so that this region doesn't take up filesystem space.
+            offset = start * self.row_size
+            count = (stop - start) * self.row_size
+            nilmdb.utils.fallocate.punch_hole(datafile, offset, count)
+
+            # Update cache.  Try to do it atomically.
+            nilmdb.utils.atomic.replace_file(cachefile,
+                                             pickle.dumps(merged, 2))
+
+    def remove(self, start, stop):
+        """Remove specified rows [start, stop) from this table.
+
+        If a file is left empty, it is fully removed.  Otherwise, a
+        parallel data file is used to remember which rows have been
+        removed, and the file is otherwise untouched."""
+        if start < 0 or start > stop or stop > self.nrows:
+            raise IndexError("Index out of range")
+
+        row = start
+        remaining = stop - start
+        while remaining:
+            # Loop through each file that we need to touch
+            (subdir, filename, offset, count) = self._offset_from_row(row)
+            if count > remaining:
+                count = remaining
+            row_offset = offset // self.row_size
+            # Mark the rows as being removed
+            self._remove_rows(subdir, filename, row_offset, row_offset + count)
+            remaining -= count
+            row += count

nilmdb/server/errors.py

@@ -0,0 +1,15 @@
+"""Exceptions"""
+
+
+class NilmDBError(Exception):
+    """Base exception for NilmDB errors"""
+    def __init__(self, msg="Unspecified error"):
+        super().__init__(msg)
+
+
+class StreamError(NilmDBError):
+    pass
+
+
+class OverlapError(NilmDBError):
+    pass

nilmdb/server/interval.pyx

@@ -0,0 +1,329 @@
+# cython: language_level=2
+
+"""Interval, IntervalSet
+
+The Interval implemented here is just like
+nilmdb.utils.interval.Interval, except implemented in Cython for
+speed.
+
+Represents an interval of time, and a set of such intervals.
+
+Intervals are half-open, ie. they include data points with timestamps
+[start, end)
+"""
+
+# First implementation kept a sorted list of intervals and used
+# biesct() to optimize some operations, but this was too slow.
+
+# Second version was based on the quicksect implementation from
+# python-bx, modified slightly to handle floating point intervals.
+# This didn't support deletion.
+
+# Third version is more similar to the first version, using a rb-tree
+# instead of a simple sorted list to maintain O(log n) operations.
+
+# Fourth version is an optimized rb-tree that stores interval starts
+# and ends directly in the tree, like bxinterval did.
+
+from ..utils.time import min_timestamp as nilmdb_min_timestamp
+from ..utils.time import max_timestamp as nilmdb_max_timestamp
+from ..utils.time import timestamp_to_string
+from ..utils.iterator import imerge
+from ..utils.interval import IntervalError
+import itertools
+
+cimport rbtree
+from libc.stdint cimport uint64_t, int64_t
+
+ctypedef int64_t timestamp_t
+
+cdef class Interval:
+    """Represents an interval of time."""
+
+    cdef public timestamp_t start, end
+
+    def __init__(self, timestamp_t start, timestamp_t end):
+        """
+        'start' and 'end' are arbitrary numbers that represent time
+        """
+        if start >= end:
+            # Explicitly disallow zero-width intervals (since they're half-open)
+            raise IntervalError("start %s must precede end %s" % (start, end))
+        self.start = start
+        self.end = end
+
+    def __repr__(self):
+        s = repr(self.start) + ", " + repr(self.end)
+        return self.__class__.__name__ + "(" + s + ")"
+
+    def __str__(self):
+        return ("[" + timestamp_to_string(self.start) +
+                " -> " + timestamp_to_string(self.end) + ")")
+
+    # Compare two intervals.  If non-equal, order by start then end
+    def __lt__(self, Interval other):
+        return (self.start, self.end) < (other.start, other.end)
+    def __gt__(self, Interval other):
+        return (self.start, self.end) > (other.start, other.end)
+    def __le__(self, Interval other):
+        return (self.start, self.end) <= (other.start, other.end)
+    def __ge__(self, Interval other):
+        return (self.start, self.end) >= (other.start, other.end)
+    def __eq__(self, Interval other):
+        return (self.start, self.end) == (other.start, other.end)
+    def __ne__(self, Interval other):
+        return (self.start, self.end) != (other.start, other.end)
+
+    cpdef intersects(self, Interval other):
+        """Return True if two Interval objects intersect"""
+        if (self.end <= other.start or self.start >= other.end):
+            return False
+        return True
+
+    cpdef subset(self, timestamp_t start, timestamp_t end):
+        """Return a new Interval that is a subset of this one"""
+        # A subclass that tracks additional data might override this.
+        if start < self.start or end > self.end:
+            raise IntervalError("not a subset")
+        return Interval(start, end)
+
+cdef class DBInterval(Interval):
+    """
+    Like Interval, but also tracks corresponding start/end times and
+    positions within the database.  These are not currently modified
+    when subsets are taken, but can be used later to help zero in on
+    database positions.
+
+    The actual 'start' and 'end' will always fall within the database
+    start and end, e.g.:
+        db_start = 100, db_startpos = 10000
+        start = 123
+        end = 150
+        db_end = 200, db_endpos = 20000
+    """
+
+    cpdef public timestamp_t db_start, db_end
+    cpdef public uint64_t db_startpos, db_endpos
+
+    def __init__(self, start, end,
+                 db_start, db_end,
+                 db_startpos, db_endpos):
+        """
+        'db_start' and 'db_end' are arbitrary numbers that represent
+        time.  They must be a strict superset of the time interval
+        covered by 'start' and 'end'.  The 'db_startpos' and
+        'db_endpos' are arbitrary database position indicators that
+        correspond to those points.
+        """
+        Interval.__init__(self, start, end)
+        self.db_start = db_start
+        self.db_end = db_end
+        self.db_startpos = db_startpos
+        self.db_endpos = db_endpos
+        if db_start > start or db_end < end:
+            raise IntervalError("database times must span the interval times")
+
+    def __repr__(self):
+        s = repr(self.start) + ", " + repr(self.end)
+        s += ", " + repr(self.db_start) + ", " + repr(self.db_end)
+        s += ", " + repr(self.db_startpos) + ", " + repr(self.db_endpos)
+        return self.__class__.__name__ + "(" + s + ")"
+
+    cpdef subset(self, timestamp_t start, timestamp_t end):
+        """
+        Return a new DBInterval that is a subset of this one
+        """
+        if start < self.start or end > self.end:
+            raise IntervalError("not a subset")
+        return DBInterval(start, end,
+                          self.db_start, self.db_end,
+                          self.db_startpos, self.db_endpos)
+
+cdef class IntervalSet:
+    """
+    A non-intersecting set of intervals.
+    """
+
+    cdef public rbtree.RBTree tree
+
+    def __init__(self, source=None):
+        """
+        'source' is an Interval or IntervalSet to add.
+        """
+        self.tree = rbtree.RBTree()
+        if source is not None:
+            self += source
+
+    def __iter__(self):
+        for node in self.tree:
+            if node.obj:
+                yield node.obj
+
+    def __len__(self):
+        return sum(1 for x in self)
+
+    def __repr__(self):
+        descs = [ repr(x) for x in self ]
+        return self.__class__.__name__ + "([" + ", ".join(descs) + "])"
+
+    def __str__(self):
+        descs = [ str(x) for x in self ]
+        return  "[" + ", ".join(descs) + "]"
+
+    def __match__(self, other):
+        # This isn't particularly efficient, but it shouldn't get used in the
+        # general case.
+        """Test equality of two IntervalSets.
+
+        Treats adjacent Intervals as equivalent to one long interval,
+        so this function really tests whether the IntervalSets cover
+        the same spans of time."""
+        i = 0
+        j = 0
+        outside = True
+
+        def is_adjacent(a, b):
+            """Return True if two Intervals are adjacent (same end or start)"""
+            if a.end == b.start or b.end == a.start:
+                return True
+            else:
+                return False
+
+        this = list(self)
+        that = list(other)
+
+        try:
+            while True:
+                if (outside):
+                    # To match, we need to be finished both sets
+                    if (i >= len(this) and j >= len(that)):
+                        return True
+                    # Or the starts need to match
+                    if (this[i].start != that[j].start):
+                        return False
+                    outside = False
+                else:
+                    # We can move on if the two interval ends match
+                    if (this[i].end == that[j].end):
+                        i += 1
+                        j += 1
+                        outside = True
+                    else:
+                        # Whichever ends first needs to be adjacent to the next
+                        if (this[i].end < that[j].end):
+                            if (not is_adjacent(this[i],this[i+1])):
+                                return False
+                            i += 1
+                        else:
+                            if (not is_adjacent(that[j],that[j+1])):
+                                return False
+                            j += 1
+        except IndexError:
+            return False
+
+    # Use __richcmp__ instead of __eq__, __ne__ for Cython.
+    def __richcmp__(self, other, int op):
+        if op == 2: # ==
+            return self.__match__(other)
+        elif op == 3: # !=
+            return not self.__match__(other)
+        return False
+    #def __eq__(self, other):
+    #    return self.__match__(other)
+    #
+    #def __ne__(self, other):
+    #    return not self.__match__(other)
+
+    def __iadd__(self, object other not None):
+        """Inplace add -- modifies self
+
+        This throws an exception if the regions being added intersect."""
+        if isinstance(other, Interval):
+            if self.intersects(other):
+                raise IntervalError("Tried to add overlapping interval "
+                                    "to this set")
+            self.tree.insert(rbtree.RBNode(other.start, other.end, other))
+        else:
+            for x in other:
+                self.__iadd__(x)
+        return self
+
+    def iadd_nocheck(self, Interval other not None):
+        """Inplace add -- modifies self.
+        'Optimized' version that doesn't check for intersection and
+        only inserts the new interval into the tree."""
+        self.tree.insert(rbtree.RBNode(other.start, other.end, other))
+
+    def __isub__(self, Interval other not None):
+        """Inplace subtract -- modifies self
+
+        Removes an interval from the set.  Must exist exactly
+        as provided -- cannot remove a subset of an existing interval."""
+        i = self.tree.find(other.start, other.end)
+        if i is None:
+            raise IntervalError("interval " + str(other) + " not in tree")
+        self.tree.delete(i)
+        return self
+
+    def __add__(self, other not None):
+        """Add -- returns a new object"""
+        new = IntervalSet(self)
+        new += IntervalSet(other)
+        return new
+
+    def __and__(self, other not None):
+        """
+        Compute a new IntervalSet from the intersection of this
+        IntervalSet with one other interval.
+
+        Output intervals are built as subsets of the intervals in the
+        first argument (self).
+        """
+        out = IntervalSet()
+        for i in self.intersection(other):
+            out.tree.insert(rbtree.RBNode(i.start, i.end, i))
+        return out
+
+    def intersection(self, Interval interval not None, orig = False):
+        """
+        Compute a sequence of intervals that correspond to the
+        intersection between `self` and the provided interval.
+        Returns a generator that yields each of these intervals
+        in turn.
+
+        Output intervals are built as subsets of the intervals in the
+        first argument (self).
+
+        If orig = True, also return the original interval that was
+        (potentially) subsetted to make the one that is being
+        returned.
+        """
+        if orig:
+            for n in self.tree.intersect(interval.start, interval.end):
+                i = n.obj
+                subset = i.subset(max(i.start, interval.start),
+                                  min(i.end, interval.end))
+                yield (subset, i)
+        else:
+            for n in self.tree.intersect(interval.start, interval.end):
+                i = n.obj
+                subset = i.subset(max(i.start, interval.start),
+                                  min(i.end, interval.end))
+                yield subset
+
+    cpdef intersects(self, Interval other):
+        """Return True if this IntervalSet intersects another interval"""
+        for n in self.tree.intersect(other.start, other.end):
+            if n.obj.intersects(other):
+                return True
+        return False
+
+    def find_end(self, timestamp_t t):
+        """
+        Return an Interval from this tree that ends at time t, or
+        None if it doesn't exist.
+        """
+        n = self.tree.find_left_end(t)
+        if n and n.obj.end == t:
+            return n.obj
+        return None

nilmdb/server/interval.pyxdep

@@ -0,0 +1 @@
+rbtree.pxd

nilmdb/server/nilmdb.py

@@ -0,0 +1,717 @@
+# -*- coding: utf-8 -*-
+
+"""NilmDB
+
+Object that represents a NILM database file.
+
+Manages both the SQL database and the table storage backend.
+"""
+
+import os
+import errno
+import sqlite3
+
+import nilmdb.utils
+from nilmdb.utils.printf import printf
+from nilmdb.utils.time import timestamp_to_bytes
+
+from nilmdb.utils.interval import IntervalError
+from nilmdb.server.interval import Interval, DBInterval, IntervalSet
+
+from nilmdb.server import bulkdata
+from nilmdb.server.errors import NilmDBError, StreamError, OverlapError
+
+# Note about performance and transactions:
+#
+# Committing a transaction in the default sync mode (PRAGMA synchronous=FULL)
+# takes about 125msec.  sqlite3 will commit transactions at 3 times:
+# 1: explicit con.commit()
+# 2: between a series of DML commands and non-DML commands, e.g.
+#    after a series of INSERT, SELECT, but before a CREATE TABLE or PRAGMA.
+# 3: at the end of an explicit transaction, e.g. "with self.con as con:"
+#
+# To speed things up, we can set 'PRAGMA synchronous=OFF'.  Or, it
+# seems that 'PRAGMA synchronous=NORMAL' and 'PRAGMA journal_mode=WAL'
+# give an equivalent speedup more safely.  That is what is used here.
+_sql_schema_updates = {
+    0: {"next": 1, "sql": """
+    -- All streams
+    CREATE TABLE streams(
+        id INTEGER PRIMARY KEY,		-- stream ID
+        path TEXT UNIQUE NOT NULL,	-- path, e.g. '/newton/prep'
+        layout TEXT NOT NULL		-- layout name, e.g. float32_8
+    );
+
+    -- Individual timestamped ranges in those streams.
+    -- For a given start_time and end_time, this tells us that the
+    -- data is stored between start_pos and end_pos.
+    -- Times are stored as μs since Unix epoch
+    -- Positions are opaque: PyTables rows, file offsets, etc.
+    --
+    -- Note: end_pos points to the row _after_ end_time, so end_pos-1
+    -- is the last valid row.
+    CREATE TABLE ranges(
+        stream_id INTEGER NOT NULL,
+        start_time INTEGER NOT NULL,
+        end_time INTEGER NOT NULL,
+        start_pos INTEGER NOT NULL,
+        end_pos INTEGER NOT NULL
+    );
+    CREATE INDEX _ranges_index ON ranges (stream_id, start_time, end_time);
+    """},
+
+    1: {"next": 3, "sql": """
+    -- Generic dictionary-type metadata that can be associated with a stream
+    CREATE TABLE metadata(
+        stream_id INTEGER NOT NULL,
+        key TEXT NOT NULL,
+        value TEXT
+    );
+    """},
+
+    2: {"error": "old format with floating-point timestamps requires "
+        "nilmdb 1.3.1 or older"},
+
+    3: {"next": None},
+}
+
+
+@nilmdb.utils.must_close()
+class NilmDB():
+    verbose = 0
+
+    def __init__(self, basepath,
+                 max_results=None,
+                 max_removals=None,
+                 max_int_removals=None,
+                 bulkdata_args=None):
+        """Initialize NilmDB at the given basepath.
+        Other arguments are for debugging / testing:
+
+        'max_results' is the max rows to send in a single
+        stream_intervals or stream_extract response.
+
+        'max_removals' is the max rows to delete at once
+        in stream_remove.
+
+        'max_int_removals' is the max intervals to delete
+        at once in stream_remove.
+
+        'bulkdata_args' is kwargs for the bulkdata module.
+        """
+        if bulkdata_args is None:
+            bulkdata_args = {}
+
+        # set up path
+        self.basepath = os.path.abspath(basepath)
+
+        # Create the database path if it doesn't exist
+        try:
+            os.makedirs(self.basepath)
+        except OSError as e:
+            if e.errno != errno.EEXIST:
+                raise IOError("can't create tree " + self.basepath)
+
+        # Our data goes inside it
+        self.data = bulkdata.BulkData(self.basepath, **bulkdata_args)
+
+        # SQLite database too
+        sqlfilename = os.path.join(self.basepath, "data.sql")
+        self.con = sqlite3.connect(sqlfilename, check_same_thread=True)
+        try:
+            self._sql_schema_update()
+        except Exception:
+            self.data.close()
+            raise
+
+        # See big comment at top about the performance implications of this
+        self.con.execute("PRAGMA synchronous=NORMAL")
+        self.con.execute("PRAGMA journal_mode=WAL")
+
+        # Approximate largest number of elements that we want to send
+        # in a single reply (for stream_intervals, stream_extract).
+        self.max_results = max_results or 16384
+
+        # Remove up to this many rows per call to stream_remove.
+        self.max_removals = max_removals or 1048576
+
+        # Remove up to this many intervals per call to stream_remove.
+        self.max_int_removals = max_int_removals or 4096
+
+    def get_basepath(self):
+        return self.basepath
+
+    def close(self):
+        if self.con:
+            self.con.commit()
+            self.con.close()
+            self.con = None
+        self.data.close()
+
+    def _sql_schema_update(self):
+        cur = self.con.cursor()
+        version = cur.execute("PRAGMA user_version").fetchone()[0]
+        oldversion = version
+
+        while True:
+            if version not in _sql_schema_updates:
+                raise Exception(self.basepath + ": unknown database version "
+                                + str(version))
+            update = _sql_schema_updates[version]
+            if "error" in update:
+                raise Exception(self.basepath + ": can't use database version "
+                                + str(version) + ": " + update["error"])
+            if update["next"] is None:
+                break
+            cur.executescript(update["sql"])
+            version = update["next"]
+            if self.verbose:
+                printf("Database schema updated to %d\n", version)
+
+        if version != oldversion:
+            with self.con:
+                cur.execute("PRAGMA user_version = {v:d}".format(v=version))
+
+    def _check_user_times(self, start, end):
+        if start is None:
+            start = nilmdb.utils.time.min_timestamp
+        if end is None:
+            end = nilmdb.utils.time.max_timestamp
+        if start >= end:
+            raise NilmDBError("start must precede end")
+        return (start, end)
+
+    @nilmdb.utils.lru_cache(size=64)
+    def _get_intervals(self, stream_id):
+        """
+        Return a mutable IntervalSet corresponding to the given stream ID.
+        """
+        iset = IntervalSet()
+        result = self.con.execute("SELECT start_time, end_time, "
+                                  "start_pos, end_pos "
+                                  "FROM ranges "
+                                  "WHERE stream_id=?", (stream_id,))
+        try:
+            for (start_time, end_time, start_pos, end_pos) in result:
+                iset += DBInterval(start_time, end_time,
+                                   start_time, end_time,
+                                   start_pos, end_pos)
+        except IntervalError:
+            raise NilmDBError("unexpected overlap in ranges table!")
+
+        return iset
+
+    def _sql_interval_insert(self, id, start, end, start_pos, end_pos):
+        """Helper that adds interval to the SQL database only"""
+        self.con.execute("INSERT INTO ranges "
+                         "(stream_id,start_time,end_time,start_pos,end_pos) "
+                         "VALUES (?,?,?,?,?)",
+                         (id, start, end, start_pos, end_pos))
+
+    def _sql_interval_delete(self, id, start, end, start_pos, end_pos):
+        """Helper that removes interval from the SQL database only"""
+        self.con.execute("DELETE FROM ranges WHERE "
+                         "stream_id=? AND start_time=? AND "
+                         "end_time=? AND start_pos=? AND end_pos=?",
+                         (id, start, end, start_pos, end_pos))
+
+    def _add_interval(self, stream_id, interval, start_pos, end_pos):
+        """
+        Add interval to the internal interval cache, and to the database.
+        Note: arguments must be ints (not numpy.int64, etc)
+        """
+        # Load this stream's intervals
+        iset = self._get_intervals(stream_id)
+
+        # Check for adjacency.  If there's a stream in the database
+        # that ends exactly when this one starts, and the database
+        # rows match up, we can make one interval that covers the
+        # time range [adjacent.start -> interval.end)
+        # and database rows [ adjacent.start_pos -> end_pos ].
+        # Only do this if the resulting interval isn't too large.
+        max_merged_rows = 8000 * 60 * 60 * 1.05  # 1.05 hours at 8 KHz
+        adjacent = iset.find_end(interval.start)
+        if (adjacent is not None and
+                start_pos == adjacent.db_endpos and
+                (end_pos - adjacent.db_startpos) < max_merged_rows):
+            # First delete the old one, both from our iset and the
+            # database
+            iset -= adjacent
+            self._sql_interval_delete(stream_id,
+                                      adjacent.db_start, adjacent.db_end,
+                                      adjacent.db_startpos, adjacent.db_endpos)
+
+            # Now update our interval so the fallthrough add is
+            # correct.
+            interval.start = adjacent.start
+            start_pos = adjacent.db_startpos
+
+        # Add the new interval to the iset
+        iset.iadd_nocheck(DBInterval(interval.start, interval.end,
+                                     interval.start, interval.end,
+                                     start_pos, end_pos))
+
+        # Insert into the database
+        self._sql_interval_insert(stream_id, interval.start, interval.end,
+                                  int(start_pos), int(end_pos))
+
+        self.con.commit()
+
+    def _remove_interval(self, stream_id, original, remove):
+        """
+        Remove an interval from the internal cache and the database.
+
+        stream_id: id of stream
+         original: original DBInterval; must be already present in DB
+        to_remove: DBInterval to remove; must be subset of 'original'
+        """
+        # Load this stream's intervals
+        iset = self._get_intervals(stream_id)
+
+        # Remove existing interval from the cached set and the database
+        iset -= original
+        self._sql_interval_delete(stream_id,
+                                  original.db_start, original.db_end,
+                                  original.db_startpos, original.db_endpos)
+
+        # Add back the intervals that would be left over if the
+        # requested interval is removed.  There may be two of them, if
+        # the removed piece was in the middle.
+        def add(iset, start, end, start_pos, end_pos):
+            iset += DBInterval(start, end, start, end, start_pos, end_pos)
+            self._sql_interval_insert(stream_id, start, end,
+                                      start_pos, end_pos)
+
+        if original.start != remove.start:
+            # Interval before the removed region
+            add(iset, original.start, remove.start,
+                original.db_startpos, remove.db_startpos)
+
+        if original.end != remove.end:
+            # Interval after the removed region
+            add(iset, remove.end, original.end,
+                remove.db_endpos, original.db_endpos)
+
+        # Commit SQL changes
+        self.con.commit()
+
+        return
+
+    def stream_list(self, path=None, layout=None, extended=False):
+        """Return list of lists of all streams in the database.
+
+        If path is specified, include only streams with a path that
+        matches the given string.
+
+        If layout is specified, include only streams with a layout
+        that matches the given string.
+
+        If extended=False, returns a list of lists containing
+        the path and layout: [ path, layout ]
+
+        If extended=True, returns a list of lists containing
+        more information:
+           path
+           layout
+           interval_min (earliest interval start)
+           interval_max (latest interval end)
+           rows         (total number of rows of data)
+           time         (total time covered by this stream, in timestamp units)
+        """
+        params = ()
+        query = "SELECT streams.path, streams.layout"
+        if extended:
+            query += ", min(ranges.start_time), max(ranges.end_time) "
+            query += ", coalesce(sum(ranges.end_pos - ranges.start_pos), 0) "
+            query += ", coalesce(sum(ranges.end_time - ranges.start_time), 0) "
+        query += " FROM streams"
+        if extended:
+            query += " LEFT JOIN ranges ON streams.id = ranges.stream_id"
+        query += " WHERE 1=1"
+        if layout is not None:
+            query += " AND streams.layout=?"
+            params += (layout,)
+        if path is not None:
+            query += " AND streams.path=?"
+            params += (path,)
+        query += " GROUP BY streams.id ORDER BY streams.path"
+        result = self.con.execute(query, params).fetchall()
+        return [list(x) for x in result]
+
+    def stream_intervals(self, path, start=None, end=None, diffpath=None):
+        """
+        List all intervals in 'path' between 'start' and 'end'.  If
+        'diffpath' is not none, list instead the set-difference
+        between the intervals in the two streams; i.e. all interval
+        ranges that are present in 'path' but not 'diffpath'.
+
+        Returns (intervals, restart) tuple.
+
+        'intervals' is a list of [start,end] timestamps of all intervals
+        that exist for path, between start and end.
+
+        'restart', if not None, means that there were too many results
+        to return in a single request.  The data is complete from the
+        starting timestamp to the point at which it was truncated, and
+        a new request with a start time of 'restart' will fetch the
+        next block of data.
+        """
+        stream_id = self._stream_id(path)
+        intervals = self._get_intervals(stream_id)
+        if diffpath:
+            diffstream_id = self._stream_id(diffpath)
+            diffintervals = self._get_intervals(diffstream_id)
+        (start, end) = self._check_user_times(start, end)
+        requested = Interval(start, end)
+        result = []
+        if diffpath:
+            getter = nilmdb.utils.interval.set_difference(
+                intervals.intersection(requested),
+                diffintervals.intersection(requested))
+        else:
+            getter = intervals.intersection(requested)
+        for n, i in enumerate(getter):
+            if n >= self.max_results:
+                restart = i.start
+                break
+            result.append([i.start, i.end])
+        else:
+            restart = None
+        return (result, restart)
+
+    def stream_create(self, path, layout_name):
+        """Create a new table in the database.
+
+        path: path to the data (e.g. '/newton/prep').
+        Paths must contain at least two elements, e.g.:
+           /newton/prep
+           /newton/raw
+           /newton/upstairs/prep
+           /newton/upstairs/raw
+
+        layout_name: string for nilmdb.layout.get_named(), e.g. 'float32_8'
+        """
+        # Create the bulk storage.  Raises ValueError on error, which we
+        # pass along.
+        self.data.create(path, layout_name)
+
+        # Insert into SQL database once the bulk storage is happy
+        with self.con as con:
+            con.execute("INSERT INTO streams (path, layout) VALUES (?,?)",
+                        (path, layout_name))
+
+    def _stream_id(self, path):
+        """Return unique stream ID"""
+        result = self.con.execute("SELECT id FROM streams WHERE path=?",
+                                  (path,)).fetchone()
+        if result is None:
+            raise StreamError("No stream at path " + path)
+        return result[0]
+
+    def stream_set_metadata(self, path, data):
+        """Set stream metadata from a dictionary, e.g.
+           { description: 'Downstairs lighting',
+             v_scaling: 123.45 }
+           This replaces all existing metadata.
+           """
+        stream_id = self._stream_id(path)
+        with self.con as con:
+            con.execute("DELETE FROM metadata WHERE stream_id=?", (stream_id,))
+            for key in data:
+                if data[key] != '':
+                    con.execute("INSERT INTO metadata VALUES (?, ?, ?)",
+                                (stream_id, key, data[key]))
+
+    def stream_get_metadata(self, path):
+        """Return stream metadata as a dictionary."""
+        stream_id = self._stream_id(path)
+        result = self.con.execute("SELECT metadata.key, metadata.value "
+                                  "FROM metadata "
+                                  "WHERE metadata.stream_id=?", (stream_id,))
+        data = {}
+        for (key, value) in result:
+            data[key] = value
+        return data
+
+    def stream_update_metadata(self, path, newdata):
+        """Update stream metadata from a dictionary"""
+        data = self.stream_get_metadata(path)
+        data.update(newdata)
+        self.stream_set_metadata(path, data)
+
+    def stream_rename(self, oldpath, newpath):
+        """Rename a stream."""
+        stream_id = self._stream_id(oldpath)
+
+        # Rename the data
+        self.data.rename(oldpath, newpath)
+
+        # Rename the stream in the database
+        with self.con as con:
+            con.execute("UPDATE streams SET path=? WHERE id=?",
+                        (newpath, stream_id))
+
+    def stream_destroy(self, path):
+        """Fully remove a table from the database.  Fails if there are
+        any intervals data present; remove them first.  Metadata is
+        also removed."""
+        stream_id = self._stream_id(path)
+
+        # Verify that no intervals are present, and clear the cache
+        iset = self._get_intervals(stream_id)
+        if iset:
+            raise NilmDBError("all intervals must be removed before "
+                              "destroying a stream")
+        self._get_intervals.cache_remove(self, stream_id)
+
+        # Delete the bulkdata storage
+        self.data.destroy(path)
+
+        # Delete metadata, stream, intervals (should be none)
+        with self.con as con:
+            con.execute("DELETE FROM metadata WHERE stream_id=?", (stream_id,))
+            con.execute("DELETE FROM ranges WHERE stream_id=?", (stream_id,))
+            con.execute("DELETE FROM streams WHERE id=?", (stream_id,))
+
+    def stream_insert(self, path, start, end, data, binary=False):
+        """Insert new data into the database.
+           path: Path at which to add the data
+           start: Starting timestamp
+           end: Ending timestamp
+           data: Textual data, formatted according to the layout of path
+
+           'binary', if True, means that 'data' is raw binary:
+           little-endian, matching the current table's layout,
+           including the int64 timestamp.
+           """
+        # First check for basic overlap using timestamp info given.
+        stream_id = self._stream_id(path)
+        iset = self._get_intervals(stream_id)
+        interval = Interval(start, end)
+        if iset.intersects(interval):
+            raise OverlapError("new data overlaps existing data at range: "
+                               + str(iset & interval))
+
+        # Tenatively append the data.  This will raise a ValueError if
+        # there are any parse errors.
+        table = self.data.getnode(path)
+        row_start = table.nrows
+        table.append_data(data, start, end, binary)
+        row_end = table.nrows
+
+        # Insert the record into the sql database.
+        self._add_interval(stream_id, interval, row_start, row_end)
+
+        # And that's all
+        return
+
+    def _bisect_left(self, a, x, lo, hi):
+        # Like bisect.bisect_left, but doesn't choke on large indices on
+        # 32-bit systems, like bisect's fast C implementation does.
+        while lo < hi:
+            mid = (lo + hi) // 2
+            if a[mid] < x:
+                lo = mid + 1
+            else:
+                hi = mid
+        return lo
+
+    def _find_start(self, table, dbinterval):
+        """
+        Given a DBInterval, find the row in the database that
+        corresponds to the start time.  Return the first database
+        position with a timestamp (first element) greater than or
+        equal to 'start'.
+        """
+        # Optimization for the common case where an interval wasn't truncated
+        if dbinterval.start == dbinterval.db_start:
+            return dbinterval.db_startpos
+        return self._bisect_left(table,
+                                 dbinterval.start,
+                                 dbinterval.db_startpos,
+                                 dbinterval.db_endpos)
+
+    def _find_end(self, table, dbinterval):
+        """
+        Given a DBInterval, find the row in the database that follows
+        the end time.  Return the first database position after the
+        row with timestamp (first element) greater than or equal
+        to 'end'.
+        """
+        # Optimization for the common case where an interval wasn't truncated
+        if dbinterval.end == dbinterval.db_end:
+            return dbinterval.db_endpos
+        # Note that we still use bisect_left here, because we don't
+        # want to include the given timestamp in the results.  This is
+        # so a queries like 1:00 -> 2:00 and 2:00 -> 3:00 return
+        # non-overlapping data.
+        return self._bisect_left(table,
+                                 dbinterval.end,
+                                 dbinterval.db_startpos,
+                                 dbinterval.db_endpos)
+
+    def stream_extract(self, path, start=None, end=None,
+                       count=False, markup=False, binary=False):
+        """
+        Returns (data, restart) tuple.
+
+        'data' is ASCII-formatted data from the database, formatted
+        according to the layout of the stream.
+
+        'restart', if not None, means that there were too many results to
+        return in a single request.  The data is complete from the
+        starting timestamp to the point at which it was truncated,
+        and a new request with a start time of 'restart' will fetch
+        the next block of data.
+
+        'count', if true, means to not return raw data, but just the count
+        of rows that would have been returned.  This is much faster
+        than actually fetching the data.  It is not limited by
+        max_results.
+
+        'markup', if true, indicates that returned data should be
+        marked with a comment denoting when a particular interval
+        starts, and another comment when an interval ends.
+
+        'binary', if true, means to return raw binary rather than
+        ASCII-formatted data.
+        """
+        stream_id = self._stream_id(path)
+        table = self.data.getnode(path)
+        intervals = self._get_intervals(stream_id)
+        (start, end) = self._check_user_times(start, end)
+        requested = Interval(start, end)
+        result = []
+        matched = 0
+        remaining = self.max_results
+        restart = None
+        if binary and (markup or count):
+            raise NilmDBError("binary mode can't be used with markup or count")
+        for interval in intervals.intersection(requested):
+            # Reading single rows from the table is too slow, so
+            # we use two bisections to find both the starting and
+            # ending row for this particular interval, then
+            # read the entire range as one slice.
+            row_start = self._find_start(table, interval)
+            row_end = self._find_end(table, interval)
+
+            if count:
+                matched += row_end - row_start
+                continue
+
+            # Shorten it if we'll hit the maximum number of results
+            row_max = row_start + remaining
+            if row_max < row_end:
+                row_end = row_max
+                restart = table[row_max]
+
+            # Add markup
+            if markup:
+                result.append(b"# interval-start " +
+                              timestamp_to_bytes(interval.start) + b"\n")
+
+            # Gather these results up
+            result.append(table.get_data(row_start, row_end, binary))
+
+            # Count them
+            remaining -= row_end - row_start
+
+            # Add markup, and exit if restart is set.
+            if restart is not None:
+                if markup:
+                    result.append(b"# interval-end " +
+                                  timestamp_to_bytes(restart) + b"\n")
+                break
+            if markup:
+                result.append(b"# interval-end " +
+                              timestamp_to_bytes(interval.end) + b"\n")
+
+        if count:
+            return matched
+        full_result = b"".join(result)
+        return (full_result, restart)
+
+    def stream_remove(self, path, start=None, end=None):
+        """
+        Remove data from the specified time interval within a stream.
+
+        Removes data in the interval [start, end), and intervals are
+        truncated or split appropriately.
+
+        Returns a (removed, restart) tuple.
+
+        'removed' is the number of data points that were removed.
+
+        'restart', if not None, means there were too many rows to
+        remove in a single request.  This function should be called
+        again with a start time of 'restart' to complete the removal.
+        """
+        stream_id = self._stream_id(path)
+        table = self.data.getnode(path)
+        intervals = self._get_intervals(stream_id)
+        (start, end) = self._check_user_times(start, end)
+        to_remove = Interval(start, end)
+        removed = 0
+        remaining = self.max_removals
+        int_remaining = self.max_int_removals
+        restart = None
+
+        # Can't remove intervals from within the iterator, so we need to
+        # remember what's currently in the intersection now.
+        all_candidates = list(intervals.intersection(to_remove, orig=True))
+
+        remove_start = None
+        remove_end = None
+
+        for (dbint, orig) in all_candidates:
+            # Stop if we've hit the max number of interval removals
+            if int_remaining <= 0:
+                restart = dbint.start
+                break
+
+            # Find row start and end
+            row_start = self._find_start(table, dbint)
+            row_end = self._find_end(table, dbint)
+
+            # Shorten it if we'll hit the maximum number of removals
+            row_max = row_start + remaining
+            if row_max < row_end:
+                row_end = row_max
+                dbint.end = table[row_max]
+                restart = dbint.end
+
+            # Adjust the DBInterval to match the newly found ends
+            dbint.db_start = dbint.start
+            dbint.db_end = dbint.end
+            dbint.db_startpos = row_start
+            dbint.db_endpos = row_end
+
+            # Remove interval from the database
+            self._remove_interval(stream_id, orig, dbint)
+
+            # Remove data from the underlying table storage,
+            # coalescing adjacent removals to reduce the number of calls
+            # to table.remove.
+            if remove_end == row_start:
+                # Extend our coalesced region
+                remove_end = row_end
+            else:
+                # Perform previous removal, then save this one
+                if remove_end is not None:
+                    table.remove(remove_start, remove_end)
+                remove_start = row_start
+                remove_end = row_end
+
+            # Count how many were removed
+            removed += row_end - row_start
+            remaining -= row_end - row_start
+            int_remaining -= 1
+
+            if restart is not None:
+                break
+
+        # Perform any final coalesced removal
+        if remove_end is not None:
+            table.remove(remove_start, remove_end)
+
+        return (removed, restart)

nilmdb/server/rbtree.pxd

@@ -0,0 +1,25 @@
+# cython: language_level=2
+
+cdef class RBNode:
+    cdef public object obj
+    cdef public double start, end
+    cdef public int red
+    cdef public RBNode left, right, parent
+
+cdef class RBTree:
+    cdef public RBNode nil, root
+
+    cpdef getroot(RBTree self)
+    cdef void __rotate_left(RBTree self, RBNode x)
+    cdef void __rotate_right(RBTree self, RBNode y)
+    cdef RBNode __successor(RBTree self, RBNode x)
+    cpdef RBNode successor(RBTree self, RBNode x)
+    cdef RBNode __predecessor(RBTree self, RBNode x)
+    cpdef RBNode predecessor(RBTree self, RBNode x)
+    cpdef insert(RBTree self, RBNode z)
+    cdef void __insert_fixup(RBTree self, RBNode x)
+    cpdef delete(RBTree self, RBNode z)
+    cdef inline void __delete_fixup(RBTree self, RBNode x)
+    cpdef RBNode find(RBTree self, double start, double end)
+    cpdef RBNode find_left_end(RBTree self, double t)
+    cpdef RBNode find_right_start(RBTree self, double t)

nilmdb/server/rbtree.pyx

@@ -0,0 +1,378 @@
+# cython: profile=False
+# cython: cdivision=True
+# cython: language_level=2
+
+"""
+Jim Paris <jim@jtan.com>
+
+Red-black tree, where keys are stored as start/end timestamps.
+This is a basic interval tree that holds half-open intervals:
+  [start, end)
+Intervals must not overlap.  Fixing that would involve making this
+into an augmented interval tree as described in CLRS 14.3.
+
+Code that assumes non-overlapping intervals is marked with the
+string 'non-overlapping'.
+"""
+
+import sys
+cimport rbtree
+
+cdef class RBNode:
+    """One node of the Red/Black tree, containing a key (start, end)
+    and value (obj)"""
+    def __init__(self, double start, double end, object obj = None):
+        self.obj = obj
+        self.start = start
+        self.end = end
+        self.red = False
+        self.left = None
+        self.right = None
+
+    def __str__(self):
+        if self.red:
+            color = "R"
+        else:
+            color = "B"
+        if self.start == sys.float_info.min:
+            return "[node nil]"
+        return ("[node ("
+                + str(self.obj) + ") "
+                + str(self.start) + " -> " + str(self.end) + " "
+                + color + "]")
+
+cdef class RBTree:
+    """Red/Black tree"""
+
+    # Init
+    def __init__(self):
+        self.nil = RBNode(start = sys.float_info.min,
+                          end = sys.float_info.min)
+        self.nil.left = self.nil
+        self.nil.right = self.nil
+        self.nil.parent = self.nil
+
+        self.root = RBNode(start = sys.float_info.max,
+                           end = sys.float_info.max)
+        self.root.left = self.nil
+        self.root.right = self.nil
+        self.root.parent = self.nil
+
+    # We have a dummy root node to simplify operations, so from an
+    # external point of view, its left child is the real root.
+    cpdef getroot(self):
+        return self.root.left
+
+    # Rotations and basic operations
+    cdef void __rotate_left(self, RBNode x):
+        """Rotate left:
+        #   x           y
+        #  / \   -->   / \
+        # z   y       x   w
+        #    / \     / \
+        #   v   w   z   v
+        """
+        cdef RBNode y = x.right
+        x.right = y.left
+        if y.left is not self.nil:
+            y.left.parent = x
+        y.parent = x.parent
+        if x is x.parent.left:
+            x.parent.left = y
+        else:
+            x.parent.right = y
+        y.left = x
+        x.parent = y
+
+    cdef void __rotate_right(self, RBNode y):
+        """Rotate right:
+        #     y           x
+        #    / \   -->   / \
+        #   x   w       z   y
+        #  / \             / \
+        # z   v           v   w
+        """
+        cdef RBNode x = y.left
+        y.left = x.right
+        if x.right is not self.nil:
+            x.right.parent = y
+        x.parent = y.parent
+        if y is y.parent.left:
+            y.parent.left = x
+        else:
+            y.parent.right = x
+        x.right = y
+        y.parent = x
+
+    cdef RBNode __successor(self, RBNode x):
+        """Returns the successor of RBNode x"""
+        cdef RBNode y = x.right
+        if y is not self.nil:
+            while y.left is not self.nil:
+                y = y.left
+        else:
+            y = x.parent
+            while x is y.right:
+                x = y
+                y = y.parent
+            if y is self.root:
+                return self.nil
+        return y
+    cpdef RBNode successor(self, RBNode x):
+        """Returns the successor of RBNode x, or None"""
+        cdef RBNode y = self.__successor(x)
+        return y if y is not self.nil else None
+
+    cdef RBNode __predecessor(self, RBNode x):
+        """Returns the predecessor of RBNode x"""
+        cdef RBNode y = x.left
+        if y is not self.nil:
+            while y.right is not self.nil:
+                y = y.right
+        else:
+            y = x.parent
+            while x is y.left:
+                if y is self.root:
+                    y = self.nil
+                    break
+                x = y
+                y = y.parent
+        return y
+    cpdef RBNode predecessor(self, RBNode x):
+        """Returns the predecessor of RBNode x, or None"""
+        cdef RBNode y = self.__predecessor(x)
+        return y if y is not self.nil else None
+
+    # Insertion
+    cpdef insert(self, RBNode z):
+        """Insert RBNode z into RBTree and rebalance as necessary"""
+        z.left = self.nil
+        z.right = self.nil
+        cdef RBNode y = self.root
+        cdef RBNode x = self.root.left
+        while x is not self.nil:
+            y = x
+            if (x.start > z.start or (x.start == z.start and x.end > z.end)):
+                x = x.left
+            else:
+                x = x.right
+        z.parent = y
+        if (y is self.root or
+            (y.start > z.start or (y.start == z.start and y.end > z.end))):
+            y.left = z
+        else:
+            y.right = z
+        # relabel/rebalance
+        self.__insert_fixup(z)
+
+    cdef void __insert_fixup(self, RBNode x):
+        """Rebalance/fix RBTree after a simple insertion of RBNode x"""
+        x.red = True
+        while x.parent.red:
+            if x.parent is x.parent.parent.left:
+                y = x.parent.parent.right
+                if y.red:
+                    x.parent.red = False
+                    y.red = False
+                    x.parent.parent.red = True
+                    x = x.parent.parent
+                else:
+                    if x is x.parent.right:
+                        x = x.parent
+                        self.__rotate_left(x)
+                    x.parent.red = False
+                    x.parent.parent.red = True
+                    self.__rotate_right(x.parent.parent)
+            else: # same as above, left/right switched
+                y = x.parent.parent.left
+                if y.red:
+                    x.parent.red = False
+                    y.red = False
+                    x.parent.parent.red = True
+                    x = x.parent.parent
+                else:
+                    if x is x.parent.left:
+                        x = x.parent
+                        self.__rotate_right(x)
+                    x.parent.red = False
+                    x.parent.parent.red = True
+                    self.__rotate_left(x.parent.parent)
+        self.root.left.red = False
+
+    # Deletion
+    cpdef delete(self, RBNode z):
+        if z.left is None or z.right is None:
+            raise AttributeError("you can only delete a node object "
+                                 + "from the tree; use find() to get one")
+        cdef RBNode x, y
+        if z.left is self.nil or z.right is self.nil:
+            y = z
+        else:
+            y = self.__successor(z)
+        if y.left is self.nil:
+            x = y.right
+        else:
+            x = y.left
+        x.parent = y.parent
+        if x.parent is self.root:
+            self.root.left = x
+        else:
+            if y is y.parent.left:
+                y.parent.left = x
+            else:
+                y.parent.right = x
+        if y is not z:
+            # y is the node to splice out, x is its child
+            y.left = z.left
+            y.right = z.right
+            y.parent = z.parent
+            z.left.parent = y
+            z.right.parent = y
+            if z is z.parent.left:
+                z.parent.left = y
+            else:
+                z.parent.right = y
+            if not y.red:
+                y.red = z.red
+                self.__delete_fixup(x)
+            else:
+                y.red = z.red
+        else:
+            if not y.red:
+                self.__delete_fixup(x)
+
+    cdef void __delete_fixup(self, RBNode x):
+        """Rebalance/fix RBTree after a deletion.  RBNode x is the
+        child of the spliced out node."""
+        cdef RBNode rootLeft = self.root.left
+        while not x.red and x is not rootLeft:
+            if x is x.parent.left:
+                w = x.parent.right
+                if w.red:
+                    w.red = False
+                    x.parent.red = True
+                    self.__rotate_left(x.parent)
+                    w = x.parent.right
+                if not w.right.red and not w.left.red:
+                    w.red = True
+                    x = x.parent
+                else:
+                    if not w.right.red:
+                        w.left.red = False
+                        w.red = True
+                        self.__rotate_right(w)
+                        w = x.parent.right
+                    w.red = x.parent.red
+                    x.parent.red = False
+                    w.right.red = False
+                    self.__rotate_left(x.parent)
+                    x = rootLeft # exit loop
+            else: # same as above, left/right switched
+                w = x.parent.left
+                if w.red:
+                    w.red = False
+                    x.parent.red = True
+                    self.__rotate_right(x.parent)
+                    w = x.parent.left
+                if not w.left.red and not w.right.red:
+                    w.red = True
+                    x = x.parent
+                else:
+                    if not w.left.red:
+                        w.right.red = False
+                        w.red = True
+                        self.__rotate_left(w)
+                        w = x.parent.left
+                    w.red = x.parent.red
+                    x.parent.red = False
+                    w.left.red = False
+                    self.__rotate_right(x.parent)
+                    x = rootLeft # exit loop
+        x.red = False
+
+    # Walking, searching
+    def __iter__(self):
+        return self.inorder()
+
+    def inorder(self, RBNode x = None):
+        """Generator that performs an inorder walk for the tree
+        rooted at RBNode x"""
+        if x is None:
+            x = self.getroot()
+        while x.left is not self.nil:
+            x = x.left
+        while x is not self.nil:
+            yield x
+            x = self.__successor(x)
+
+    cpdef RBNode find(self, double start, double end):
+        """Return the node with exactly the given start and end."""
+        cdef RBNode x = self.getroot()
+        while x is not self.nil:
+            if start < x.start:
+                x = x.left
+            elif start == x.start:
+                if end == x.end:
+                    break # found it
+                elif end < x.end:
+                    x = x.left
+                else:
+                    x = x.right
+            else:
+                x = x.right
+        return x if x is not self.nil else None
+
+    cpdef RBNode find_left_end(self, double t):
+        """Find the leftmode node with end >= t.  With non-overlapping
+        intervals, this is the first node that might overlap time t.
+
+        Note that this relies on non-overlapping intervals, since
+        it assumes that we can use the endpoints to traverse the
+        tree even though it was created using the start points."""
+        cdef RBNode x = self.getroot()
+        while x is not self.nil:
+            if t < x.end:
+                if x.left is self.nil:
+                    break
+                x = x.left
+            elif t == x.end:
+                break
+            else:
+                if x.right is self.nil:
+                    x = self.__successor(x)
+                    break
+                x = x.right
+        return x if x is not self.nil else None
+
+    cpdef RBNode find_right_start(self, double t):
+        """Find the rightmode node with start <= t.  With non-overlapping
+        intervals, this is the last node that might overlap time t."""
+        cdef RBNode x = self.getroot()
+        while x is not self.nil:
+            if t < x.start:
+                if x.left is self.nil:
+                    x = self.__predecessor(x)
+                    break
+                x = x.left
+            elif t == x.start:
+                break
+            else:
+                if x.right is self.nil:
+                    break
+                x = x.right
+        return x if x is not self.nil else None
+
+    # Intersections
+    def intersect(self, double start, double end):
+        """Generator that returns nodes that overlap the given
+        (start,end) range.  Assumes non-overlapping intervals."""
+        # Start with the leftmode node that ends after start
+        cdef RBNode n = self.find_left_end(start)
+        while n is not None:
+            if n.start >= end:
+                # this node starts after the requested end; we're done
+                break
+            if start < n.end:
+                # this node overlaps our requested area
+                yield n
+            n = self.successor(n)

nilmdb/server/rbtree.pyxdep

@@ -0,0 +1 @@
+rbtree.pxd

nilmdb/server/rocket.c

@@ -0,0 +1,806 @@
+#include <Python.h>
+#include <structmember.h>
+#include <endian.h>
+
+#include <ctype.h>
+#include <stdint.h>
+
+#define __STDC_FORMAT_MACROS
+#include <inttypes.h>
+
+/* Values missing from stdint.h */
+#define UINT8_MIN 0
+#define UINT16_MIN 0
+#define UINT32_MIN 0
+#define UINT64_MIN 0
+
+/* Marker values (if min == max, skip range check) */
+#define FLOAT32_MIN 0
+#define FLOAT32_MAX 0
+#define FLOAT64_MIN 0
+#define FLOAT64_MAX 0
+
+typedef int64_t timestamp_t;
+
+/* Somewhat arbitrary, just so we can use fixed sizes for strings
+   etc. */
+static const int MAX_LAYOUT_COUNT = 1024;
+
+/* Error object and constants */
+static PyObject *ParseError;
+typedef enum {
+	ERR_OTHER,
+	ERR_NON_MONOTONIC,
+	ERR_OUT_OF_INTERVAL,
+} parseerror_code_t;
+static void add_parseerror_codes(PyObject *module)
+{
+	PyModule_AddIntMacro(module, ERR_OTHER);
+	PyModule_AddIntMacro(module, ERR_NON_MONOTONIC);
+	PyModule_AddIntMacro(module, ERR_OUT_OF_INTERVAL);
+}
+
+/* Helpers to raise ParseErrors.  Use "return raise_str(...)" etc. */
+static PyObject *raise_str(int line, int col, int code, const char *string)
+{
+	PyObject *o;
+	o = Py_BuildValue("(iiis)", line, col, code, string);
+	if (o != NULL) {
+		PyErr_SetObject(ParseError, o);
+		Py_DECREF(o);
+	}
+	return NULL;
+}
+static PyObject *raise_int(int line, int col, int code, int64_t num)
+{
+	PyObject *o;
+	o = Py_BuildValue("(iiiL)", line, col, code, (long long)num);
+	if (o != NULL) {
+		PyErr_SetObject(ParseError, o);
+		Py_DECREF(o);
+	}
+	return NULL;
+}
+
+/****
+ * Layout and type helpers
+ */
+typedef union {
+	int8_t i;
+	uint8_t u;
+} union8_t;
+typedef union {
+	int16_t i;
+	uint16_t u;
+} union16_t;
+typedef union {
+	int32_t i;
+	uint32_t u;
+	float f;
+} union32_t;
+typedef union {
+	int64_t i;
+	uint64_t u;
+	double d;
+} union64_t;
+
+typedef enum {
+	LAYOUT_TYPE_NONE,
+	LAYOUT_TYPE_INT8,
+	LAYOUT_TYPE_UINT8,
+	LAYOUT_TYPE_INT16,
+	LAYOUT_TYPE_UINT16,
+	LAYOUT_TYPE_INT32,
+	LAYOUT_TYPE_UINT32,
+	LAYOUT_TYPE_INT64,
+	LAYOUT_TYPE_UINT64,
+	LAYOUT_TYPE_FLOAT32,
+	LAYOUT_TYPE_FLOAT64,
+} layout_type_t;
+
+struct {
+	char *string;
+	layout_type_t layout;
+	int size;
+} type_lookup[] = {
+	{ "int8",    LAYOUT_TYPE_INT8,    1 },
+	{ "uint8",   LAYOUT_TYPE_UINT8,   1 },
+	{ "int16",   LAYOUT_TYPE_INT16,   2 },
+	{ "uint16",  LAYOUT_TYPE_UINT16,  2 },
+	{ "int32",   LAYOUT_TYPE_INT32,   4 },
+	{ "uint32",  LAYOUT_TYPE_UINT32,  4 },
+	{ "int64",   LAYOUT_TYPE_INT64,   8 },
+	{ "uint64",  LAYOUT_TYPE_UINT64,  8 },
+	{ "float32", LAYOUT_TYPE_FLOAT32, 4 },
+	{ "float64", LAYOUT_TYPE_FLOAT64, 8 },
+	{ NULL }
+};
+
+/****
+ * Object definition, init, etc
+ */
+
+/* Rocket object */
+typedef struct {
+	PyObject_HEAD
+	layout_type_t layout_type;
+	int layout_count;
+	int binary_size;
+	FILE *file;
+	int file_size;
+} Rocket;
+
+/* Dealloc / new */
+static void Rocket_dealloc(Rocket *self)
+{
+	if (self->file) {
+		fprintf(stderr, "rocket: file wasn't closed\n");
+		fclose(self->file);
+		self->file = NULL;
+	}
+	Py_TYPE(self)->tp_free((PyObject *)self);
+}
+
+static PyObject *Rocket_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
+{
+	Rocket *self;
+
+	self = (Rocket *)type->tp_alloc(type, 0);
+	if (!self)
+		return NULL;
+	self->layout_type = LAYOUT_TYPE_NONE;
+	self->layout_count = 0;
+	self->binary_size = 0;
+	self->file = NULL;
+	self->file_size = -1;
+	return (PyObject *)self;
+}
+
+/* .__init__(layout, file) */
+static int Rocket_init(Rocket *self, PyObject *args, PyObject *kwds)
+{
+	const char *layout, *path;
+        int pathlen;
+	static char *kwlist[] = { "layout", "file", NULL };
+	if (!PyArg_ParseTupleAndKeywords(args, kwds, "sz#", kwlist,
+					 &layout, &path, &pathlen))
+		return -1;
+	if (!layout)
+		return -1;
+	if (path) {
+                if (strlen(path) != (size_t)pathlen) {
+                        PyErr_SetString(PyExc_ValueError, "path must not "
+                                        "contain NUL characters");
+                        return -1;
+                }
+		if ((self->file = fopen(path, "a+b")) == NULL) {
+			PyErr_SetFromErrno(PyExc_OSError);
+			return -1;
+		}
+		self->file_size = -1;
+	} else {
+		self->file = NULL;
+	}
+
+	const char *under;
+	char *tmp;
+	under = strchr(layout, '_');
+	if (!under) {
+		PyErr_SetString(PyExc_ValueError, "no such layout: "
+				"badly formatted string");
+		return -1;
+	}
+	self->layout_count = strtoul(under+1, &tmp, 10);
+	if (self->layout_count < 1 || *tmp != '\0') {
+		PyErr_SetString(PyExc_ValueError, "no such layout: "
+				"bad count");
+		return -1;
+	}
+	if (self->layout_count >= MAX_LAYOUT_COUNT) {
+		PyErr_SetString(PyExc_ValueError, "no such layout: "
+				"count too high");
+		return -1;
+	}
+
+	int i;
+	for (i = 0; type_lookup[i].string; i++)
+		if (strncmp(layout, type_lookup[i].string, under-layout) == 0)
+			break;
+	if (!type_lookup[i].string) {
+		PyErr_SetString(PyExc_ValueError, "no such layout: "
+				"bad data type");
+		return -1;
+	}
+	self->layout_type = type_lookup[i].layout;
+	self->binary_size = 8 + (type_lookup[i].size * self->layout_count);
+
+	return 0;
+}
+
+/* .close() */
+static PyObject *Rocket_close(Rocket *self)
+{
+	if (self->file) {
+		fclose(self->file);
+		self->file = NULL;
+	}
+	Py_INCREF(Py_None);
+	return Py_None;
+}
+
+/* .file_size property */
+static PyObject *Rocket_get_file_size(Rocket *self)
+{
+	if (!self->file) {
+		PyErr_SetString(PyExc_AttributeError, "no file");
+		return NULL;
+	}
+	if (self->file_size < 0) {
+		int oldpos;
+		if (((oldpos = ftell(self->file)) < 0) ||
+		    (fseek(self->file, 0, SEEK_END) < 0) ||
+		    ((self->file_size = ftell(self->file)) < 0) ||
+		    (fseek(self->file, oldpos, SEEK_SET) < 0)) {
+			PyErr_SetFromErrno(PyExc_OSError);
+			return NULL;
+		}
+	}
+	return PyLong_FromLong(self->file_size);
+}
+
+/****
+ * Append from string
+ */
+static inline long int strtoll10(const char *nptr, char **endptr) {
+	return strtoll(nptr, endptr, 10);
+}
+static inline long int strtoull10(const char *nptr, char **endptr) {
+	return strtoull(nptr, endptr, 10);
+}
+
+/* .append_string(count, data, offset, linenum, start, end, last_timestamp) */
+static PyObject *Rocket_append_string(Rocket *self, PyObject *args)
+{
+	int count;
+	const char *data;
+	int offset;
+	const char *linestart;
+	int linenum;
+        long long ll1, ll2, ll3;
+	timestamp_t start;
+	timestamp_t end;
+	timestamp_t last_timestamp;
+
+	int written = 0;
+	char *endptr;
+	union8_t t8;
+	union16_t t16;
+	union32_t t32;
+	union64_t t64;
+	int i;
+
+        /* Input data is bytes.  Using 'y#' instead of 'y' might be
+           preferable, but strto* requires the null terminator. */
+	if (!PyArg_ParseTuple(args, "iyiiLLL:append_string", &count,
+			      &data, &offset, &linenum,
+			      &ll1, &ll2, &ll3))
+		return NULL;
+        start = ll1;
+        end = ll2;
+        last_timestamp = ll3;
+
+	/* Skip spaces, but don't skip over a newline. */
+#define SKIP_BLANK(buf) do {			\
+	while (isspace(*buf)) {			\
+		if (*buf == '\n')		\
+			break;			\
+		buf++;				\
+	} } while(0)
+
+	const char *buf = &data[offset];
+	while (written < count && *buf)
+	{
+		linestart = buf;
+		linenum++;
+
+		/* Skip leading whitespace and commented lines */
+		SKIP_BLANK(buf);
+		if (*buf == '#') {
+			while (*buf && *buf != '\n')
+				buf++;
+			if (*buf)
+				buf++;
+			continue;
+		}
+
+		/* Extract timestamp */
+		t64.i = strtoll(buf, &endptr, 10);
+		if (endptr == buf || !isspace(*endptr)) {
+			/* Try parsing as a double instead */
+			t64.d = strtod(buf, &endptr);
+			if (endptr == buf)
+				goto bad_timestamp;
+			if (!isspace(*endptr))
+				goto cant_parse_value;
+			t64.i = round(t64.d);
+		}
+		if (t64.i <= last_timestamp)
+			return raise_int(linenum, buf - linestart + 1,
+					 ERR_NON_MONOTONIC, t64.i);
+		last_timestamp = t64.i;
+		if (t64.i < start || t64.i >= end)
+			return raise_int(linenum, buf - linestart + 1,
+					 ERR_OUT_OF_INTERVAL, t64.i);
+		t64.u = le64toh(t64.u);
+		if (fwrite(&t64.u, 8, 1, self->file) != 1)
+			goto err;
+		buf = endptr;
+
+		/* Parse all values in the line */
+		switch (self->layout_type) {
+#define CS(type, parsefunc, parsetype, realtype, disktype, letoh, bytes) \
+		case LAYOUT_TYPE_##type:				\
+			/* parse and write in a loop */			\
+			for (i = 0; i < self->layout_count; i++) {	\
+				/* skip non-newlines */			\
+				SKIP_BLANK(buf);			\
+				if (*buf == '\n')			\
+					goto wrong_number_of_values;	\
+				/* parse number */			\
+				parsetype = parsefunc(buf, &endptr);	\
+				if (*endptr && !isspace(*endptr))	\
+					goto cant_parse_value;		\
+				/* check limits */			\
+				if (type##_MIN != type##_MAX &&		\
+				    (parsetype < type##_MIN ||		\
+				     parsetype > type##_MAX))		\
+					goto value_out_of_range;	\
+				/* convert to disk representation */	\
+				realtype = parsetype;			\
+				disktype = letoh(disktype);		\
+				/* write it */				\
+				if (fwrite(&disktype, bytes,		\
+					   1, self->file) != 1)		\
+					goto err;			\
+				/* advance buf */			\
+				buf = endptr;				\
+			}						\
+			/* Skip trailing whitespace and comments */	\
+			SKIP_BLANK(buf);				\
+			if (*buf == '#')				\
+				while (*buf && *buf != '\n')		\
+					buf++;				\
+			if (*buf == '\n')				\
+				buf++;					\
+			else if (*buf != '\0')				\
+				goto extra_data_on_line;		\
+			break
+
+			CS(INT8,   strtoll10,  t64.i, t8.i,  t8.u,         , 1);
+			CS(UINT8,  strtoull10, t64.u, t8.u,  t8.u,         , 1);
+			CS(INT16,  strtoll10,  t64.i, t16.i, t16.u, le16toh, 2);
+			CS(UINT16, strtoull10, t64.u, t16.u, t16.u, le16toh, 2);
+			CS(INT32,  strtoll10,  t64.i, t32.i, t32.u, le32toh, 4);
+			CS(UINT32, strtoull10, t64.u, t32.u, t32.u, le32toh, 4);
+			CS(INT64,  strtoll10,  t64.i, t64.i, t64.u, le64toh, 8);
+			CS(UINT64, strtoull10, t64.u, t64.u, t64.u, le64toh, 8);
+			CS(FLOAT32, strtod,   t64.d, t32.f, t32.u, le32toh, 4);
+			CS(FLOAT64, strtod,   t64.d, t64.d, t64.u, le64toh, 8);
+#undef CS
+		default:
+			PyErr_SetString(PyExc_TypeError, "unknown type");
+			return NULL;
+		}
+
+		/* Done this line */
+		written++;
+	}
+
+	fflush(self->file);
+
+	/* Build return value and return */
+	offset = buf - data;
+	PyObject *o;
+	o = Py_BuildValue("(iiLi)", written, offset,
+                          (long long)last_timestamp, linenum);
+	return o;
+err:
+	PyErr_SetFromErrno(PyExc_OSError);
+	return NULL;
+bad_timestamp:
+	return raise_str(linenum, buf - linestart + 1,
+			 ERR_OTHER, "bad timestamp");
+cant_parse_value:
+	return raise_str(linenum, buf - linestart + 1,
+			 ERR_OTHER, "can't parse value");
+wrong_number_of_values:
+	return raise_str(linenum, buf - linestart + 1,
+			 ERR_OTHER, "wrong number of values");
+value_out_of_range:
+	return raise_str(linenum, buf - linestart + 1,
+			 ERR_OTHER, "value out of range");
+extra_data_on_line:
+	return raise_str(linenum, buf - linestart + 1,
+			 ERR_OTHER, "extra data on line");
+}
+
+/****
+ * Append from binary data
+ */
+
+/* .append_binary(count, data, offset, linenum, start, end, last_timestamp) */
+static PyObject *Rocket_append_binary(Rocket *self, PyObject *args)
+{
+        int count;
+	const uint8_t *data;
+        int data_len;
+        int linenum;
+	int offset;
+        long long ll1, ll2, ll3;
+	timestamp_t start;
+	timestamp_t end;
+	timestamp_t last_timestamp;
+
+	if (!PyArg_ParseTuple(args, "iy#iiLLL:append_binary",
+                              &count, &data, &data_len, &offset,
+                              &linenum, &ll1, &ll2, &ll3))
+		return NULL;
+        start = ll1;
+        end = ll2;
+        last_timestamp = ll3;
+
+        /* Advance to offset */
+        if (offset > data_len)
+                return raise_str(0, 0, ERR_OTHER, "bad offset");
+        data += offset;
+        data_len -= offset;
+
+        /* Figure out max number of rows to insert */
+        int rows = data_len / self->binary_size;
+        if (rows > count)
+                rows = count;
+
+        /* Check timestamps */
+        timestamp_t ts;
+	int i;
+        for (i = 0; i < rows; i++) {
+                /* Read raw timestamp, byteswap if needed */
+                memcpy(&ts, &data[i * self->binary_size], 8);
+                ts = le64toh(ts);
+
+                /* Check limits */
+                if (ts <= last_timestamp)
+                        return raise_int(i, 0, ERR_NON_MONOTONIC, ts);
+                last_timestamp = ts;
+                if (ts < start || ts >= end)
+                        return raise_int(i, 0, ERR_OUT_OF_INTERVAL, ts);
+        }
+
+        /* Write binary data */
+        if (fwrite(data, self->binary_size, rows, self->file) != (size_t)rows) {
+                PyErr_SetFromErrno(PyExc_OSError);
+                return NULL;
+        }
+	fflush(self->file);
+
+	/* Build return value and return */
+	PyObject *o;
+	o = Py_BuildValue("(iiLi)", rows, offset + rows * self->binary_size,
+                          (long long)last_timestamp, linenum);
+	return o;
+}
+
+/****
+ * Extract to binary bytes object containing ASCII text-formatted data
+ */
+
+static PyObject *Rocket_extract_string(Rocket *self, PyObject *args)
+{
+	long count;
+	long offset;
+
+	if (!PyArg_ParseTuple(args, "ll", &offset, &count))
+		return NULL;
+	if (!self->file) {
+		PyErr_SetString(PyExc_Exception, "no file");
+		return NULL;
+	}
+	/* Seek to target location */
+	if (fseek(self->file, offset, SEEK_SET) < 0) {
+		PyErr_SetFromErrno(PyExc_OSError);
+		return NULL;
+	}
+
+	char *str = NULL, *new;
+	long len_alloc = 0;
+	long len = 0;
+	int ret;
+
+	/* min space free in string (and the maximum length of one
+	   line); this is generous */
+	const int min_free = 32 * MAX_LAYOUT_COUNT;
+
+	/* how much to allocate at once */
+	const int alloc_size = 1048576;
+
+	int row, i;
+	union8_t t8;
+	union16_t t16;
+	union32_t t32;
+	union64_t t64;
+	for (row = 0; row < count; row++) {
+		/* Make sure there's space for a line */
+		if ((len_alloc - len) < min_free) {
+			/* grow by 1 meg at a time */
+			len_alloc += alloc_size;
+			new = realloc(str, len_alloc);
+			if (new == NULL)
+				goto err;
+			str = new;
+		}
+
+		/* Read and print timestamp */
+		if (fread(&t64.u, 8, 1, self->file) != 1)
+			goto err;
+		t64.u = le64toh(t64.u);
+		ret = sprintf(&str[len], "%" PRId64, t64.i);
+		if (ret <= 0)
+			goto err;
+		len += ret;
+
+		/* Read and print values */
+		switch (self->layout_type) {
+#define CASE(type, fmt, fmttype, disktype, letoh, bytes)		\
+		case LAYOUT_TYPE_##type:				\
+			/* read and format in a loop */			\
+			for (i = 0; i < self->layout_count; i++) {	\
+				if (fread(&disktype, bytes,		\
+					  1, self->file) != 1)		\
+					goto err;			\
+				disktype = letoh(disktype);		\
+				ret = sprintf(&str[len], " " fmt,	\
+					      fmttype);			\
+				if (ret <= 0)				\
+					goto err;			\
+				len += ret;				\
+			}						\
+			break
+			CASE(INT8,   "%" PRId8,  t8.i,  t8.u,         , 1);
+			CASE(UINT8,  "%" PRIu8,  t8.u,  t8.u,         , 1);
+			CASE(INT16,  "%" PRId16, t16.i, t16.u, le16toh, 2);
+			CASE(UINT16, "%" PRIu16, t16.u, t16.u, le16toh, 2);
+			CASE(INT32,  "%" PRId32, t32.i, t32.u, le32toh, 4);
+			CASE(UINT32, "%" PRIu32, t32.u, t32.u, le32toh, 4);
+			CASE(INT64,  "%" PRId64, t64.i, t64.u, le64toh, 8);
+			CASE(UINT64, "%" PRIu64, t64.u, t64.u, le64toh, 8);
+			/* These next two are a bit debatable.  floats
+			   are 6-9 significant figures, so we print 7.
+			   Doubles are 15-19, so we print 17.  This is
+			   similar to the old prep format for float32.
+			*/
+			CASE(FLOAT32, "%.6e",  t32.f, t32.u, le32toh, 4);
+			CASE(FLOAT64, "%.16e", t64.d, t64.u, le64toh, 8);
+#undef CASE
+		default:
+			PyErr_SetString(PyExc_TypeError, "unknown type");
+			if (str) free(str);
+			return NULL;
+		}
+		str[len++] = '\n';
+	}
+
+	PyObject *pystr = PyBytes_FromStringAndSize(str, len);
+	free(str);
+	return pystr;
+err:
+	if (str) free(str);
+	PyErr_SetFromErrno(PyExc_OSError);
+	return NULL;
+}
+
+/****
+ * Extract to binary bytes object containing raw little-endian binary data
+ */
+static PyObject *Rocket_extract_binary(Rocket *self, PyObject *args)
+{
+	long count;
+	long offset;
+
+	if (!PyArg_ParseTuple(args, "ll", &offset, &count))
+		return NULL;
+	if (!self->file) {
+		PyErr_SetString(PyExc_Exception, "no file");
+		return NULL;
+	}
+	/* Seek to target location */
+	if (fseek(self->file, offset, SEEK_SET) < 0) {
+		PyErr_SetFromErrno(PyExc_OSError);
+		return NULL;
+	}
+
+        uint8_t *str;
+        int len = count * self->binary_size;
+        str = malloc(len);
+        if (str == NULL) {
+                PyErr_SetFromErrno(PyExc_OSError);
+                return NULL;
+        }
+
+        /* Data in the file is already in the desired little-endian
+           binary format, so just read it directly. */
+        if (fread(str, self->binary_size, count, self->file) != (size_t)count) {
+                free(str);
+                PyErr_SetFromErrno(PyExc_OSError);
+                return NULL;
+        }
+
+	PyObject *pystr = PyBytes_FromStringAndSize((char *)str, len);
+	free(str);
+	return pystr;
+}
+
+/****
+ * Extract timestamp
+ */
+static PyObject *Rocket_extract_timestamp(Rocket *self, PyObject *args)
+{
+	long offset;
+	union64_t t64;
+	if (!PyArg_ParseTuple(args, "l", &offset))
+		return NULL;
+	if (!self->file) {
+		PyErr_SetString(PyExc_Exception, "no file");
+		return NULL;
+	}
+
+	/* Seek to target location and read timestamp */
+	if ((fseek(self->file, offset, SEEK_SET) < 0) ||
+	    (fread(&t64.u, 8, 1, self->file) != 1)) {
+		PyErr_SetFromErrno(PyExc_OSError);
+		return NULL;
+	}
+
+	/* Convert and return */
+	t64.u = le64toh(t64.u);
+	return Py_BuildValue("L", (long long)t64.i);
+}
+
+/****
+ * Module and type setup
+ */
+
+static PyGetSetDef Rocket_getsetters[] = {
+	{ "file_size", (getter)Rocket_get_file_size, NULL,
+	  "file size in bytes", NULL },
+	{ NULL },
+};
+
+static PyMemberDef Rocket_members[] = {
+	{ "binary_size", T_INT, offsetof(Rocket, binary_size), 0,
+	  "binary size per row" },
+	{ NULL },
+};
+
+static PyMethodDef Rocket_methods[] = {
+	{ "close",
+          (PyCFunction)Rocket_close, METH_NOARGS,
+	  "close(self)\n\n"
+	  "Close file handle" },
+
+	{ "append_string",
+          (PyCFunction)Rocket_append_string, METH_VARARGS,
+	  "append_string(self, count, data, offset, line, start, end, ts)\n\n"
+          "Parse string and append data.\n"
+	  "\n"
+	  "  count: maximum number of rows to add\n"
+          "  data: string data\n"
+          "  offset: byte offset into data to start parsing\n"
+          "  line: current line number of data\n"
+          "  start: starting timestamp for interval\n"
+          "  end: end timestamp for interval\n"
+          "  ts: last timestamp that was previously parsed\n"
+	  "\n"
+	  "Raises ParseError if timestamps are non-monotonic, outside\n"
+	  "the start/end interval etc.\n"
+	  "\n"
+          "On success, return a tuple:\n"
+          "  added_rows: how many rows were added from the file\n"
+          "  data_offset: current offset into the data string\n"
+          "  last_timestamp: last timestamp we parsed\n"
+          "  linenum: current line number" },
+
+	{ "append_binary",
+	  (PyCFunction)Rocket_append_binary, METH_VARARGS,
+	  "append_binary(self, count, data, offset, line, start, end, ts)\n\n"
+          "Append binary data, which must match the data layout.\n"
+	  "\n"
+	  "  count: maximum number of rows to add\n"
+          "  data: binary data\n"
+          "  offset: byte offset into data to start adding\n"
+          "  line: current line number (unused)\n"
+          "  start: starting timestamp for interval\n"
+          "  end: end timestamp for interval\n"
+          "  ts: last timestamp that was previously parsed\n"
+	  "\n"
+	  "Raises ParseError if timestamps are non-monotonic, outside\n"
+	  "the start/end interval etc.\n"
+	  "\n"
+          "On success, return a tuple:\n"
+          "  added_rows: how many rows were added from the file\n"
+          "  data_offset: current offset into the data string\n"
+          "  last_timestamp: last timestamp we parsed\n"
+          "  linenum: current line number (copied from argument)" },
+
+	{ "extract_string",
+          (PyCFunction)Rocket_extract_string, METH_VARARGS,
+	  "extract_string(self, offset, count)\n\n"
+	  "Extract count rows of data from the file at offset offset.\n"
+	  "Return an ascii formatted string according to the layout" },
+
+	{ "extract_binary",
+	  (PyCFunction)Rocket_extract_binary, METH_VARARGS,
+	  "extract_binary(self, offset, count)\n\n"
+	  "Extract count rows of data from the file at offset offset.\n"
+	  "Return a raw binary string of data matching the data layout." },
+
+	{ "extract_timestamp",
+	  (PyCFunction)Rocket_extract_timestamp, METH_VARARGS,
+	  "extract_timestamp(self, offset)\n\n"
+	  "Extract a single timestamp from the file" },
+
+	{ NULL },
+};
+
+static PyTypeObject RocketType = {
+	PyVarObject_HEAD_INIT(NULL, 0)
+
+	.tp_name	= "rocket.Rocket",
+	.tp_basicsize	= sizeof(Rocket),
+	.tp_flags	= Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE,
+
+	.tp_new		= Rocket_new,
+	.tp_dealloc	= (destructor)Rocket_dealloc,
+	.tp_init	= (initproc)Rocket_init,
+	.tp_methods	= Rocket_methods,
+	.tp_members	= Rocket_members,
+	.tp_getset	= Rocket_getsetters,
+
+	.tp_doc		= ("rocket.Rocket(layout, file)\n\n"
+			   "C implementation of the \"rocket\" data parsing\n"
+			   "interface, which translates between the binary\n"
+			   "format on disk and the ASCII or Python list\n"
+			   "format used when communicating with the rest of\n"
+			   "the system.")
+};
+
+static PyMethodDef module_methods[] = {
+	{ NULL },
+};
+
+static struct PyModuleDef moduledef = {
+        PyModuleDef_HEAD_INIT,
+        .m_name        = "rocker",
+        .m_doc         = "Rocket data parsing and formatting module",
+        .m_size        = -1,
+        .m_methods     = module_methods,
+};
+
+PyMODINIT_FUNC PyInit_rocket(void)
+{
+	PyObject *module;
+
+	RocketType.tp_new = PyType_GenericNew;
+	if (PyType_Ready(&RocketType) < 0)
+		return NULL;
+
+	module = PyModule_Create(&moduledef);
+	Py_INCREF(&RocketType);
+	PyModule_AddObject(module, "Rocket", (PyObject *)&RocketType);
+
+	ParseError = PyErr_NewException("rocket.ParseError", NULL, NULL);
+	Py_INCREF(ParseError);
+	PyModule_AddObject(module, "ParseError", ParseError);
+	add_parseerror_codes(module);
+
+	return module;
+}

nilmdb/server/server.py

@@ -0,0 +1,546 @@
+"""CherryPy-based server for accessing NILM database via HTTP"""
+
+import os
+import json
+import socket
+import traceback
+
+import psutil
+import cherrypy
+
+import nilmdb.server
+from nilmdb.utils.printf import sprintf
+from nilmdb.server.errors import NilmDBError
+from nilmdb.utils.time import string_to_timestamp
+
+from nilmdb.server.serverutil import (
+    chunked_response,
+    response_type,
+    exception_to_httperror,
+    CORS_allow,
+    json_to_request_params,
+    json_error_page,
+    cherrypy_start,
+    cherrypy_stop,
+    bool_param,
+    )
+
+# Add CORS_allow tool
+cherrypy.tools.CORS_allow = cherrypy.Tool('on_start_resource', CORS_allow)
+
+
+class NilmApp():
+    def __init__(self, db):
+        self.db = db
+
+
+# CherryPy apps
+class Root(NilmApp):
+    """Root application for NILM database"""
+    # /
+    @cherrypy.expose
+    def index(self):
+        cherrypy.response.headers['Content-Type'] = 'text/plain'
+        msg = sprintf("This is NilmDB version %s, running on host %s.\n",
+                      nilmdb.__version__, socket.getfqdn())
+        return msg
+
+    # /favicon.ico
+    @cherrypy.expose
+    def favicon_ico(self):
+        raise cherrypy.NotFound()
+
+    # /version
+    @cherrypy.expose
+    @cherrypy.tools.json_out()
+    def version(self):
+        return nilmdb.__version__
+
+    # /dbinfo
+    @cherrypy.expose
+    @cherrypy.tools.json_out()
+    def dbinfo(self):
+        """Return a dictionary with the database path,
+        size of the database in bytes, and free disk space in bytes"""
+        path = self.db.get_basepath()
+        usage = psutil.disk_usage(path)
+        dbsize = nilmdb.utils.du(path)
+        return {
+            "path": path,
+            "size": dbsize,
+            "other": max(usage.used - dbsize, 0),
+            "reserved": max(usage.total - usage.used - usage.free, 0),
+            "free": usage.free
+        }
+
+
+class Stream(NilmApp):
+    """Stream-specific operations"""
+
+    # Helpers
+    def _get_times(self, start_param, end_param):
+        (start, end) = (None, None)
+        try:
+            if start_param is not None:
+                start = string_to_timestamp(start_param)
+        except Exception:
+            raise cherrypy.HTTPError("400 Bad Request", sprintf(
+                "invalid start (%s): must be a numeric timestamp",
+                start_param))
+        try:
+            if end_param is not None:
+                end = string_to_timestamp(end_param)
+        except Exception:
+            raise cherrypy.HTTPError("400 Bad Request", sprintf(
+                "invalid end (%s): must be a numeric timestamp", end_param))
+        if start is not None and end is not None:
+            if start >= end:
+                raise cherrypy.HTTPError(
+                    "400 Bad Request",
+                    sprintf("start must precede end (%s >= %s)",
+                            start_param, end_param))
+        return (start, end)
+
+    # /stream/list
+    # /stream/list?layout=float32_8
+    # /stream/list?path=/newton/prep&extended=1
+    @cherrypy.expose
+    @cherrypy.tools.json_out()
+    def list(self, path=None, layout=None, extended=None):
+        """List all streams in the database.  With optional path or
+        layout parameter, just list streams that match the given path
+        or layout.
+
+        If extended is missing or zero, returns a list of lists
+        containing the path and layout: [ path, layout ]
+
+        If extended is true, returns a list of lists containing
+        extended info: [ path, layout, extent_min, extent_max,
+        total_rows, total_seconds ].  More data may be added.
+        """
+        return self.db.stream_list(path, layout, bool(extended))
+
+    # /stream/create?path=/newton/prep&layout=float32_8
+    @cherrypy.expose
+    @cherrypy.tools.json_in()
+    @cherrypy.tools.json_out()
+    @exception_to_httperror(NilmDBError, ValueError)
+    @cherrypy.tools.CORS_allow(methods=["POST"])
+    def create(self, path, layout):
+        """Create a new stream in the database.  Provide path
+        and one of the nilmdb.layout.layouts keys.
+        """
+        return self.db.stream_create(path, layout)
+
+    # /stream/destroy?path=/newton/prep
+    @cherrypy.expose
+    @cherrypy.tools.json_in()
+    @cherrypy.tools.json_out()
+    @exception_to_httperror(NilmDBError)
+    @cherrypy.tools.CORS_allow(methods=["POST"])
+    def destroy(self, path):
+        """Delete a stream.  Fails if any data is still present."""
+        return self.db.stream_destroy(path)
+
+    # /stream/rename?oldpath=/newton/prep&newpath=/newton/prep/1
+    @cherrypy.expose
+    @cherrypy.tools.json_in()
+    @cherrypy.tools.json_out()
+    @exception_to_httperror(NilmDBError, ValueError)
+    @cherrypy.tools.CORS_allow(methods=["POST"])
+    def rename(self, oldpath, newpath):
+        """Rename a stream."""
+        return self.db.stream_rename(oldpath, newpath)
+
+    # /stream/get_metadata?path=/newton/prep
+    # /stream/get_metadata?path=/newton/prep&key=foo&key=bar
+    @cherrypy.expose
+    @cherrypy.tools.json_out()
+    def get_metadata(self, path, key=None):
+        """Get metadata for the named stream.  If optional
+        key parameters are specified, only return metadata
+        matching the given keys."""
+        try:
+            data = self.db.stream_get_metadata(path)
+        except nilmdb.server.nilmdb.StreamError as e:
+            raise cherrypy.HTTPError("404 Not Found", str(e))
+        if key is None:  # If no keys specified, return them all
+            key = list(data.keys())
+        elif not isinstance(key, list):
+            key = [key]
+        result = {}
+        for k in key:
+            if k in data:
+                result[k] = data[k]
+            else:  # Return "None" for keys with no matching value
+                result[k] = None
+        return result
+
+    # Helper for set_metadata and get_metadata
+    def _metadata_helper(self, function, path, data):
+        if not isinstance(data, dict):
+            try:
+                data = dict(json.loads(data))
+            except TypeError as e:
+                raise NilmDBError("can't parse 'data' parameter: " + str(e))
+        for key in data:
+            if not isinstance(data[key], (str, float, int)):
+                raise NilmDBError("metadata values must be a string or number")
+        function(path, data)
+
+    # /stream/set_metadata?path=/newton/prep&data=<json>
+    @cherrypy.expose
+    @cherrypy.tools.json_in()
+    @cherrypy.tools.json_out()
+    @exception_to_httperror(NilmDBError, LookupError)
+    @cherrypy.tools.CORS_allow(methods=["POST"])
+    def set_metadata(self, path, data):
+        """Set metadata for the named stream, replacing any existing
+        metadata.  Data can be json-encoded or a plain dictionary."""
+        self._metadata_helper(self.db.stream_set_metadata, path, data)
+
+    # /stream/update_metadata?path=/newton/prep&data=<json>
+    @cherrypy.expose
+    @cherrypy.tools.json_in()
+    @cherrypy.tools.json_out()
+    @exception_to_httperror(NilmDBError, LookupError, ValueError)
+    @cherrypy.tools.CORS_allow(methods=["POST"])
+    def update_metadata(self, path, data):
+        """Set metadata for the named stream, replacing any existing
+        metadata.  Data can be json-encoded or a plain dictionary."""
+        self._metadata_helper(self.db.stream_update_metadata, path, data)
+
+    # /stream/insert?path=/newton/prep
+    @cherrypy.expose
+    @cherrypy.tools.json_out()
+    @exception_to_httperror(NilmDBError, ValueError)
+    @cherrypy.tools.CORS_allow(methods=["PUT"])
+    def insert(self, path, start, end, binary=False):
+        """
+        Insert new data into the database.  Provide textual data
+        (matching the path's layout) as a HTTP PUT.
+
+        If 'binary' is True, expect raw binary data, rather than lines
+        of ASCII-formatted data.  Raw binary data is always
+        little-endian and matches the database types (including an
+        int64 timestamp).
+        """
+        binary = bool_param(binary)
+
+        # Important that we always read the input before throwing any
+        # errors, to keep lengths happy for persistent connections.
+        # Note that CherryPy 3.2.2 has a bug where this fails for GET
+        # requests, if we ever want to handle those (issue #1134)
+        body = cherrypy.request.body.read()
+
+        # Verify content type for binary data
+        content_type = cherrypy.request.headers.get('content-type')
+        if binary and content_type:
+            if content_type != "application/octet-stream":
+                raise cherrypy.HTTPError("400", "Content type must be "
+                                         "application/octet-stream for "
+                                         "binary data, not " + content_type)
+
+        # Note that non-binary data is *not* decoded from bytes to string,
+        # but rather passed directly to stream_insert.
+
+        # Check path and get layout
+        if len(self.db.stream_list(path=path)) != 1:
+            raise cherrypy.HTTPError("404", "No such stream: " + path)
+
+        # Check limits
+        (start, end) = self._get_times(start, end)
+
+        # Pass the data directly to nilmdb, which will parse it and
+        # raise a ValueError if there are any problems.
+        self.db.stream_insert(path, start, end, body, binary)
+
+        # Done
+        return
+
+    # /stream/remove?path=/newton/prep
+    # /stream/remove?path=/newton/prep&start=1234567890.0&end=1234567899.0
+    @cherrypy.expose
+    @cherrypy.tools.json_in()
+    @cherrypy.tools.CORS_allow(methods=["POST"])
+    @chunked_response
+    @response_type("application/x-json-stream")
+    def remove(self, path, start=None, end=None):
+        """
+        Remove data from the backend database.  Removes all data in
+        the interval [start, end).
+
+        Returns the number of data points removed.  Since this is a potentially
+        long-running operation, multiple numbers may be returned as the
+        data gets removed from the backend database.  The total number of
+        points removed is the sum of all of these numbers.
+        """
+        (start, end) = self._get_times(start, end)
+
+        if len(self.db.stream_list(path=path)) != 1:
+            raise cherrypy.HTTPError("404", "No such stream: " + path)
+
+        def content(start, end):
+            # Note: disable chunked responses to see tracebacks from here.
+            while True:
+                (removed, restart) = self.db.stream_remove(path, start, end)
+                response = json.dumps(removed) + "\r\n"
+                yield response.encode('utf-8')
+                if restart is None:
+                    break
+                start = restart
+        return content(start, end)
+
+    # /stream/intervals?path=/newton/prep
+    # /stream/intervals?path=/newton/prep&start=1234567890.0&end=1234567899.0
+    # /stream/intervals?path=/newton/prep&diffpath=/newton/prep2
+    @cherrypy.expose
+    @chunked_response
+    @response_type("application/x-json-stream")
+    def intervals(self, path, start=None, end=None, diffpath=None):
+        """
+        Get intervals from backend database.  Streams the resulting
+        intervals as JSON strings separated by CR LF pairs.  This may
+        make multiple requests to the nilmdb backend to avoid causing
+        it to block for too long.
+
+        Returns intervals between 'start' and 'end' belonging to
+        'path'.  If 'diff' is provided, the set-difference between
+        intervals in 'path' and intervals in 'diffpath' are
+        returned instead.
+
+        Note that the response type is the non-standard
+        'application/x-json-stream' for lack of a better option.
+        """
+        (start, end) = self._get_times(start, end)
+
+        if len(self.db.stream_list(path=path)) != 1:
+            raise cherrypy.HTTPError("404", "No such stream: " + path)
+
+        if diffpath and len(self.db.stream_list(path=diffpath)) != 1:
+            raise cherrypy.HTTPError("404", "No such stream: " + diffpath)
+
+        def content(start, end):
+            # Note: disable chunked responses to see tracebacks from here.
+            while True:
+                (ints, restart) = self.db.stream_intervals(path, start, end,
+                                                           diffpath)
+                response = ''.join([json.dumps(i) + "\r\n" for i in ints])
+                yield response.encode('utf-8')
+                if restart is None:
+                    break
+                start = restart
+        return content(start, end)
+
+    # /stream/extract?path=/newton/prep&start=1234567890.0&end=1234567899.0
+    @cherrypy.expose
+    @chunked_response
+    def extract(self, path, start=None, end=None,
+                count=False, markup=False, binary=False):
+        """
+        Extract data from backend database.  Streams the resulting
+        entries as ASCII text lines separated by newlines.  This may
+        make multiple requests to the nilmdb backend to avoid causing
+        it to block for too long.
+
+        If 'count' is True, returns a count rather than actual data.
+
+        If 'markup' is True, adds comments to the stream denoting each
+        interval's start and end timestamp.
+
+        If 'binary' is True, return raw binary data, rather than lines
+        of ASCII-formatted data.  Raw binary data is always
+        little-endian and matches the database types (including an
+        int64 timestamp).
+        """
+        binary = bool_param(binary)
+        markup = bool_param(markup)
+        count = bool_param(count)
+
+        (start, end) = self._get_times(start, end)
+
+        # Check path and get layout
+        if len(self.db.stream_list(path=path)) != 1:
+            raise cherrypy.HTTPError("404", "No such stream: " + path)
+
+        if binary:
+            content_type = "application/octet-stream"
+            if markup or count:
+                raise cherrypy.HTTPError("400", "can't mix binary and "
+                                         "markup or count modes")
+        else:
+            content_type = "text/plain"
+        cherrypy.response.headers['Content-Type'] = content_type
+
+        def content(start, end):
+            # Note: disable chunked responses to see tracebacks from here.
+            if count:
+                matched = self.db.stream_extract(path, start, end,
+                                                 count=True)
+                yield sprintf(b"%d\n", matched)
+                return
+
+            while True:
+                (data, restart) = self.db.stream_extract(
+                    path, start, end, count=False,
+                    markup=markup, binary=binary)
+                yield data
+
+                if restart is None:
+                    return
+                start = restart
+        return content(start, end)
+
+
+class Exiter():
+    """App that exits the server, for testing"""
+    @cherrypy.expose
+    def index(self):
+        cherrypy.response.headers['Content-Type'] = 'text/plain'
+
+        def content():
+            yield b'Exiting by request'
+            raise SystemExit
+
+        return content()
+    index._cp_config = {'response.stream': True}
+
+
+class Server():
+    def __init__(self, db, host='127.0.0.1', port=8080,
+                 stoppable=False,        # whether /exit URL exists
+                 fast_shutdown=False,    # don't wait for clients to disconn.
+                 force_traceback=False,  # include traceback in all errors
+                 basepath='',            # base URL path for cherrypy.tree
+                 ):
+        # Save server version, just for verification during tests
+        self.version = nilmdb.__version__
+
+        self.db = db
+        if not getattr(db, "_thread_safe", None):
+            raise KeyError("Database object " + str(db) + " doesn't claim "
+                           "to be thread safe.  You should pass "
+                           "nilmdb.utils.serializer_proxy(NilmDB)(args) "
+                           "rather than NilmDB(args).")
+
+        # Build up global server configuration
+        cherrypy.config.update({
+            'environment': 'embedded',
+            'server.socket_host': host,
+            'server.socket_port': port,
+            'engine.autoreload.on': False,
+            'server.max_request_body_size': 8*1024*1024,
+            })
+
+        # Build up application specific configuration
+        app_config = {}
+        app_config.update({
+            'error_page.default': self.json_error_page,
+            })
+
+        # Some default headers to just help identify that things are working
+        app_config.update({'response.headers.X-Jim-Is-Awesome': 'yeah'})
+
+        # Set up Cross-Origin Resource Sharing (CORS) handler so we
+        # can correctly respond to browsers' CORS preflight requests.
+        # This also limits verbs to GET and HEAD by default.
+        app_config.update({'tools.CORS_allow.on': True,
+                           'tools.CORS_allow.methods': ['GET', 'HEAD']})
+
+        # Configure the 'json_in' tool to also allow other content-types
+        # (like x-www-form-urlencoded), and to treat JSON as a dict that
+        # fills requests.param.
+        app_config.update({'tools.json_in.force': False,
+                           'tools.json_in.processor': json_to_request_params})
+
+        # Send tracebacks in error responses.  They're hidden by the
+        # error_page function for client errors (code 400-499).
+        app_config.update({'request.show_tracebacks': True})
+        self.force_traceback = force_traceback
+
+        # Patch CherryPy error handler to never pad out error messages.
+        # This isn't necessary, but then again, neither is padding the
+        # error messages.
+        cherrypy._cperror._ie_friendly_error_sizes = {}
+
+        # Build up the application and mount it
+        root = Root(self.db)
+        root.stream = Stream(self.db)
+        if stoppable:
+            root.exit = Exiter()
+        cherrypy.tree.apps = {}
+        cherrypy.tree.mount(root, basepath, config={"/": app_config})
+
+        # Shutdowns normally wait for clients to disconnect.  To speed
+        # up tests, set fast_shutdown = True
+        if fast_shutdown:
+            cherrypy.server.shutdown_timeout = 0
+        else:
+            cherrypy.server.shutdown_timeout = 5
+
+        # Set up the WSGI application pointer for external programs
+        self.wsgi_application = cherrypy.tree
+
+    def json_error_page(self, status, message, traceback, version):
+        """Return a custom error page in JSON so the client can parse it"""
+        return json_error_page(status, message, traceback, version,
+                               self.force_traceback)
+
+    def start(self, blocking=False, event=None):
+        cherrypy_start(blocking, event)
+
+    def stop(self):
+        cherrypy_stop()
+
+
+# Use a single global nilmdb.server.NilmDB and nilmdb.server.Server
+# instance since the database can only be opened once.  For this to
+# work, the web server must use only a single process and single
+# Python interpreter.  Multiple threads are OK.
+_wsgi_server = None
+
+
+def wsgi_application(dbpath, basepath):
+    """Return a WSGI application object with a database at the
+    specified path.
+
+    'dbpath' is a filesystem location, e.g. /home/nilm/db
+
+    'basepath' is the URL path of the application base, which
+    is the same as the first argument to Apache's WSGIScriptAlias
+    directive.
+    """
+    def application(environ, start_response):
+        global _wsgi_server
+        if _wsgi_server is None:
+            # Try to start the server
+            try:
+                db = nilmdb.utils.serializer_proxy(
+                    nilmdb.server.NilmDB)(dbpath)
+                _wsgi_server = nilmdb.server.Server(
+                    db, basepath=basepath.rstrip('/'))
+            except Exception:
+                # Build an error message on failure
+                import pprint
+                err = sprintf("Initializing database at path '%s' failed:\n\n",
+                              dbpath)
+                err += traceback.format_exc()
+                import pwd
+                import grp
+                err += sprintf("\nRunning as: uid=%d (%s), gid=%d (%s) "
+                               "on host %s, pid %d\n",
+                               os.getuid(), pwd.getpwuid(os.getuid())[0],
+                               os.getgid(), grp.getgrgid(os.getgid())[0],
+                               socket.gethostname(), os.getpid())
+                err += sprintf("\nEnvironment:\n%s\n", pprint.pformat(environ))
+        if _wsgi_server is None:
+            # Serve up the error with our own mini WSGI app.
+            err_b = err.encode('utf-8')
+            headers = [('Content-type', 'text/plain; charset=utf-8'),
+                       ('Content-length', str(len(err_b)))]
+            start_response("500 Internal Server Error", headers)
+            return [err_b]
+
+        # Call the normal application
+        return _wsgi_server.wsgi_application(environ, start_response)
+    return application

nilmdb/server/serverutil.py

@@ -0,0 +1,225 @@
+"""Miscellaneous decorators and other helpers for running a CherryPy
+server"""
+
+import os
+import sys
+import json
+import decorator
+import functools
+import threading
+
+import cherrypy
+
+
+# Helper to parse parameters into booleans
+def bool_param(s):
+    """Return a bool indicating whether parameter 's' was True or False,
+    supporting a few different types for 's'."""
+    try:
+        ss = s.lower()
+        if ss in ["0", "false", "f", "no", "n"]:
+            return False
+        if ss in ["1", "true", "t", "yes", "y"]:
+            return True
+    except Exception:
+        return bool(s)
+    raise cherrypy.HTTPError("400 Bad Request",
+                             "can't parse parameter: " + ss)
+
+
+# Decorators
+def chunked_response(func):
+    """Decorator to enable chunked responses."""
+    # Set this to False to get better tracebacks from some requests
+    # (/stream/extract, /stream/intervals).
+    func._cp_config = {'response.stream': True}
+    return func
+
+
+def response_type(content_type):
+    """Return a decorator-generating function that sets the
+    response type to the specified string."""
+    def wrapper(func, *args, **kwargs):
+        cherrypy.response.headers['Content-Type'] = content_type
+        return func(*args, **kwargs)
+    return decorator.decorator(wrapper)
+
+
+def exception_to_httperror(*expected):
+    """Return a decorator-generating function that catches expected
+    errors and throws a HTTPError describing it instead.
+
+        @exception_to_httperror(NilmDBError, ValueError)
+        def foo():
+            pass
+    """
+    def wrapper(func, *args, **kwargs):
+        exc_info = None
+        try:
+            return func(*args, **kwargs)
+        except expected:
+            # Re-raise it, but maintain the original traceback
+            exc_info = sys.exc_info()
+            new_exc = cherrypy.HTTPError("400 Bad Request", str(exc_info[1]))
+            raise new_exc.with_traceback(exc_info[2])
+        finally:
+            del exc_info
+    # We need to preserve the function's argspecs for CherryPy to
+    # handle argument errors correctly.  Decorator.decorator takes
+    # care of that.
+    return decorator.decorator(wrapper)
+
+
+# Custom CherryPy tools
+def CORS_allow(methods):
+    """This does several things:
+
+    Handles CORS preflight requests.
+    Adds Allow: header to all requests.
+    Raise 405 if request.method not in method.
+
+    It is similar to cherrypy.tools.allow, with the CORS stuff added.
+
+    Add this to CherryPy with:
+    cherrypy.tools.CORS_allow = cherrypy.Tool('on_start_resource', CORS_allow)
+    """
+    request = cherrypy.request.headers
+    response = cherrypy.response.headers
+
+    if not isinstance(methods, (tuple, list)):
+        methods = [methods]
+    methods = [m.upper() for m in methods if m]
+    if not methods:
+        methods = ['GET', 'HEAD']
+    elif 'GET' in methods and 'HEAD' not in methods:
+        methods.append('HEAD')
+    response['Allow'] = ', '.join(methods)
+
+    # Allow all origins
+    if 'Origin' in request:
+        response['Access-Control-Allow-Origin'] = request['Origin']
+
+    # If it's a CORS request, send response.
+    request_method = request.get("Access-Control-Request-Method", None)
+    request_headers = request.get("Access-Control-Request-Headers", None)
+    if (cherrypy.request.method == "OPTIONS" and
+            request_method and request_headers):
+        response['Access-Control-Allow-Headers'] = request_headers
+        response['Access-Control-Allow-Methods'] = ', '.join(methods)
+        # Try to stop further processing and return a 200 OK
+        cherrypy.response.status = "200 OK"
+        cherrypy.response.body = b""
+        cherrypy.request.handler = lambda: ""
+        return
+
+    # Reject methods that were not explicitly allowed
+    if cherrypy.request.method not in methods:
+        raise cherrypy.HTTPError(405)
+
+
+# Helper for json_in tool to process JSON data into normal request
+# parameters.
+def json_to_request_params(body):
+    cherrypy.lib.jsontools.json_processor(body)
+    if not isinstance(cherrypy.request.json, dict):
+        raise cherrypy.HTTPError(415)
+    cherrypy.request.params.update(cherrypy.request.json)
+
+
+# Used as an "error_page.default" handler
+def json_error_page(status, message, traceback, version,
+                    force_traceback=False):
+    """Return a custom error page in JSON so the client can parse it"""
+    errordata = {"status": status,
+                 "message": message,
+                 "version": version,
+                 "traceback": traceback}
+    # Don't send a traceback if the error was 400-499 (client's fault)
+    code = int(status.split()[0])
+    if not force_traceback:
+        if 400 <= code <= 499:
+            errordata["traceback"] = ""
+    # Override the response type, which was previously set to text/html
+    cherrypy.serving.response.headers['Content-Type'] = (
+        "application/json;charset=utf-8")
+    # Undo the HTML escaping that cherrypy's get_error_page function applies
+    # (cherrypy issue 1135)
+    for k, v in errordata.items():
+        v = v.replace("&lt;", "<")
+        v = v.replace("&gt;", ">")
+        v = v.replace("&amp;", "&")
+        errordata[k] = v
+    return json.dumps(errordata, separators=(',', ':'))
+
+
+class CherryPyExit(SystemExit):
+    pass
+
+
+def cherrypy_patch_exit():
+    # Cherrypy stupidly calls os._exit(70) when it can't bind the port
+    # and exits.  Instead of that, raise a CherryPyExit (derived from
+    # SystemExit).  This exception may not make it back up to the caller
+    # due to internal thread use in the CherryPy engine, but there should
+    # be at least some indication that it happened.
+    bus = cherrypy.process.wspbus.bus
+    if "_patched_exit" in bus.__dict__:
+        return
+    bus._patched_exit = True
+
+    def patched_exit(orig):
+        real_exit = os._exit
+
+        def fake_exit(code):
+            raise CherryPyExit(code)
+        os._exit = fake_exit
+        try:
+            orig()
+        finally:
+            os._exit = real_exit
+    bus.exit = functools.partial(patched_exit, bus.exit)
+
+    # A behavior change in Python 3.8 means that some thread exceptions,
+    # derived from SystemExit, now print tracebacks where they didn't
+    # used to: https://bugs.python.org/issue1230540
+    # Install a thread exception hook that ignores CherryPyExit;
+    # to make this match the behavior where we didn't set
+    # threading.excepthook, we also need to ignore SystemExit.
+    def hook(args):
+        if args.exc_type == CherryPyExit or args.exc_type == SystemExit:
+            return
+        sys.excepthook(args.exc_type, args.exc_value,
+                       args.exc_traceback)  # pragma: no cover
+    threading.excepthook = hook
+
+
+# Start/stop CherryPy standalone server
+def cherrypy_start(blocking=False, event=False):
+    """Start the CherryPy server, handling errors and signals
+    somewhat gracefully."""
+
+    cherrypy_patch_exit()
+
+    # Start the server
+    cherrypy.engine.start()
+
+    # Signal that the engine has started successfully
+    if event is not None:
+        event.set()
+
+    if blocking:
+        try:
+            cherrypy.engine.wait(cherrypy.engine.states.EXITING,
+                                 interval=0.1, channel='main')
+        except (KeyboardInterrupt, IOError):
+            cherrypy.engine.log('Keyboard Interrupt: shutting down')
+            cherrypy.engine.exit()
+        except SystemExit:
+            cherrypy.engine.log('SystemExit raised: shutting down')
+            cherrypy.engine.exit()
+            raise
+
+
+# Stop CherryPy server
+def cherrypy_stop():
+    cherrypy.engine.exit()

nilmdb/test_fileinterval.py

@@ -1,46 +0,0 @@
-from nilmdb import Interval, IntervalSet, IntervalError, FileInterval
-from datetime import datetime
-from nose.tools import assert_raises
-
-from test_interval import iset
-
-def fiset(string):
-    """Like iset, but builds with FileIntervals instead of Intervals"""
-    iset = IntervalSet()
-    for i, c in enumerate(string):
-        day = datetime.strptime("{0:04d}".format(i+2000), "%Y")
-        if (c == "["):
-            start = day
-        elif (c == "|"):
-            iset += FileInterval(start, day, "test.dat")
-            start = day
-        elif (c == "]"):
-            iset += FileInterval(start, day, "test.dat")
-            del start
-    return iset
-
-def test_fileinterval_vs_interval():
-    """Test FileInterval/Interval inheritance"""
-
-    i = iset("[--]")
-    f = fiset("[--]")
-
-    # check types
-    assert(isinstance(i[0], Interval))
-    assert(not isinstance(i[0], FileInterval))
-    assert(isinstance(f[0], Interval))
-    assert(isinstance(f[0], FileInterval))
-
-    # when doing an intersection, result should be a subset of the first arg
-    u = (i & f)
-    assert(isinstance(u[0], Interval))
-    assert(not isinstance(u[0], FileInterval))
-    u = (f & i)
-    assert(isinstance(u[0], Interval))
-    assert(isinstance(u[0], FileInterval))
-
-    # they're still the same though
-    assert(i == f == u)
-
-    # just for coverage
-    assert_raises(IntervalError, fiset("[]")[0].subset, f[0].start, f[0].end)

nilmdb/test_interval.py

@@ -1,189 +0,0 @@
-from nilmdb import Interval, IntervalSet, IntervalError
-from datetime import datetime
-from nose.tools import assert_raises
-import itertools 
-
-def test_interval():
-    """Test the Interval class"""
-    d1 = datetime.strptime("19801205","%Y%m%d")
-    d2 = datetime.strptime("19900216","%Y%m%d")
-    d3 = datetime.strptime("20111205","%Y%m%d")
-
-    # basic construction
-    i = Interval(d1, d1)
-    i = Interval(d1, d3)
-    assert(i.start == d1)
-    assert(i.end == d3)
-
-    # assignment should work
-    i.start = d2
-    try:
-        i.end = d1
-        raise Exception("should have died there")
-    except IntervalError:
-        pass
-    i.start = d1
-    i.end = d2
-
-    # end before start
-    assert_raises(IntervalError, Interval, d3, d1)
-
-    # wrong type
-    assert_raises(IntervalError, Interval, 1, 2)
-
-    # compare
-    assert(Interval(d1, d2) == Interval(d1, d2))
-    assert(Interval(d1, d2) < Interval(d1, d3))
-    assert(Interval(d1, d3) > Interval(d1, d2))
-    assert(Interval(d1, d2) < Interval(d2, d3))
-    assert(Interval(d1, d3) < Interval(d2, d3))
-    assert(Interval(d2, d2) > Interval(d1, d3))
-    assert(Interval(d3, d3) == Interval(d3, d3))
-    assert_raises(TypeError, cmp, i, 123)
-
-    # subset
-    assert(Interval(d1, d3).subset(d1, d2) == Interval(d1, d2))
-    assert_raises(IntervalError, Interval(d2, d3).subset, d1, d2)
-
-    # append
-    assert(Interval(d1, d2).is_adjacent(Interval(d2,d3)))
-    assert(Interval(d2, d3).is_adjacent(Interval(d1,d2)))
-    assert(not Interval(d2, d3).is_adjacent(Interval(d1,d3)))
-    assert_raises(TypeError, Interval(d1, d2).is_adjacent, 1)
-
-    # misc
-    assert(repr(i) == repr(eval(repr(i).replace("datetime.",""))))
-    assert(str(i) == "[1980-12-05 00:00:00 -> 1990-02-16 00:00:00]")
-
-def test_interval_intersect():
-    """Test Interval intersections"""
-    dates = [ datetime.strptime(year, "%y") for year in [ "00", "01", "02", "03" ] ]
-    perm = list(itertools.permutations(dates, 2))
-    prod = list(itertools.product(perm, perm))
-    should_intersect = {
-        False: [4, 5, 8, 20, 48, 56, 60, 96, 97, 100],
-        True: [0, 1, 2, 12, 13, 14, 16, 17, 24, 25, 26, 28, 29,
-               32, 49, 50, 52, 53, 61, 62, 64, 65, 68, 98, 101, 104]}
-    for i,((a,b),(c,d)) in enumerate(prod):
-        try:
-            i1 = Interval(a, b)
-            i2 = Interval(c, d)
-            assert(i1.intersects(i2) == i2.intersects(i1))
-            assert(i in should_intersect[i1.intersects(i2)])
-        except IntervalError:
-            assert(i not in should_intersect[True] and
-                   i not in should_intersect[False])
-    assert_raises(TypeError, i1.intersects, 1234)
-
-def test_intervalset_construct():
-    """Test interval set construction"""
-    dates = [ datetime.strptime(year, "%y") for year in [ "00", "01", "02", "03" ]]
-
-    a = Interval(dates[0], dates[1])
-    b = Interval(dates[1], dates[2])
-    c = Interval(dates[0], dates[2])
-    d = Interval(dates[2], dates[3])
-
-    iseta = IntervalSet(a)
-    isetb = IntervalSet([a, b])
-    isetc = IntervalSet([a])
-    assert(iseta != isetb)
-    assert(iseta == isetc)
-    assert(iseta != 3)
-    assert(IntervalSet(a) != IntervalSet(b))
-
-    print iseta == None
-    assert_raises(TypeError, cmp, iseta, isetb)
-    assert_raises(IntervalError, IntervalSet, [a, b, c])
-    assert_raises(TypeError, IntervalSet, [1, 2])
-
-    iset = IntervalSet(isetb)   # test iterator
-    assert(iset == isetb)
-    assert(len(iset) == 2)
-    assert(len(IntervalSet()) == 0)
-
-    # Test adding
-    iset = IntervalSet(a)
-    iset += IntervalSet(b)
-    assert(iset == IntervalSet([a, b]))
-    iset = IntervalSet(a)
-    iset += b
-    assert(iset == IntervalSet([a, b]))
-    iset = IntervalSet(a) + IntervalSet(b)
-    assert(iset == IntervalSet([a, b]))
-    iset = IntervalSet(b) + a
-    assert(iset == IntervalSet([a, b]))
-
-    # A set consisting of [0-1],[1-2] should match a set consisting of [0-2]
-    assert(IntervalSet([a,b]) == IntervalSet([c]))
-    # Etc
-    assert(IntervalSet([a,d]) != IntervalSet([c]))
-    assert(IntervalSet([c]) != IntervalSet([a,d]))
-    assert(IntervalSet([c,d]) != IntervalSet([b,d]))
-    
-    # misc
-    assert(repr(iset) == repr(eval(repr(iset).replace("datetime.",""))))
-
-def iset(string):
-    """Build an IntervalSet from a string, for testing purposes
-
-    Each character is a year
-    [ = interval start
-    | = interval end + adjacent start
-    ] = interval end
-    anything else is ignored
-    """
-    iset = IntervalSet()
-    for i, c in enumerate(string):
-        day = datetime.strptime("{0:04d}".format(i+2000), "%Y")
-        if (c == "["):
-            start = day
-        elif (c == "|"):
-            iset += Interval(start, day)
-            start = day
-        elif (c == "]"):
-            iset += Interval(start, day)
-            del start
-    return iset
-
-def test_intervalset_iset():
-    """Test basic iset construction"""
-    assert(iset("  [----]   ") ==
-           iset("  [-|--]   "))
-
-    assert(iset("[]  [--]   ") +
-           iset(" []    [--]") ==
-           iset("[|] [-----]"))
-
-def test_intervalset_intsersect():
-    """Test intersection (&)"""
-    assert_raises(TypeError, iset("[--]").__and__, 1234)
-    
-    assert(iset("[---------]") &
-           iset(" [---]     ") ==
-           iset(" [---]     "))
-
-    assert(iset(" [---]     ") &
-           iset("[---------]") ==
-           iset(" [---]     "))
-
-    assert(iset("    [-----]") &
-           iset(" [-----]   ") ==
-           iset("    [--]   "))
-
-    assert(iset("      [---]") &
-           iset(" [--]      ") ==
-           iset("           "))
-
-    assert(iset("    [-|---]") &
-           iset(" [-----|-] ") ==
-           iset("    [----] "))
-
-    assert(iset("    [-|-]  ") &
-           iset(" [-|--|--] ") ==
-           iset("    [---]  "))
-
-    assert(iset(" [----][--]") &
-           iset("[-] [--] []") ==
-           iset(" [] [-]  []"))
-

nilmdb/utils/__init__.py

@@ -0,0 +1,16 @@
+"""NilmDB utilities"""
+
+
+from nilmdb.utils.timer import Timer
+from nilmdb.utils.serializer import serializer_proxy
+from nilmdb.utils.lrucache import lru_cache
+from nilmdb.utils.diskusage import du, human_size
+from nilmdb.utils.mustclose import must_close
+from nilmdb.utils import atomic
+import nilmdb.utils.threadsafety
+import nilmdb.utils.fallocate
+import nilmdb.utils.time
+import nilmdb.utils.iterator
+import nilmdb.utils.interval
+import nilmdb.utils.lock
+import nilmdb.utils.sort

nilmdb/utils/atomic.py

@@ -0,0 +1,19 @@
+# Atomic file writing helper.
+
+import os
+
+
+def replace_file(filename, content):
+    """Attempt to atomically and durably replace the filename with the
+    given contents"""
+
+    newfilename = filename + b".new"
+
+    # Write to new file, flush it
+    with open(newfilename, "wb") as f:
+        f.write(content)
+        f.flush()
+        os.fsync(f.fileno())
+
+    # Move new file over old one
+    os.replace(newfilename, filename)

nilmdb/utils/diskusage.py

@@ -0,0 +1,36 @@
+import os
+import errno
+from math import log
+
+
+def human_size(num):
+    """Human friendly file size"""
+    unit_list = list(zip(['bytes', 'kiB', 'MiB', 'GiB', 'TiB'],
+                         [0, 0, 1, 2, 2]))
+    if num == 0:
+        return '0 bytes'
+    if num == 1:
+        return '1 byte'
+    exponent = min(int(log(num, 1024)), len(unit_list) - 1)
+    quotient = float(num) / 1024**exponent
+    unit, num_decimals = unit_list[exponent]
+    format_string = '{:.%sf} {}' % (num_decimals)
+    return format_string.format(quotient, unit)
+
+
+def du(path):
+    """Like du -sb, returns total size of path in bytes.  Ignore
+    errors that might occur if we encounter broken symlinks or
+    files in the process of being removed."""
+    try:
+        st = os.stat(path)
+        size = st.st_blocks * 512
+        if os.path.isdir(path):
+            for thisfile in os.listdir(path):
+                filepath = os.path.join(path, thisfile)
+                size += du(filepath)
+        return size
+    except OSError as e:
+        if e.errno != errno.ENOENT:
+            raise
+        return 0

nilmdb/utils/fallocate.py

@@ -0,0 +1,20 @@
+# Implementation of hole punching via fallocate, if the OS
+# and filesystem support it.
+
+import fallocate
+
+
+def punch_hole(filename, offset, length, ignore_errors=True):
+    """Punch a hole in the file.  This isn't well supported, so errors
+    are ignored by default."""
+    try:
+        with open(filename, "r+") as f:
+            fallocate.fallocate(
+                f.fileno(),
+                offset,
+                length,
+                fallocate.FALLOC_FL_KEEP_SIZE | fallocate.FALLOC_FL_PUNCH_HOLE)
+    except Exception:
+        if ignore_errors:
+            return
+        raise

nilmdb/utils/interval.py

@@ -0,0 +1,168 @@
+"""Interval.  Like nilmdb.server.interval, but re-implemented here
+in plain Python so clients have easier access to it, and with a few
+helper functions.
+
+Intervals are half-open, ie. they include data points with timestamps
+[start, end)
+"""
+
+import nilmdb.utils.time
+import nilmdb.utils.iterator
+
+
+class IntervalError(Exception):
+    """Error due to interval overlap, etc"""
+    pass
+
+
+# Interval
+class Interval:
+    """Represents an interval of time."""
+
+    def __init__(self, start, end):
+        """
+        'start' and 'end' are arbitrary numbers that represent time
+        """
+        if start >= end:
+            # Explicitly disallow zero-width intervals, since they're half-open
+            raise IntervalError("start %s must precede end %s" % (start, end))
+        self.start = start
+        self.end = end
+
+    def __repr__(self):
+        s = repr(self.start) + ", " + repr(self.end)
+        return self.__class__.__name__ + "(" + s + ")"
+
+    def __str__(self):
+        return ("[" + nilmdb.utils.time.timestamp_to_string(self.start) +
+                " -> " + nilmdb.utils.time.timestamp_to_string(self.end) + ")")
+
+    def human_string(self):
+        return ("[ " + nilmdb.utils.time.timestamp_to_human(self.start) +
+                " -> " + nilmdb.utils.time.timestamp_to_human(self.end) + " ]")
+
+    # Compare two intervals.  If non-equal, order by start then end
+    def __lt__(self, other):
+        return (self.start, self.end) < (other.start, other.end)
+
+    def __gt__(self, other):
+        return (self.start, self.end) > (other.start, other.end)
+
+    def __le__(self, other):
+        return (self.start, self.end) <= (other.start, other.end)
+
+    def __ge__(self, other):
+        return (self.start, self.end) >= (other.start, other.end)
+
+    def __eq__(self, other):
+        return (self.start, self.end) == (other.start, other.end)
+
+    def __ne__(self, other):
+        return (self.start, self.end) != (other.start, other.end)
+
+    def intersects(self, other):
+        """Return True if two Interval objects intersect"""
+        if not isinstance(other, Interval):
+            raise TypeError("need an Interval")
+        if self.end <= other.start or self.start >= other.end:
+            return False
+        return True
+
+    def subset(self, start, end):
+        """Return a new Interval that is a subset of this one"""
+        # A subclass that tracks additional data might override this.
+        if start < self.start or end > self.end:
+            raise IntervalError("not a subset")
+        return Interval(start, end)
+
+
+def _interval_math_helper(a, b, op, subset=True):
+    """Helper for set_difference, intersection functions,
+    to compute interval subsets based on a math operator on ranges
+    present in A and B.  Subsets are computed from A, or new intervals
+    are generated if subset = False."""
+    # Iterate through all starts and ends in sorted order.  Add a
+    # tag to the iterator so that we can figure out which one they
+    # were, after sorting.
+    def decorate(it, key_start, key_end):
+        for i in it:
+            yield i.start, key_start, i
+            yield i.end, key_end, i
+    a_iter = decorate(iter(a), 0, 2)
+    b_iter = decorate(iter(b), 1, 3)
+
+    # Now iterate over the timestamps of each start and end.
+    # At each point, evaluate which type of end it is, to determine
+    # how to build up the output intervals.
+    a_interval = None
+    in_a = False
+    in_b = False
+    out_start = None
+    for (ts, k, i) in nilmdb.utils.iterator.imerge(a_iter, b_iter):
+        if k == 0:
+            a_interval = i
+            in_a = True
+        elif k == 1:
+            in_b = True
+        elif k == 2:
+            in_a = False
+        else:  # k == 3
+            in_b = False
+        include = op(in_a, in_b)
+        if include and out_start is None:
+            out_start = ts
+        elif not include:
+            if out_start is not None and out_start != ts:
+                if subset:
+                    yield a_interval.subset(out_start, ts)
+                else:
+                    yield Interval(out_start, ts)
+            out_start = None
+
+
+def set_difference(a, b):
+    """
+    Compute the difference (a \\ b) between the intervals in 'a' and
+    the intervals in 'b'; i.e., the ranges that are present in 'self'
+    but not 'other'.
+
+    'a' and 'b' must both be iterables.
+
+    Returns a generator that yields each interval in turn.
+    Output intervals are built as subsets of the intervals in the
+    first argument (a).
+    """
+    return _interval_math_helper(a, b, (lambda a, b: a and not b))
+
+
+def intersection(a, b):
+    """
+    Compute the intersection between the intervals in 'a' and the
+    intervals in 'b'; i.e., the ranges that are present in both 'a'
+    and 'b'.
+
+    'a' and 'b' must both be iterables.
+
+    Returns a generator that yields each interval in turn.
+    Output intervals are built as subsets of the intervals in the
+    first argument (a).
+    """
+    return _interval_math_helper(a, b, (lambda a, b: a and b))
+
+
+def optimize(it):
+    """
+    Given an iterable 'it' with intervals, optimize them by joining
+    together intervals that are adjacent in time, and return a generator
+    that yields the new intervals.
+    """
+    saved_int = None
+    for interval in it:
+        if saved_int is not None:
+            if saved_int.end == interval.start:
+                interval.start = saved_int.start
+            else:
+                yield saved_int
+        saved_int = interval
+    if saved_int is not None:
+        yield saved_int

nilmdb/utils/iterator.py

@@ -0,0 +1,38 @@
+# Misc iterator tools
+
+# Iterator merging, based on http://code.activestate.com/recipes/491285/
+import heapq
+
+
+def imerge(*iterables):
+    '''Merge multiple sorted inputs into a single sorted output.
+
+    Equivalent to:  sorted(itertools.chain(*iterables))
+
+    >>> list(imerge([1,3,5,7], [0,2,4,8], [5,10,15,20], [], [25]))
+    [0, 1, 2, 3, 4, 5, 5, 7, 8, 10, 15, 20, 25]
+
+    '''
+    heappop, siftup, _Stop = heapq.heappop, heapq._siftup, StopIteration
+
+    h = []
+    h_append = h.append
+    for it in map(iter, iterables):
+        try:
+            nexter = it.__next__
+            h_append([nexter(), nexter])
+        except _Stop:
+            pass
+    heapq.heapify(h)
+
+    while 1:
+        try:
+            while 1:
+                v, nexter = s = h[0]    # raises IndexError when h is empty
+                yield v
+                s[0] = nexter()         # raises StopIteration when exhausted
+                siftup(h, 0)            # restore heap condition
+        except _Stop:
+            heappop(h)                  # remove empty iterator
+        except IndexError:
+            return

nilmdb/utils/lock.py

@@ -0,0 +1,22 @@
+# File locking
+
+import fcntl
+import errno
+
+
+def exclusive_lock(f):
+    """Acquire an exclusive lock.  Returns True on successful
+    lock, or False on error."""
+    try:
+        fcntl.flock(f.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB)
+    except IOError as e:
+        if e.errno in (errno.EACCES, errno.EAGAIN):
+            return False
+        else:
+            raise
+    return True
+
+
+def exclusive_unlock(f):
+    """Release an exclusive lock."""
+    fcntl.flock(f.fileno(), fcntl.LOCK_UN)

nilmdb/utils/lrucache.py

@@ -0,0 +1,80 @@
+# Memoize a function's return value with a least-recently-used cache
+# Based on:
+#   http://code.activestate.com/recipes/498245-lru-and-lfu-cache-decorators/
+# with added 'destructor' functionality.
+
+import collections
+import decorator
+
+
+def lru_cache(size=10, onremove=None, keys=slice(None)):
+    """Least-recently-used cache decorator.
+
+    @lru_cache(size=10, onremove=None)
+    def f(...):
+        pass
+
+    Given a function and arguments, memoize its return value.  Up to
+    'size' elements are cached.  'keys' is a slice object that
+    represents which arguments are used as the cache key.
+
+    When evicting a value from the cache, call the function
+    'onremove' with the value that's being evicted.
+
+    Call f.cache_remove(...) to evict the cache entry with the given
+    arguments.  Call f.cache_remove_all() to evict all entries.
+    f.cache_hits and f.cache_misses give statistics.
+    """
+
+    def decorate(func):
+        cache = collections.OrderedDict()  # order: least- to most-recent
+
+        def evict(value):
+            if onremove:
+                onremove(value)
+
+        def wrapper(orig, *args, **kwargs):
+            if kwargs:
+                raise NotImplementedError("kwargs not supported")
+            key = args[keys]
+            try:
+                value = cache.pop(key)
+                orig.cache_hits += 1
+            except KeyError:
+                value = orig(*args)
+                orig.cache_misses += 1
+                if len(cache) >= size:
+                    evict(cache.popitem(0)[1])  # evict LRU cache entry
+            cache[key] = value                  # (re-)insert this key at end
+            return value
+
+        def cache_remove(*args):
+            """Remove the described key from this cache, if present."""
+            key = args
+            if key in cache:
+                evict(cache.pop(key))
+            else:
+                if cache:
+                    if len(args) != len(next(iter(cache.keys()))):
+                        raise KeyError("trying to remove from LRU cache, but "
+                                       "number of arguments doesn't match the "
+                                       "cache key length")
+
+        def cache_remove_all():
+            nonlocal cache
+            for key in cache:
+                evict(cache[key])
+            cache = collections.OrderedDict()
+
+        def cache_info():
+            return (func.cache_hits, func.cache_misses)
+
+        new = decorator.decorator(wrapper, func)
+        func.cache_hits = 0
+        func.cache_misses = 0
+        new.cache_info = cache_info
+        new.cache_remove = cache_remove
+        new.cache_remove_all = cache_remove_all
+        return new
+
+    return decorate

nilmdb/utils/mustclose.py

@@ -0,0 +1,71 @@
+import sys
+import inspect
+import decorator
+from nilmdb.utils.printf import fprintf
+
+
+def must_close(errorfile=sys.stderr, wrap_verify=False):
+    """Class decorator that warns on 'errorfile' at deletion time if
+    the class's close() member wasn't called.
+
+    If 'wrap_verify' is True, every class method is wrapped with a
+    verifier that will raise AssertionError if the .close() method has
+    already been called."""
+    def class_decorator(cls):
+
+        def is_method_or_function(x):
+            return inspect.ismethod(x) or inspect.isfunction(x)
+
+        def wrap_class_method(wrapper):
+            try:
+                orig = getattr(cls, wrapper.__name__)
+            except AttributeError:
+                orig = lambda x: None
+            if is_method_or_function(orig):
+                setattr(cls, wrapper.__name__,
+                        decorator.decorator(wrapper, orig))
+
+        @wrap_class_method
+        def __init__(orig, self, *args, **kwargs):
+            ret = orig(self, *args, **kwargs)
+            self.__dict__["_must_close"] = True
+            self.__dict__["_must_close_initialized"] = True
+            return ret
+
+        @wrap_class_method
+        def __del__(orig, self, *args, **kwargs):
+            try:
+                if "_must_close" in self.__dict__:
+                    fprintf(errorfile, "error: %s.close() wasn't called!\n",
+                            self.__class__.__name__)
+                return orig(self, *args, **kwargs)
+            except:
+                pass
+
+        @wrap_class_method
+        def close(orig, self, *args, **kwargs):
+            if "_must_close" in self.__dict__:
+                del self._must_close
+            return orig(self, *args, **kwargs)
+
+        # Optionally wrap all other functions
+        def verifier(orig, self, *args, **kwargs):
+            if ("_must_close" not in self.__dict__ and
+                    "_must_close_initialized" in self.__dict__):
+                raise AssertionError("called " + str(orig) + " after close")
+            return orig(self, *args, **kwargs)
+        if wrap_verify:
+            for (name, method) in inspect.getmembers(cls,
+                                                     is_method_or_function):
+                # Skip some methods
+                if name in ["__del__", "__init__"]:
+                    continue
+                # Set up wrapper
+                if inspect.ismethod(method):
+                    func = method.__func__
+                else:
+                    func = method
+                setattr(cls, name, decorator.decorator(verifier, func))
+
+        return cls
+    return class_decorator

nilmdb/utils/printf.py

@@ -0,0 +1,13 @@
+"""printf, fprintf, sprintf"""
+
+
+def printf(_str, *args):
+    print(_str % args, end='')
+
+
+def fprintf(_file, _str, *args):
+    print(_str % args, end='', file=_file)
+
+
+def sprintf(_str, *args):
+    return (_str % args)

nilmdb/utils/serializer.py

@@ -0,0 +1,134 @@
+import queue
+import threading
+import sys
+
+# This file provides a class that will wrap an object and serialize
+# all calls to its methods.  All calls to that object will be queued
+# and executed from a single thread, regardless of which thread makes
+# the call.
+
+# Based partially on http://stackoverflow.com/questions/2642515/
+
+
+class SerializerThread(threading.Thread):
+    """Thread that retrieves call information from the queue, makes the
+    call, and returns the results."""
+    def __init__(self, classname, call_queue):
+        threading.Thread.__init__(self)
+        self.name = "Serializer-" + classname + "-" + self.name
+        self.call_queue = call_queue
+
+    def run(self):
+        while True:
+            result_queue, func, args, kwargs = self.call_queue.get()
+            # Terminate if result_queue is None
+            if result_queue is None:
+                return
+            exception = None
+            result = None
+            try:
+                result = func(*args, **kwargs)  # wrapped
+            except:
+                exception = sys.exc_info()
+            # Ensure we delete these before returning a result, so
+            # we don't unncessarily hold onto a reference while
+            # we're waiting for the next call.
+            del func, args, kwargs
+            result_queue.put((exception, result))
+            del exception, result
+
+
+def serializer_proxy(obj_or_type):
+    """Wrap the given object or type in a SerializerObjectProxy.
+
+    Returns a SerializerObjectProxy object that proxies all method
+    calls to the object, as well as attribute retrievals.
+
+    The proxied requests, including instantiation, are performed in a
+    single thread and serialized between caller threads.
+    """
+    class SerializerCallProxy():
+        def __init__(self, call_queue, func, objectproxy):
+            self.call_queue = call_queue
+            self.func = func
+            # Need to hold a reference to object proxy so it doesn't
+            # go away (and kill the thread) until after get called.
+            self.objectproxy = objectproxy
+
+        def __call__(self, *args, **kwargs):
+            result_queue = queue.Queue()
+            self.call_queue.put((result_queue, self.func, args, kwargs))
+            (exc_info, result) = result_queue.get()
+            if exc_info is None:
+                return result
+            else:
+                raise exc_info[1].with_traceback(exc_info[2])
+
+    class SerializerObjectProxy():
+        def __init__(self, obj_or_type, *args, **kwargs):
+            self.__object = obj_or_type
+            if isinstance(obj_or_type, type):
+                classname = obj_or_type.__name__
+            else:
+                classname = obj_or_type.__class__.__name__
+            self.__call_queue = queue.Queue()
+            self.__thread = SerializerThread(classname, self.__call_queue)
+            self.__thread.daemon = True
+            self.__thread.start()
+            self._thread_safe = True
+
+        def __getattr__(self, key):
+            # If the attribute is a function, we want to return a
+            # proxy that will perform the call through the serializer
+            # when called.  Otherwise, we want to return the value
+            # directly.  This means we need to grab the attribute once,
+            # and therefore self.__object.__getattr__ may be called
+            # in an unsafe way, from the caller's thread.
+            attr = getattr(self.__object, key)
+            if not callable(attr):
+                # It's not callable, so perform the getattr from within
+                # the serializer thread, then return its value.
+                # That may differ from the "attr" value we just grabbed
+                # from here, due to forced ordering in the serializer.
+                getter = SerializerCallProxy(self.__call_queue, getattr, self)
+                return getter(self.__object, key)
+            else:
+                # It is callable, so return an object that will proxy through
+                # the serializer when called.
+                r = SerializerCallProxy(self.__call_queue, attr, self)
+                return r
+
+        # For an interable object, on __iter__(), save the object's
+        # iterator and return this proxy.  On next(), call the object's
+        # iterator through this proxy.
+        def __iter__(self):
+            attr = getattr(self.__object, "__iter__")
+            self.__iter = SerializerCallProxy(self.__call_queue, attr, self)()
+            return self
+
+        def __next__(self):
+            return SerializerCallProxy(self.__call_queue,
+                                       self.__iter.__next__, self)()
+
+        def __getitem__(self, key):
+            return self.__getattr__("__getitem__")(key)
+
+        def __call__(self, *args, **kwargs):
+            """Call this to instantiate the type, if a type was passed
+            to serializer_proxy.  Otherwise, pass the call through."""
+            ret = SerializerCallProxy(self.__call_queue,
+                                      self.__object, self)(*args, **kwargs)
+            if isinstance(self.__object, type):
+                # Instantiation
+                self.__object = ret
+                return self
+            return ret
+
+        def __del__(self):
+            try:
+                # Signal thread to exit, but don't wait for it.
+                self.__call_queue.put((None, None, None, None))
+            except:
+                pass
+
+    return SerializerObjectProxy(obj_or_type)

nilmdb/utils/sort.py

@@ -0,0 +1,19 @@
+import re
+
+
+def sort_human(items, key=None):
+    """Human-friendly sort (/stream/2 before /stream/10)"""
+    def to_num(val):
+        try:
+            return int(val)
+        except Exception:
+            return val
+
+    def human_key(text):
+        if key:
+            text = key(text)
+        # Break into character and numeric chunks.
+        chunks = re.split(r'([0-9]+)', text)
+        return [to_num(c) for c in chunks]
+
+    return sorted(items, key=human_key)

nilmdb/utils/threadsafety.py

@@ -0,0 +1,97 @@
+import threading
+from nilmdb.utils.printf import sprintf
+
+
+def verify_proxy(obj_or_type, check_thread=True,
+                 check_concurrent=True):
+    """Wrap the given object or type in a VerifyObjectProxy.
+
+    Returns a VerifyObjectProxy that proxies all method calls to the
+    given object, as well as attribute retrievals.
+
+    When calling methods, the following checks are performed.  On
+    failure, an exception is raised.
+
+    check_thread = True     # Fail if two different threads call methods.
+    check_concurrent = True # Fail if two functions are concurrently
+                            # run through this proxy
+    """
+    class Namespace():
+        pass
+
+    class VerifyCallProxy():
+        def __init__(self, func, parent_namespace):
+            self.func = func
+            self.parent_namespace = parent_namespace
+
+        def __call__(self, *args, **kwargs):
+            p = self.parent_namespace
+            this = threading.current_thread()
+            try:
+                callee = self.func.__name__
+            except AttributeError:
+                callee = "???"
+
+            if p.thread is None:
+                p.thread = this
+                p.thread_callee = callee
+
+            if check_thread and p.thread != this:
+                err = sprintf("unsafe threading: %s called %s.%s,"
+                              " but %s called %s.%s",
+                              p.thread.name, p.classname, p.thread_callee,
+                              this.name, p.classname, callee)
+                raise AssertionError(err)
+
+            need_concur_unlock = False
+            if check_concurrent:
+                if not p.concur_lock.acquire(False):
+                    err = sprintf("unsafe concurrency: %s called %s.%s "
+                                  "while %s is still in %s.%s",
+                                  this.name, p.classname, callee,
+                                  p.concur_tname, p.classname, p.concur_callee)
+                    raise AssertionError(err)
+                else:
+                    p.concur_tname = this.name
+                    p.concur_callee = callee
+                    need_concur_unlock = True
+
+            try:
+                ret = self.func(*args, **kwargs)
+            finally:
+                if need_concur_unlock:
+                    p.concur_lock.release()
+            return ret
+
+    class VerifyObjectProxy():
+        def __init__(self, obj_or_type, *args, **kwargs):
+            p = Namespace()
+            self.__ns = p
+            p.thread = None
+            p.thread_callee = None
+            p.concur_lock = threading.Lock()
+            p.concur_tname = None
+            p.concur_callee = None
+            self.__obj = obj_or_type
+            if isinstance(obj_or_type, type):
+                p.classname = self.__obj.__name__
+            else:
+                p.classname = self.__obj.__class__.__name__
+
+        def __getattr__(self, key):
+            attr = getattr(self.__obj, key)
+            if not callable(attr):
+                return VerifyCallProxy(getattr, self.__ns)(self.__obj, key)
+            return VerifyCallProxy(attr, self.__ns)
+
+        def __call__(self, *args, **kwargs):
+            """Call this to instantiate the type, if a type was passed
+            to verify_proxy.  Otherwise, pass the call through."""
+            ret = VerifyCallProxy(self.__obj, self.__ns)(*args, **kwargs)
+            if isinstance(self.__obj, type):
+                # Instantiation
+                self.__obj = ret
+                return self
+            return ret
+
+    return VerifyObjectProxy(obj_or_type)

nilmdb/utils/time.py

@@ -0,0 +1,148 @@
+import re
+import time
+import datetime_tz
+
+# Range
+min_timestamp = (-2**63)
+max_timestamp = (2**63 - 1)
+
+# Smallest representable step
+epsilon = 1
+
+
+def string_to_timestamp(string):
+    """Convert a string that represents an integer number of microseconds
+    since epoch."""
+    try:
+        # Parse a string like "1234567890123456" and return an integer
+        return int(string)
+    except ValueError:
+        # Try parsing as a float, in case it's "1234567890123456.0"
+        return int(round(float(string)))
+
+
+def timestamp_to_string(timestamp):
+    """Convert a timestamp (integer microseconds since epoch) to a string"""
+    if isinstance(timestamp, float):
+        return str(int(round(timestamp)))
+    else:
+        return str(timestamp)
+
+
+def timestamp_to_bytes(timestamp):
+    """Convert a timestamp (integer microseconds since epoch) to a Python
+    bytes object"""
+    return timestamp_to_string(timestamp).encode('utf-8')
+
+
+def timestamp_to_human(timestamp):
+    """Convert a timestamp (integer microseconds since epoch) to a
+    human-readable string, using the local timezone for display
+    (e.g. from the TZ env var)."""
+    if timestamp == min_timestamp:
+        return "(minimum)"
+    if timestamp == max_timestamp:
+        return "(maximum)"
+    dt = datetime_tz.datetime_tz.fromtimestamp(timestamp_to_unix(timestamp))
+    return dt.strftime("%a, %d %b %Y %H:%M:%S.%f %z")
+
+
+def unix_to_timestamp(unix):
+    """Convert a Unix timestamp (floating point seconds since epoch)
+    into a NILM timestamp (integer microseconds since epoch)"""
+    return int(round(unix * 1e6))
+
+
+def timestamp_to_unix(timestamp):
+    """Convert a NILM timestamp (integer microseconds since epoch)
+    into a Unix timestamp (floating point seconds since epoch)"""
+    return timestamp / 1e6
+
+
+seconds_to_timestamp = unix_to_timestamp
+timestamp_to_seconds = timestamp_to_unix
+
+
+def rate_to_period(hz, cycles=1):
+    """Convert a rate (in Hz) to a period (in timestamp units).
+    Returns an integer."""
+    period = unix_to_timestamp(cycles) / float(hz)
+    return int(round(period))
+
+
+def parse_time(toparse):
+    """
+    Parse a free-form time string and return a nilmdb timestamp
+    (integer microseconds since epoch).  If the string doesn't contain a
+    timestamp, the current local timezone is assumed (e.g. from the TZ
+    env var).
+    """
+    if toparse == "min":
+        return min_timestamp
+    if toparse == "max":
+        return max_timestamp
+
+    # If it starts with @, treat it as a NILM timestamp
+    # (integer microseconds since epoch)
+    try:
+        if toparse[0] == '@':
+            return int(toparse[1:])
+    except (ValueError, KeyError, IndexError):
+        pass
+
+    # If string isn't "now" and doesn't contain at least 4 digits,
+    # consider it invalid.  smartparse might otherwise accept
+    # empty strings and strings with just separators.
+    if toparse != "now" and len(re.findall(r"\d", toparse)) < 4:
+        raise ValueError("not enough digits for a timestamp")
+
+    # Try to just parse the time as given
+    try:
+        return unix_to_timestamp(datetime_tz.datetime_tz.
+                                 smartparse(toparse).totimestamp())
+    except (ValueError, OverflowError, TypeError):
+        pass
+
+    # If it's parseable as a float, treat it as a Unix or NILM
+    # timestamp based on its range.
+    try:
+        val = float(toparse)
+        # range is from about year 2001 - 2128
+        if 1e9 < val < 5e9:
+            return unix_to_timestamp(val)
+        if 1e15 < val < 5e15:
+            return val
+    except ValueError:
+        pass
+
+    # Try to extract a substring in a condensed format that we expect
+    # to see in a filename or header comment
+    res = re.search(r"(^|[^\d])("            # non-numeric or SOL
+                    r"(199\d|2\d\d\d)"       # year
+                    r"[-/]?"                 # separator
+                    r"(0[1-9]|1[012])"       # month
+                    r"[-/]?"                 # separator
+                    r"([012]\d|3[01])"       # day
+                    r"[-T ]?"                # separator
+                    r"([01]\d|2[0-3])"       # hour
+                    r"[:]?"                  # separator
+                    r"([0-5]\d)"             # minute
+                    r"[:]?"                  # separator
+                    r"([0-5]\d)?"            # second
+                    r"([-+]\d\d\d\d)?"       # timezone
+                    r")", toparse)
+    if res is not None:
+        try:
+            return unix_to_timestamp(datetime_tz.datetime_tz.
+                                     smartparse(res.group(2)).totimestamp())
+        except ValueError:
+            pass
+
+    # Could also try to successively parse substrings, but let's
+    # just give up for now.
+    raise ValueError("unable to parse timestamp")
+
+
+def now():
+    """Return current timestamp"""
+    return unix_to_timestamp(time.time())

nilmdb/utils/timer.py

@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+
+# Simple timer to time a block of code, for optimization debugging
+# use like:
+#   with nilmdb.utils.Timer("flush"):
+#       foo.flush()
+
+import contextlib
+import time
+
+
+@contextlib.contextmanager
+def Timer(name=None, tosyslog=False):
+    start = time.time()
+    yield
+    elapsed = int((time.time() - start) * 1000)
+    msg = (name or 'elapsed') + ": " + str(elapsed) + " ms"
+    if tosyslog:
+        import syslog
+        syslog.syslog(msg)
+    else:
+        print(msg)

nilmdb/utils/timestamper.py

@@ -0,0 +1,103 @@
+"""File-like objects that add timestamps to the input lines"""
+
+from nilmdb.utils.printf import sprintf
+import nilmdb.utils.time
+
+
+class Timestamper():
+    """A file-like object that adds timestamps to lines of an input file."""
+    def __init__(self, infile, ts_iter):
+        """file: filename, or another file-like object
+           ts_iter: iterator that returns a timestamp string for
+           each line of the file"""
+        if isinstance(infile, str):
+            self.file = open(infile, "rb")
+        else:
+            self.file = infile
+        self.ts_iter = ts_iter
+
+    def close(self):
+        self.file.close()
+
+    def readline(self, *args):
+        while True:
+            line = self.file.readline(*args)
+            if not line:
+                return b""
+            if line[0:1] == b'#':
+                continue
+            # For some reason, coverage on python 3.8 reports that
+            # we never hit this break, even though we definitely do.
+            break  # pragma: no cover
+        try:
+            return next(self.ts_iter) + line
+        except StopIteration:
+            return b""
+
+    def readlines(self, size=None):
+        out = b""
+        while True:
+            line = self.readline()
+            out += line
+            if not line or (size and len(out) >= size):
+                break
+        return out
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        result = self.readline()
+        if not result:
+            raise StopIteration
+        return result
+
+
+class TimestamperRate(Timestamper):
+    """Timestamper that uses a start time and a fixed rate"""
+    def __init__(self, infile, start, rate, end=None):
+        """
+        file: file name or object
+
+        start: Unix timestamp for the first value
+
+        rate: 1/rate is added to the timestamp for each line
+
+        end: If specified, raise StopIteration before outputting a value
+             greater than this."""
+        timestamp_to_bytes = nilmdb.utils.time.timestamp_to_bytes
+        rate_to_period = nilmdb.utils.time.rate_to_period
+
+        def iterator(start, rate, end):
+            n = 0
+            rate = float(rate)
+            while True:
+                now = start + rate_to_period(rate, n)
+                if end and now >= end:
+                    return
+                yield timestamp_to_bytes(now) + b" "
+                n += 1
+        Timestamper.__init__(self, infile, iterator(start, rate, end))
+        self.start = start
+        self.rate = rate
+
+    def __str__(self):
+        return sprintf("TimestamperRate(..., start=\"%s\", rate=%g)",
+                       nilmdb.utils.time.timestamp_to_human(self.start),
+                       self.rate)
+
+
+class TimestamperNow(Timestamper):
+    """Timestamper that uses current time"""
+    def __init__(self, infile):
+        timestamp_to_bytes = nilmdb.utils.time.timestamp_to_bytes
+        get_now = nilmdb.utils.time.now
+
+        def iterator():
+            while True:
+                yield timestamp_to_bytes(get_now()) + b" "
+
+        Timestamper.__init__(self, infile, iterator())
+
+    def __str__(self):
+        return "TimestamperNow(...)"

pytables-test/Makefile

@@ -1,5 +0,0 @@
-all:
-	time python test-indexed-read.py
-
-clean:
-	rm -f *pyc

pytables-test/README.jim

@@ -1,2 +0,0 @@
-New version from:
-  http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=641485#15

pytables-test/TODO.txt

@@ -1,12 +0,0 @@
-- Make CherryPy server that can handle simple GET/POST,
-  and a httplib client that can talk to that server.
-  Steps:
-  - Make server handle GET
-  - Make client send request, get response
-  - Add request streaming to server
-  - Add request streaming to client
-  - Make server handle POST
-  - Make client send request, get response
-  - Add request streaming to server
-  - Add request streaming to client
-  - Integrate into a server process that also keeps database open.

pytables-test/indexing-notes

@@ -1,3 +0,0 @@
-Indexing time64 doesn't seem to work -- needed to do "time >= 1243052015" even though the actual database times
-should be something like 1243052015.847000.  Let's switch to just using a 64-bit integer counting e.g.
-microseconds since 1970-01-01 

pytables-test/sample-query

@@ -1,3 +0,0 @@
-timestamp > 1243052015
-took 394.5 minutes in vitables
-(2340 rows matched)

pytables-test/server.py

@@ -1,53 +0,0 @@
-import sys
-import tables
-import nilmdb
-
-try:
-    import cherrypy
-    cherrypy.tools.json_out
-except:
-    sys.stderr.write("Cherrypy 3.2+ required\n")
-    sys.exit(1)
-
-class NilmApp:
-    def __init__(self, db):
-        self.db = db
-
-class Root(NilmApp):
-    """NILM Database"""
-
-    server_version = "1.0"
-
-    @cherrypy.expose
-    def index(self):
-        raise cherrypy.NotFound()
-
-    @cherrypy.expose
-    def favicon_ico(self):
-        raise cherrypy.NotFound()
-
-    @cherrypy.expose
-    @cherrypy.tools.json_out()
-    def version(self):
-        return self.server_version
-
-class Stream(NilmApp):
-    """Stream operations"""
-
-    @cherrypy.expose
-    @cherrypy.tools.json_out()
-    def list(self):
-        return 
-    
-cherrypy.config.update({
-    'server.socket_host': '127.0.0.1',
-    'server.socket_port': 12380
-    })
-
-db = nilmdb.nilmdb()
-cherrypy.tree.mount(Root(db), "/")
-cherrypy.tree.mount(Stream(db), "/stream")
-
-if __name__ == "__main__":
-    cherrypy.engine.start()
-    cherrypy.engine.block()

pytables-test/speed-pytables.py

@@ -1,16 +0,0 @@
-import tables
-import numpy
-
-class RawSample(tables.IsDescription):
-    timestamp = tables.UInt64Col()
-    voltage   = tables.UInt16Col(shape = 3)
-    current   = tables.UInt16Col(shape = 3)
-
-h5file = tables.openFile("test.h5", mode = "w", title = "Test")
-group = h5file.createGroup("/", "raw", "Raw Data")
-table = h5file.createTable(group, "nilm1", RawSample, "NILM 1")
-
-print repr(h5file)
-
-# write rows
-

pytables-test/test-indexing.py

@@ -1,54 +0,0 @@
-#!/usr/bin/python
-
-from tables import *
-import re
-import time
-
-# A class to describe our data
-class PrepData(IsDescription):
-    timestamp = Int64Col()
-    p1 = Float32Col()
-    q1 = Float32Col()
-    p3 = Float32Col()
-    q3 = Float32Col()
-    p5 = Float32Col()
-    q5 = Float32Col()
-    p7 = Float32Col()
-    q7 = Float32Col()
-
-filename = "test.h5"
-h5file = openFile(filename, mode = "w", title = "NILM Test")
-
-group = h5file.createGroup("/", "newton", "Newton school")
-table = h5file.createTable(group, "prep", PrepData, "Prep Data", expectedrows = 120 * 86400 * 90)
-
-table.cols.timestamp.createIndex()
-
-for i in range(0, 80):
-    # Open file
-    data = open("data/alldata")
-    count = 0
-    oldtime = time.time()
-    prep = table.row
-    for line in data:
-        count = count + 1
-        if count % 1000000 == 0:
-            print str(i) + ": " + str((time.time() - oldtime)) + ", total " + str(count/1000000) + "m lines"
-            oldtime = time.time()
-        v = re.split('\s+', line)
-        prep['timestamp'] = int(v[0]) + 500000000 * i
-        prep['p1'] = v[1]
-        prep['q1'] = v[2]
-        prep['p3'] = v[3]
-        prep['q3'] = v[4]
-        prep['p5'] = v[5]
-        prep['q5'] = v[6]
-        prep['p7'] = v[7]
-        prep['q7'] = v[8]
-        prep.append()
-    data.close()
-
-h5file.close()
-
-
-

pytables-test/test-write.py

@@ -1,54 +0,0 @@
-#!/usr/bin/python
-
-from tables import *
-import re
-import time
-
-# A class to describe our data
-class PrepData(IsDescription):
-    timestamp = Time64Col()
-    p1 = Float32Col()
-    q1 = Float32Col()
-    p3 = Float32Col()
-    q3 = Float32Col()
-    p5 = Float32Col()
-    q5 = Float32Col()
-    p7 = Float32Col()
-    q7 = Float32Col()
-
-filename = "test.h5"
-h5file = openFile(filename, mode = "w", title = "NILM Test")
-
-group = h5file.createGroup("/", "newton", "Newton school")
-table = h5file.createTable(group, "prep", PrepData, "Prep Data")
-
-table.cols.timestamp.createIndex()
-
-for i in range(0, 80):
-    # Open file
-    data = open("data/alldata")
-    count = 0
-    oldtime = time.time()
-    prep = table.row
-    for line in data:
-        count = count + 1
-        if count % 1000000 == 0:
-            print str(i) + ": " + str((time.time() - oldtime)) + ", total " + str(count/1000000) + "m lines"
-            oldtime = time.time()
-        v = re.split('\s+', line)
-        prep['timestamp'] = float(v[0]) / 1000.0 + 500000 * i
-        prep['p1'] = v[1]
-        prep['q1'] = v[2]
-        prep['p3'] = v[3]
-        prep['q3'] = v[4]
-        prep['p5'] = v[5]
-        prep['q5'] = v[6]
-        prep['p7'] = v[7]
-        prep['q7'] = v[8]
-        prep.append()
-    data.close()
-
-h5file.close()
-
-
-

requirements.txt

@@ -0,0 +1,41 @@
+argcomplete==1.12.0
+CherryPy==18.6.0
+coverage==5.2.1
+Cython==0.29.21
+decorator==4.4.2
+fallocate==1.6.4
+flake8==3.8.3
+nose==1.3.7
+numpy==1.19.1
+progressbar==2.5
+psutil==5.7.2
+python-datetime-tz==0.5.4
+python-dateutil==2.8.1
+requests==2.24.0
+tz==0.2.2
+yappi==1.2.5
+
+## The following requirements were added by pip freeze:
+beautifulsoup4==4.9.1
+certifi==2020.6.20
+chardet==3.0.4
+cheroot==8.4.2
+idna==2.10
+jaraco.classes==3.1.0
+jaraco.collections==3.0.0
+jaraco.functools==3.0.1
+jaraco.text==3.2.0
+mccabe==0.6.1
+more-itertools==8.4.0
+portend==2.6
+pycodestyle==2.6.0
+pyflakes==2.2.0
+pytz==2020.1
+six==1.15.0
+soupsieve==2.0.1
+tempora==4.0.0
+urllib3==1.25.10
+waitress==1.4.4
+WebOb==1.8.6
+WebTest==2.0.35
+zc.lockfile==2.0

setup.cfg

@@ -1,9 +1,59 @@
+[aliases]
+test = nosetests
+
 [nosetests]
+# Note: values must be set to 1, and have no comments on the same line,
+# for "python setup.py nosetests" to work correctly.
 nocapture=1
+# Comment this out to see CherryPy logs on failure:
+nologcapture=1
 with-coverage=1
 cover-inclusive=1
 cover-package=nilmdb
 cover-erase=1
+# this works, puts html output in cover/ dir:
+# cover-html=1
+#debug=nose
+#debug-log=nose.log
 stop=1
 verbosity=2
+tests=tests
+#tests=tests/test_threadsafety.py
+#tests=tests/test_bulkdata.py
+#tests=tests/test_mustclose.py
+#tests=tests/test_lrucache.py
+#tests=tests/test_cmdline.py
+#tests=tests/test_layout.py
+#tests=tests/test_rbtree.py
+#tests=tests/test_interval.py
+#tests=tests/test_rbtree.py,tests/test_interval.py
+#tests=tests/test_interval.py
+#tests=tests/test_client.py
+#tests=tests/test_timestamper.py
+#tests=tests/test_serializer.py
+#tests=tests/test_iteratorizer.py
+#tests=tests/test_client.py:TestClient.test_client_nilmdb
+#tests=tests/test_nilmdb.py
+#with-profile=1
+#profile-sort=time
+##profile-restrict=10  # doesn't work right, treated as string or something
 
+[versioneer]
+VCS=git
+style=pep440
+versionfile_source=nilmdb/_version.py
+versionfile_build=nilmdb/_version.py
+tag_prefix=nilmdb-
+parentdir_prefix=nilmdb-
+
+[flake8]
+exclude=_version.py
+extend-ignore=E731
+per-file-ignores=__init__.py:F401,E402 \
+        serializer.py:E722 \
+        mustclose.py:E722 \
+        fsck.py:E266
+
+[pylint]
+ignore=_version.py
+disable=C0103,C0111,R0913,R0914

setup.py

@@ -1,9 +1,64 @@
-#!/usr/bin/python
+#!/usr/bin/env python3
 
-from distutils.core import setup
+# To release a new version, tag it:
+#   git tag -a nilmdb-1.1 -m "Version 1.1"
+#   git push --tags
+# Then just package it up:
+#   python3 setup.py sdist
 
+import sys
+import os
+from setuptools import setup
+from distutils.extension import Extension
+
+# Versioneer manages version numbers from git tags.
+# https://github.com/warner/python-versioneer
+import versioneer
+
+# External modules that need to be built
+ext_modules = [ Extension('nilmdb.server.rocket', ['nilmdb/server/rocket.c' ]) ]
+
+# Use Cython.
+cython_modules = [ 'nilmdb.server.interval', 'nilmdb.server.rbtree' ]
+import Cython
+from Cython.Build import cythonize
+for modulename in cython_modules:
+    filename = modulename.replace('.','/')
+    ext_modules.extend(cythonize(filename + ".pyx"))
+
+# Get list of requirements to use in `install_requires` below.  Note
+# that we don't make a distinction between things that are actually
+# required for end-users vs developers (or use `test_requires` or
+# anything else) -- just install everything for simplicity.
+install_requires = open('requirements.txt').readlines()
+
+# Run setup
 setup(name='nilmdb',
-      version = '1.0',
+      version = versioneer.get_version(),
-      scripts = [ 'bin/nilm-test.py' ],
+      cmdclass = versioneer.get_cmdclass(),
-      packages = [ 'nilmdb' ],
+      url = 'https://git.jim.sh/nilm/nilmdb.git',
+      author = 'Jim Paris',
+      description = "NILM Database",
+      long_description = "NILM Database",
+      license = "Proprietary",
+      author_email = 'jim@jtan.com',
+      setup_requires = [ 'setuptools' ],
+      install_requires = install_requires,
+      packages = [ 'nilmdb',
+                   'nilmdb.utils',
+                   'nilmdb.server',
+                   'nilmdb.client',
+                   'nilmdb.cmdline',
+                   'nilmdb.scripts',
+                   'nilmdb.fsck',
+                   ],
+      entry_points = {
+          'console_scripts': [
+              'nilmtool = nilmdb.scripts.nilmtool:main',
+              'nilmdb-server = nilmdb.scripts.nilmdb_server:main',
+              'nilmdb-fsck = nilmdb.scripts.nilmdb_fsck:main',
+              ],
+          },
+      ext_modules = ext_modules,
+      zip_safe = False,
       )

test/Makefile

@@ -1,5 +0,0 @@
-all:
-	python speed-readascii.py
-
-clean:
-	rm -f *pyc

test/printf.py

@@ -1,4 +0,0 @@
-from __future__ import print_function
-def printf(str, *args):
-    print(str % args, end='')
-    

test/speed-readascii.py

@@ -1,67 +0,0 @@
-#!/usr/bin/python
-
-from printf import printf
-import time
-import re
-import numpy as np
-import itertools
-
-class Timer():
-    def __init__(self, arg):
-        self.arg = arg
-    def __enter__(self): self.start = time.time()
-    def __exit__(self, *args): printf("%s: %f lines/sec\n", self.arg, 1e6 / (time.time() - self.start))
-
-def test_split():
-    for n, line in enumerate(open('1m.raw', 'r')):
-        out = [0]*6
-        tmp = [ int(i) for i in line.partition('#')[0].split() ]
-        out[0:len(tmp)] = tmp
-        if (n % 100000 == 0):
-            printf("line %d = %s\n", n, str(out))
-
-def test_split2():
-    for n, line in enumerate(open('1m.raw', 'r')):
-        out = [0]*6
-        tmp = [ int(i,10) for i in line.partition('#')[0].split() ]
-        out[0:len(tmp)] = tmp
-        if (n % 100000 == 0):
-            printf("line %d = %s\n", n, str(out))
-
-def test_regex():
-    for n, line in enumerate(open('1m.raw', 'r')):
-        out = [0]*6
-        tmp = [ int(x) for x in re.findall('(\d+)\s+',line.partition('#')[0]) ]
-        out[0:len(tmp)] = tmp
-        if (n % 100000 == 0):
-            printf("line %d = %s\n", n, str(out))
-
-def test_bigregex():
-    regex = re.compile('^(?:\s*)' + '(?:(\d+)\s+)?' * 6)
-    for n, line in enumerate(open('1m.raw', 'r')):
-        out = [ int(x or 0) for x in re.match(regex, line).groups() ]
-        if (n % 100000 == 0):
-            printf("line %d = %s\n", n, str(out))
-
-def test_numpy():
-    out = np.genfromtxt(open('1m.raw', 'r'),
-                        dtype = np.dtype('i2,i2,i2,i2,i2,i2'))
-    
-with Timer("numpy"):
-    test_numpy() # 106k/sec
-
-with Timer("regex"):
-    test_regex() # 121k/sec
-
-with Timer("split"):
-    test_split() # 219k/sec
-
-with Timer("split2"):
-    test_split2() # 328k/sec
-
-with Timer("bigregex"):
-    test_bigregex() # 130k/sec
-
-# The "int" operation takes quite a while -- int(x,10) is twice as fast
-# Perl does about 500k/sec
-

test/speed-readbinary.py

@@ -1,74 +0,0 @@
-#!/usr/bin/python
-
-from printf import printf
-import time
-import re
-import numpy as np
-import itertools
-import struct
-import array
-import os
-import mmap
-
-class Timer():
-    def __init__(self, arg):
-        self.arg = arg
-    def __enter__(self): self.start = time.time()
-    def __exit__(self, *args): printf("%s: %f klines/sec\n", self.arg, 1e3 / (time.time() - self.start))
-
-def test_struct1():
-    """read with struct.unpack"""
-    f = open('1m.bin', 'rb')
-    f.seek(0,os.SEEK_END)
-    filesize = f.tell()
-    f.seek(0,os.SEEK_SET)
-    packer = struct.Struct('!dHHHHHH')
-    items = filesize / packer.size
-    for n in xrange(items):
-        s = f.read(packer.size)
-        out = packer.unpack(s)
-        if (n % 100000 == 0):
-            printf("line %d = %s\n", n, str(out))
-
-def test_struct2():
-    """read with struct.unpack, convert to string"""
-    f = open('1m.bin', 'rb')
-    f.seek(0,os.SEEK_END)
-    filesize = f.tell()
-    f.seek(0,os.SEEK_SET)
-    packer = struct.Struct('!dHHHHHH')
-    items = filesize / packer.size
-    for n in xrange(items):
-        s = f.read(packer.size)
-        out = packer.unpack(s)
-        x = str(out)
-        if (n % 100000 == 0):
-            printf("line %d = %s\n", n, str(out))
-
-def test_mmap():
-    """struct.unpack with mmap"""
-    with open('1m.bin', 'rb') as f:
-        f.seek(0,os.SEEK_END)
-        filesize = f.tell()
-        f.seek(0,os.SEEK_SET)
-        m = mmap.mmap(f.fileno(), filesize, access=mmap.ACCESS_READ)
-        packer = struct.Struct('!dHHHHHH')
-        items = filesize / packer.size
-        for n in xrange(items):
-            out = packer.unpack(m[packer.size*n : packer.size*(n+1)])
-            if (n % 100000 == 0):
-                printf("line %d = %s\n", n, str(out))
-
-with Timer("mmap"):
-    test_mmap()  # 1600k
-
-with Timer("struct1"):
-    test_struct1()  # 1460k
-
-with Timer("struct2"):
-    test_struct2()  # 210k
-
-# Reading from the file is again much quicker than converting to string
-# Use mmap, it's good
-
-

test/speed-writebinary.py

@@ -1,76 +0,0 @@
-#!/usr/bin/python
-
-from printf import printf
-import time
-import re
-import numpy as np
-import itertools
-import struct
-import array
-
-class Timer():
-    def __init__(self, arg):
-        self.arg = arg
-    def __enter__(self): self.start = time.time()
-    def __exit__(self, *args): printf("%s: %f klines/sec\n", self.arg, 1e3 / (time.time() - self.start))
-
-def read_ascii():
-    for n in xrange(1000000):
-        yield (1234, 2345, 3456, 4576, 5678, 6789)
-#    for n, line in enumerate(open('1m.raw', 'r')):
-#        out = [0]*6
-#        tmp = [ int(i,10) for i in line.partition('#')[0].split() ]
-#        out[0:len(tmp)] = tmp
-#        if (n % 100000 == 0):
-#            printf("line %d = %s\n", n, str(out))
-#        yield out
-
-def test_struct1():
-    """write with struct.pack"""
-    f = open('1m.bin', 'wb')
-    for out in read_ascii():
-        s = struct.pack('!HHHHHH', *out)
-        f.write(s)
-
-def test_struct2():
-    """use constant format string"""
-    f = open('1m.bin', 'wb')
-    packer = struct.Struct('!HHHHHH')
-    for out in read_ascii():
-        f.write(packer.pack(*out))
-    f.close()
-    printf("size was %d\n", packer.size)
-
-def test_struct3():
-    """like struct1, with timestamp"""
-    f = open('1m.bin', 'wb')
-    for out in read_ascii():
-        s = struct.pack('!dHHHHHH', time.time(), *out)
-        f.write(s)
-
-def test_struct4():
-    """like struct2, with timestamp"""
-    f = open('1m.bin', 'wb')
-    packer = struct.Struct('!dHHHHHH')
-    for out in read_ascii():
-        f.write(packer.pack(time.time(), *out))
-    f.close()
-    printf("size was %d\n", packer.size)
-
-#raise Exception('done')
-
-with Timer("struct1"):
-    test_struct1() # 1089k
-
-with Timer("struct2"):
-    test_struct2() # 1249k
-
-with Timer("struct3"):
-    test_struct3() # 845k
-
-with Timer("struct4"):
-    test_struct4() # 922k
-
-# This seems fast enough for writing new data, since it's faster than
-# we read ascii data anyway.  Use e.g. struct4
-

test/test-struct-pack.py

@@ -1,11 +0,0 @@
-#!/usr/bin/python
-
-import struct
-import mmap
-
-f = open("test.dat", "rb+")
-mm = mmap.mmap(f.fileno(),3)
-
-print len(mm)
-print "first 3 bytes: " + mm[0:3];
-

test/test.dat

@@ -1 +0,0 @@
-

















test/todo.md

@@ -1,7 +0,0 @@
-- Have a class representing the file contents
-- Looks like an array
-  - len(), get(), index
-  - some form of bisect search
-  - get_extents = return [0].timestamp, [-1].timestamp
-  - 
-- Can append?  Sure, why not.  Just write to the file, extend mmap accordingly.

tests/data/extract-1

@@ -0,0 +1,124 @@
+# path: /newton/prep
+# layout: float32_8
+# start: Fri, 23 Mar 2012 10:00:30.000000 +0000
+# end: Fri, 23 Mar 2012 10:00:31.000000 +0000
+1332496830000000 2.517740e+05 2.242410e+05 5.688100e+03 1.915530e+03 9.329220e+03 4.183710e+03 1.212350e+03 2.641790e+03
+1332496830008333 2.595670e+05 2.226980e+05 6.207600e+03 6.786720e+02 9.380230e+03 4.575580e+03 2.830610e+03 2.688630e+03
+1332496830016667 2.630730e+05 2.233040e+05 4.961640e+03 2.197120e+03 7.687310e+03 4.861860e+03 2.732780e+03 3.008540e+03
+1332496830025000 2.576140e+05 2.233230e+05 5.003660e+03 3.525140e+03 7.165310e+03 4.685620e+03 1.715380e+03 3.440480e+03
+1332496830033333 2.557800e+05 2.219150e+05 6.357310e+03 2.145290e+03 8.426970e+03 3.775350e+03 1.475390e+03 3.797240e+03
+1332496830041667 2.601660e+05 2.230080e+05 6.702590e+03 1.484960e+03 9.288100e+03 3.330830e+03 1.228500e+03 3.214320e+03
+1332496830050000 2.612310e+05 2.264260e+05 4.980060e+03 2.982380e+03 8.499630e+03 4.267670e+03 9.940890e+02 2.292890e+03
+1332496830058333 2.551170e+05 2.266420e+05 4.584410e+03 4.656440e+03 7.860150e+03 5.317310e+03 1.473600e+03 2.111690e+03
+1332496830066667 2.533000e+05 2.235540e+05 6.455090e+03 3.036650e+03 8.869750e+03 4.986310e+03 2.607360e+03 2.839590e+03
+1332496830075000 2.610610e+05 2.212630e+05 6.951980e+03 1.500240e+03 9.386100e+03 3.791680e+03 2.677010e+03 3.980630e+03
+1332496830083333 2.665030e+05 2.231980e+05 5.189610e+03 2.594560e+03 8.571530e+03 3.175000e+03 9.198400e+02 3.792010e+03
+1332496830091667 2.606920e+05 2.251840e+05 3.782480e+03 4.642880e+03 7.662960e+03 3.917790e+03 -2.510970e+02 2.907060e+03
+1332496830100000 2.539630e+05 2.250810e+05 5.123530e+03 3.839550e+03 8.669030e+03 4.877820e+03 9.437240e+02 2.527450e+03
+1332496830108333 2.565550e+05 2.241690e+05 5.930600e+03 2.298540e+03 8.906710e+03 5.331680e+03 2.549910e+03 3.053560e+03
+1332496830116667 2.608890e+05 2.250100e+05 4.681130e+03 2.971870e+03 7.900040e+03 4.874080e+03 2.322430e+03 3.649120e+03
+1332496830125000 2.579440e+05 2.249230e+05 3.291140e+03 4.357090e+03 7.131590e+03 4.385560e+03 1.077050e+03 3.664040e+03
+1332496830133333 2.550090e+05 2.230180e+05 4.584820e+03 2.864000e+03 8.469490e+03 3.625580e+03 9.855570e+02 3.504230e+03
+1332496830141667 2.601140e+05 2.219470e+05 5.676190e+03 1.210340e+03 9.393780e+03 3.390240e+03 1.654020e+03 3.018700e+03
+1332496830150000 2.642770e+05 2.244380e+05 4.446620e+03 2.176720e+03 8.142090e+03 4.584880e+03 2.327830e+03 2.615800e+03
+1332496830158333 2.592210e+05 2.264710e+05 2.734440e+03 4.182760e+03 6.389550e+03 5.540520e+03 1.958880e+03 2.720120e+03
+1332496830166667 2.526500e+05 2.248310e+05 4.163640e+03 2.989990e+03 7.179200e+03 5.213060e+03 1.929550e+03 3.457660e+03
+1332496830175000 2.570830e+05 2.220480e+05 5.759040e+03 7.024410e+02 8.566550e+03 3.552020e+03 1.832940e+03 3.956190e+03
+1332496830183333 2.631300e+05 2.229670e+05 5.141140e+03 1.166120e+03 8.666960e+03 2.720370e+03 9.713740e+02 3.479730e+03
+1332496830191667 2.602360e+05 2.252650e+05 3.425140e+03 3.339080e+03 7.853610e+03 3.674950e+03 5.259080e+02 2.443310e+03
+1332496830200000 2.535030e+05 2.245270e+05 4.398130e+03 2.927430e+03 8.110280e+03 4.842470e+03 1.513870e+03 2.467100e+03
+1332496830208333 2.561260e+05 2.226930e+05 6.043530e+03 6.562240e+02 8.797560e+03 4.832410e+03 2.832370e+03 3.426140e+03
+1332496830216667 2.616770e+05 2.236080e+05 5.830460e+03 1.033910e+03 8.123940e+03 3.980690e+03 1.927960e+03 4.092720e+03
+1332496830225000 2.594570e+05 2.255360e+05 4.015570e+03 2.995990e+03 7.135440e+03 3.713550e+03 3.072200e+02 3.849430e+03
+1332496830233333 2.533520e+05 2.242160e+05 4.650560e+03 3.196620e+03 8.131280e+03 3.586160e+03 7.083230e+01 3.074180e+03
+1332496830241667 2.561240e+05 2.215130e+05 6.100480e+03 8.219800e+02 9.757540e+03 3.474510e+03 1.647520e+03 2.559860e+03
+1332496830250000 2.630240e+05 2.215590e+05 5.789960e+03 6.994170e+02 9.129740e+03 4.153080e+03 2.829250e+03 2.677270e+03
+1332496830258333 2.617200e+05 2.240150e+05 4.358500e+03 2.645360e+03 7.414110e+03 4.810670e+03 2.225990e+03 3.185990e+03
+1332496830266667 2.547560e+05 2.242400e+05 4.857380e+03 3.229680e+03 7.539310e+03 4.769140e+03 1.507130e+03 3.668260e+03
+1332496830275000 2.568890e+05 2.226580e+05 6.473420e+03 1.214110e+03 9.010760e+03 3.848730e+03 1.303840e+03 3.778500e+03
+1332496830283333 2.642080e+05 2.233160e+05 5.700450e+03 1.116560e+03 9.087610e+03 3.846680e+03 1.293590e+03 2.891560e+03
+1332496830291667 2.633100e+05 2.257190e+05 3.936120e+03 3.252360e+03 7.552850e+03 4.897860e+03 1.156630e+03 2.037160e+03
+1332496830300000 2.550790e+05 2.250860e+05 4.536450e+03 3.960110e+03 7.454590e+03 5.479070e+03 1.596360e+03 2.190800e+03
+1332496830308333 2.544870e+05 2.225080e+05 6.635860e+03 1.758850e+03 8.732970e+03 4.466970e+03 2.650360e+03 3.139310e+03
+1332496830316667 2.612410e+05 2.224320e+05 6.702270e+03 1.085130e+03 8.989230e+03 3.112990e+03 1.933560e+03 3.828410e+03
+1332496830325000 2.621190e+05 2.255870e+05 4.714950e+03 2.892360e+03 8.107820e+03 2.961310e+03 2.399780e+02 3.273720e+03
+1332496830333333 2.549990e+05 2.265140e+05 4.532090e+03 4.126900e+03 8.200130e+03 3.872590e+03 5.608900e+01 2.370580e+03
+1332496830341667 2.542890e+05 2.240330e+05 6.538810e+03 2.251440e+03 9.419430e+03 4.564450e+03 2.077810e+03 2.508170e+03
+1332496830350000 2.618900e+05 2.219600e+05 6.846090e+03 1.475270e+03 9.125590e+03 4.598290e+03 3.299220e+03 3.475420e+03
+1332496830358333 2.645020e+05 2.230850e+05 5.066380e+03 3.270560e+03 7.933170e+03 4.173710e+03 1.908910e+03 3.867460e+03
+1332496830366667 2.578890e+05 2.236560e+05 4.201660e+03 4.473640e+03 7.688340e+03 4.161580e+03 6.875790e+02 3.653690e+03
+1332496830375000 2.542700e+05 2.231510e+05 5.715140e+03 2.752140e+03 9.273320e+03 3.772950e+03 8.964040e+02 3.256060e+03
+1332496830383333 2.582570e+05 2.242170e+05 6.114310e+03 1.856860e+03 9.604320e+03 4.200490e+03 1.764380e+03 2.939220e+03
+1332496830391667 2.600200e+05 2.268680e+05 4.237530e+03 3.605880e+03 8.066220e+03 5.430250e+03 2.138580e+03 2.696710e+03
+1332496830400000 2.550830e+05 2.259240e+05 3.350310e+03 4.853070e+03 7.045820e+03 5.925200e+03 1.893610e+03 2.897340e+03
+1332496830408333 2.544530e+05 2.221270e+05 5.271330e+03 2.491500e+03 8.436680e+03 5.032080e+03 2.436050e+03 3.724590e+03
+1332496830416667 2.625880e+05 2.199500e+05 5.994620e+03 7.892740e+02 9.029650e+03 3.515740e+03 1.953570e+03 4.014520e+03
+1332496830425000 2.656100e+05 2.233330e+05 4.391410e+03 2.400960e+03 8.146460e+03 3.536960e+03 5.302320e+02 3.133920e+03
+1332496830433333 2.574700e+05 2.269770e+05 2.975320e+03 4.633530e+03 7.278560e+03 4.640100e+03 -5.015020e+01 2.024960e+03
+1332496830441667 2.506870e+05 2.263310e+05 4.517860e+03 3.183800e+03 8.072600e+03 5.281660e+03 1.605140e+03 2.335140e+03
+1332496830450000 2.555630e+05 2.244950e+05 5.551000e+03 1.101300e+03 8.461490e+03 4.725700e+03 2.726670e+03 3.480540e+03
+1332496830458333 2.613350e+05 2.246450e+05 4.764680e+03 1.557020e+03 7.833350e+03 3.524810e+03 1.577410e+03 4.038620e+03
+1332496830466667 2.602690e+05 2.240080e+05 3.558030e+03 2.987610e+03 7.362440e+03 3.279230e+03 5.624420e+02 3.786550e+03
+1332496830475000 2.574350e+05 2.217770e+05 4.972600e+03 2.166880e+03 8.481440e+03 3.328720e+03 1.037130e+03 3.271370e+03
+1332496830483333 2.610460e+05 2.215500e+05 5.816180e+03 5.902170e+02 9.120930e+03 3.895400e+03 2.382670e+03 2.824170e+03
+1332496830491667 2.627660e+05 2.244730e+05 4.835050e+03 1.785770e+03 7.880760e+03 4.745620e+03 2.443660e+03 3.229550e+03
+1332496830500000 2.565090e+05 2.264130e+05 3.758870e+03 3.461200e+03 6.743770e+03 4.928960e+03 1.536620e+03 3.546690e+03
+1332496830508333 2.507930e+05 2.243720e+05 5.218490e+03 2.865260e+03 7.803960e+03 4.351090e+03 1.333820e+03 3.680490e+03
+1332496830516667 2.563190e+05 2.220660e+05 6.403970e+03 7.323450e+02 9.627760e+03 3.089300e+03 1.516780e+03 3.653690e+03
+1332496830525000 2.633430e+05 2.232350e+05 5.200430e+03 1.388580e+03 9.372850e+03 3.371230e+03 1.450390e+03 2.678910e+03
+1332496830533333 2.609030e+05 2.251100e+05 3.722580e+03 3.246660e+03 7.876540e+03 4.716810e+03 1.498440e+03 2.116520e+03
+1332496830541667 2.544160e+05 2.237690e+05 4.841650e+03 2.956400e+03 8.115920e+03 5.392360e+03 2.142810e+03 2.652320e+03
+1332496830550000 2.566980e+05 2.221720e+05 6.471230e+03 9.703960e+02 8.834980e+03 4.816840e+03 2.376630e+03 3.605860e+03
+1332496830558333 2.618410e+05 2.235370e+05 5.500740e+03 1.189660e+03 8.365730e+03 4.016470e+03 1.042270e+03 3.821200e+03
+1332496830566667 2.595030e+05 2.258400e+05 3.827930e+03 3.088840e+03 7.676140e+03 3.978310e+03 -3.570070e+02 3.016420e+03
+1332496830575000 2.534570e+05 2.246360e+05 4.914610e+03 3.097450e+03 8.224900e+03 4.321440e+03 1.713740e+02 2.412360e+03
+1332496830583333 2.560290e+05 2.222210e+05 6.841800e+03 1.028500e+03 9.252300e+03 4.387570e+03 2.418140e+03 2.510100e+03
+1332496830591667 2.628400e+05 2.225500e+05 6.210250e+03 1.410730e+03 8.538900e+03 4.152580e+03 3.009300e+03 3.219760e+03
+1332496830600000 2.616330e+05 2.250650e+05 4.284530e+03 3.357210e+03 7.282170e+03 3.823590e+03 1.402840e+03 3.644670e+03
+1332496830608333 2.545910e+05 2.251090e+05 4.693160e+03 3.647740e+03 7.745160e+03 3.686380e+03 4.901610e+02 3.448860e+03
+1332496830616667 2.547800e+05 2.235990e+05 6.527380e+03 1.569870e+03 9.438430e+03 3.456580e+03 1.162520e+03 3.252010e+03
+1332496830625000 2.606390e+05 2.241070e+05 6.531050e+03 1.633050e+03 9.283720e+03 4.174020e+03 2.089550e+03 2.775750e+03
+1332496830633333 2.611080e+05 2.254720e+05 4.968260e+03 3.527850e+03 7.692870e+03 5.137100e+03 2.207390e+03 2.436660e+03
+1332496830641667 2.557750e+05 2.237080e+05 4.963450e+03 4.017370e+03 7.701420e+03 5.269650e+03 2.284400e+03 2.842080e+03
+1332496830650000 2.573980e+05 2.209470e+05 6.767500e+03 1.645710e+03 9.107070e+03 4.000180e+03 2.548860e+03 3.624770e+03
+1332496830658333 2.649240e+05 2.215590e+05 6.471460e+03 1.110330e+03 9.459650e+03 3.108170e+03 1.696970e+03 3.893440e+03
+1332496830666667 2.653390e+05 2.257330e+05 4.348800e+03 3.459510e+03 8.475300e+03 4.031240e+03 5.733470e+02 2.910270e+03
+1332496830675000 2.568140e+05 2.269950e+05 3.479540e+03 4.949790e+03 7.499910e+03 5.624710e+03 7.516560e+02 2.347710e+03
+1332496830683333 2.533160e+05 2.251610e+05 5.147060e+03 3.218430e+03 8.460160e+03 5.869300e+03 2.336320e+03 2.987960e+03
+1332496830691667 2.593600e+05 2.231010e+05 5.549120e+03 1.869950e+03 8.740760e+03 4.668940e+03 2.457910e+03 3.758820e+03
+1332496830700000 2.620120e+05 2.240160e+05 4.173610e+03 3.004130e+03 8.157040e+03 3.704730e+03 9.879640e+02 3.652750e+03
+1332496830708333 2.571760e+05 2.244200e+05 3.517300e+03 4.118750e+03 7.822240e+03 3.718230e+03 3.726490e+01 2.953680e+03
+1332496830716667 2.551460e+05 2.233220e+05 4.923980e+03 2.330680e+03 9.095910e+03 3.792400e+03 1.013070e+03 2.711240e+03
+1332496830725000 2.605240e+05 2.236510e+05 5.413630e+03 1.146210e+03 8.817170e+03 4.419650e+03 2.446650e+03 2.832050e+03
+1332496830733333 2.620980e+05 2.257520e+05 4.262980e+03 2.270970e+03 7.135480e+03 5.067120e+03 2.294680e+03 3.376620e+03
+1332496830741667 2.568890e+05 2.253790e+05 3.606460e+03 3.568190e+03 6.552650e+03 4.970270e+03 1.516380e+03 3.662570e+03
+1332496830750000 2.539480e+05 2.226310e+05 5.511700e+03 2.066300e+03 7.952660e+03 4.019910e+03 1.513140e+03 3.752630e+03
+1332496830758333 2.597990e+05 2.220670e+05 5.873500e+03 6.085840e+02 9.253780e+03 2.870740e+03 1.348240e+03 3.344200e+03
+1332496830766667 2.625470e+05 2.249010e+05 4.346080e+03 1.928100e+03 8.590970e+03 3.455460e+03 9.043910e+02 2.379270e+03
+1332496830775000 2.561370e+05 2.267610e+05 3.423560e+03 3.379080e+03 7.471150e+03 4.894170e+03 1.153540e+03 2.031410e+03
+1332496830783333 2.503260e+05 2.250130e+05 5.519980e+03 2.423970e+03 7.991760e+03 5.117950e+03 2.098790e+03 3.099240e+03
+1332496830791667 2.554540e+05 2.229920e+05 6.547950e+03 4.964960e+02 8.751340e+03 3.900560e+03 2.132290e+03 4.076810e+03
+1332496830800000 2.612860e+05 2.234890e+05 5.152850e+03 1.501510e+03 8.425610e+03 2.888030e+03 7.761140e+02 3.786360e+03
+1332496830808333 2.589690e+05 2.240690e+05 3.832610e+03 3.001980e+03 7.979260e+03 3.182310e+03 5.271600e+01 2.874800e+03
+1332496830816667 2.549460e+05 2.220350e+05 5.317880e+03 2.139800e+03 9.103140e+03 3.955610e+03 1.235170e+03 2.394150e+03
+1332496830825000 2.586760e+05 2.212050e+05 6.594910e+03 5.053440e+02 9.423360e+03 4.562470e+03 2.913740e+03 2.892350e+03
+1332496830833333 2.621250e+05 2.235660e+05 5.116750e+03 1.773600e+03 8.082200e+03 4.776370e+03 2.386390e+03 3.659730e+03
+1332496830841667 2.578350e+05 2.259180e+05 3.714300e+03 3.477080e+03 7.205370e+03 4.554610e+03 7.115390e+02 3.878420e+03
+1332496830850000 2.536600e+05 2.243710e+05 5.022450e+03 2.592430e+03 8.277200e+03 4.119370e+03 4.865080e+02 3.666740e+03
+1332496830858333 2.595030e+05 2.220610e+05 6.589950e+03 6.599360e+02 9.596920e+03 3.598100e+03 1.702490e+03 3.036600e+03
+1332496830866667 2.654950e+05 2.228430e+05 5.541850e+03 1.728430e+03 8.459960e+03 4.492000e+03 2.231970e+03 2.430620e+03
+1332496830875000 2.609290e+05 2.249960e+05 4.000950e+03 3.745990e+03 6.983790e+03 5.430860e+03 1.855260e+03 2.533380e+03
+1332496830883333 2.527160e+05 2.243350e+05 5.086560e+03 3.401150e+03 7.597970e+03 5.196120e+03 1.755720e+03 3.079760e+03
+1332496830891667 2.541100e+05 2.231110e+05 6.822190e+03 1.229080e+03 9.164340e+03 3.761230e+03 1.679390e+03 3.584880e+03
+1332496830900000 2.599690e+05 2.246930e+05 6.183950e+03 1.538500e+03 9.222080e+03 3.139170e+03 9.499020e+02 3.180800e+03
+1332496830908333 2.590780e+05 2.269130e+05 4.388890e+03 3.694820e+03 8.195020e+03 3.933000e+03 4.260800e+02 2.388450e+03
+1332496830916667 2.545630e+05 2.247600e+05 5.168440e+03 4.020940e+03 8.450270e+03 4.758910e+03 1.458900e+03 2.286430e+03
+1332496830925000 2.580590e+05 2.212170e+05 6.883460e+03 1.649530e+03 9.232780e+03 4.457650e+03 3.057820e+03 3.031950e+03
+1332496830933333 2.646670e+05 2.211770e+05 6.218510e+03 1.645730e+03 8.657180e+03 3.663500e+03 2.528280e+03 3.978340e+03
+1332496830941667 2.629250e+05 2.243820e+05 4.627500e+03 3.635930e+03 7.892800e+03 3.431320e+03 6.045090e+02 3.901370e+03
+1332496830950000 2.547080e+05 2.254480e+05 4.408250e+03 4.461040e+03 8.197170e+03 3.953750e+03 -4.453460e+01 3.154870e+03
+1332496830958333 2.537020e+05 2.246350e+05 5.825770e+03 2.577050e+03 9.590050e+03 4.569250e+03 1.460270e+03 2.785170e+03
+1332496830966667 2.602060e+05 2.241400e+05 5.387980e+03 1.951160e+03 8.789510e+03 5.131660e+03 2.706380e+03 2.972480e+03
+1332496830975000 2.612400e+05 2.247370e+05 3.860810e+03 3.418310e+03 7.414530e+03 5.284520e+03 2.271380e+03 3.183150e+03
+1332496830983333 2.561400e+05 2.232520e+05 3.850010e+03 3.957140e+03 7.262650e+03 4.964640e+03 1.499510e+03 3.453130e+03
+1332496830991667 2.561160e+05 2.213490e+05 5.594480e+03 2.054400e+03 8.835130e+03 3.662010e+03 1.485510e+03 3.613010e+03

tests/data/extract-2

@@ -0,0 +1,119 @@
+1332496830008333 2.595670e+05 2.226980e+05 6.207600e+03 6.786720e+02 9.380230e+03 4.575580e+03 2.830610e+03 2.688630e+03
+1332496830016667 2.630730e+05 2.233040e+05 4.961640e+03 2.197120e+03 7.687310e+03 4.861860e+03 2.732780e+03 3.008540e+03
+1332496830025000 2.576140e+05 2.233230e+05 5.003660e+03 3.525140e+03 7.165310e+03 4.685620e+03 1.715380e+03 3.440480e+03
+1332496830033333 2.557800e+05 2.219150e+05 6.357310e+03 2.145290e+03 8.426970e+03 3.775350e+03 1.475390e+03 3.797240e+03
+1332496830041667 2.601660e+05 2.230080e+05 6.702590e+03 1.484960e+03 9.288100e+03 3.330830e+03 1.228500e+03 3.214320e+03
+1332496830050000 2.612310e+05 2.264260e+05 4.980060e+03 2.982380e+03 8.499630e+03 4.267670e+03 9.940890e+02 2.292890e+03
+1332496830058333 2.551170e+05 2.266420e+05 4.584410e+03 4.656440e+03 7.860150e+03 5.317310e+03 1.473600e+03 2.111690e+03
+1332496830066667 2.533000e+05 2.235540e+05 6.455090e+03 3.036650e+03 8.869750e+03 4.986310e+03 2.607360e+03 2.839590e+03
+1332496830075000 2.610610e+05 2.212630e+05 6.951980e+03 1.500240e+03 9.386100e+03 3.791680e+03 2.677010e+03 3.980630e+03
+1332496830083333 2.665030e+05 2.231980e+05 5.189610e+03 2.594560e+03 8.571530e+03 3.175000e+03 9.198400e+02 3.792010e+03
+1332496830091667 2.606920e+05 2.251840e+05 3.782480e+03 4.642880e+03 7.662960e+03 3.917790e+03 -2.510970e+02 2.907060e+03
+1332496830100000 2.539630e+05 2.250810e+05 5.123530e+03 3.839550e+03 8.669030e+03 4.877820e+03 9.437240e+02 2.527450e+03
+1332496830108333 2.565550e+05 2.241690e+05 5.930600e+03 2.298540e+03 8.906710e+03 5.331680e+03 2.549910e+03 3.053560e+03
+1332496830116667 2.608890e+05 2.250100e+05 4.681130e+03 2.971870e+03 7.900040e+03 4.874080e+03 2.322430e+03 3.649120e+03
+1332496830125000 2.579440e+05 2.249230e+05 3.291140e+03 4.357090e+03 7.131590e+03 4.385560e+03 1.077050e+03 3.664040e+03
+1332496830133333 2.550090e+05 2.230180e+05 4.584820e+03 2.864000e+03 8.469490e+03 3.625580e+03 9.855570e+02 3.504230e+03
+1332496830141667 2.601140e+05 2.219470e+05 5.676190e+03 1.210340e+03 9.393780e+03 3.390240e+03 1.654020e+03 3.018700e+03
+1332496830150000 2.642770e+05 2.244380e+05 4.446620e+03 2.176720e+03 8.142090e+03 4.584880e+03 2.327830e+03 2.615800e+03
+1332496830158333 2.592210e+05 2.264710e+05 2.734440e+03 4.182760e+03 6.389550e+03 5.540520e+03 1.958880e+03 2.720120e+03
+1332496830166667 2.526500e+05 2.248310e+05 4.163640e+03 2.989990e+03 7.179200e+03 5.213060e+03 1.929550e+03 3.457660e+03
+1332496830175000 2.570830e+05 2.220480e+05 5.759040e+03 7.024410e+02 8.566550e+03 3.552020e+03 1.832940e+03 3.956190e+03
+1332496830183333 2.631300e+05 2.229670e+05 5.141140e+03 1.166120e+03 8.666960e+03 2.720370e+03 9.713740e+02 3.479730e+03
+1332496830191667 2.602360e+05 2.252650e+05 3.425140e+03 3.339080e+03 7.853610e+03 3.674950e+03 5.259080e+02 2.443310e+03
+1332496830200000 2.535030e+05 2.245270e+05 4.398130e+03 2.927430e+03 8.110280e+03 4.842470e+03 1.513870e+03 2.467100e+03
+1332496830208333 2.561260e+05 2.226930e+05 6.043530e+03 6.562240e+02 8.797560e+03 4.832410e+03 2.832370e+03 3.426140e+03
+1332496830216667 2.616770e+05 2.236080e+05 5.830460e+03 1.033910e+03 8.123940e+03 3.980690e+03 1.927960e+03 4.092720e+03
+1332496830225000 2.594570e+05 2.255360e+05 4.015570e+03 2.995990e+03 7.135440e+03 3.713550e+03 3.072200e+02 3.849430e+03
+1332496830233333 2.533520e+05 2.242160e+05 4.650560e+03 3.196620e+03 8.131280e+03 3.586160e+03 7.083230e+01 3.074180e+03
+1332496830241667 2.561240e+05 2.215130e+05 6.100480e+03 8.219800e+02 9.757540e+03 3.474510e+03 1.647520e+03 2.559860e+03
+1332496830250000 2.630240e+05 2.215590e+05 5.789960e+03 6.994170e+02 9.129740e+03 4.153080e+03 2.829250e+03 2.677270e+03
+1332496830258333 2.617200e+05 2.240150e+05 4.358500e+03 2.645360e+03 7.414110e+03 4.810670e+03 2.225990e+03 3.185990e+03
+1332496830266667 2.547560e+05 2.242400e+05 4.857380e+03 3.229680e+03 7.539310e+03 4.769140e+03 1.507130e+03 3.668260e+03
+1332496830275000 2.568890e+05 2.226580e+05 6.473420e+03 1.214110e+03 9.010760e+03 3.848730e+03 1.303840e+03 3.778500e+03
+1332496830283333 2.642080e+05 2.233160e+05 5.700450e+03 1.116560e+03 9.087610e+03 3.846680e+03 1.293590e+03 2.891560e+03
+1332496830291667 2.633100e+05 2.257190e+05 3.936120e+03 3.252360e+03 7.552850e+03 4.897860e+03 1.156630e+03 2.037160e+03
+1332496830300000 2.550790e+05 2.250860e+05 4.536450e+03 3.960110e+03 7.454590e+03 5.479070e+03 1.596360e+03 2.190800e+03
+1332496830308333 2.544870e+05 2.225080e+05 6.635860e+03 1.758850e+03 8.732970e+03 4.466970e+03 2.650360e+03 3.139310e+03
+1332496830316667 2.612410e+05 2.224320e+05 6.702270e+03 1.085130e+03 8.989230e+03 3.112990e+03 1.933560e+03 3.828410e+03
+1332496830325000 2.621190e+05 2.255870e+05 4.714950e+03 2.892360e+03 8.107820e+03 2.961310e+03 2.399780e+02 3.273720e+03
+1332496830333333 2.549990e+05 2.265140e+05 4.532090e+03 4.126900e+03 8.200130e+03 3.872590e+03 5.608900e+01 2.370580e+03
+1332496830341667 2.542890e+05 2.240330e+05 6.538810e+03 2.251440e+03 9.419430e+03 4.564450e+03 2.077810e+03 2.508170e+03
+1332496830350000 2.618900e+05 2.219600e+05 6.846090e+03 1.475270e+03 9.125590e+03 4.598290e+03 3.299220e+03 3.475420e+03
+1332496830358333 2.645020e+05 2.230850e+05 5.066380e+03 3.270560e+03 7.933170e+03 4.173710e+03 1.908910e+03 3.867460e+03
+1332496830366667 2.578890e+05 2.236560e+05 4.201660e+03 4.473640e+03 7.688340e+03 4.161580e+03 6.875790e+02 3.653690e+03
+1332496830375000 2.542700e+05 2.231510e+05 5.715140e+03 2.752140e+03 9.273320e+03 3.772950e+03 8.964040e+02 3.256060e+03
+1332496830383333 2.582570e+05 2.242170e+05 6.114310e+03 1.856860e+03 9.604320e+03 4.200490e+03 1.764380e+03 2.939220e+03
+1332496830391667 2.600200e+05 2.268680e+05 4.237530e+03 3.605880e+03 8.066220e+03 5.430250e+03 2.138580e+03 2.696710e+03
+1332496830400000 2.550830e+05 2.259240e+05 3.350310e+03 4.853070e+03 7.045820e+03 5.925200e+03 1.893610e+03 2.897340e+03
+1332496830408333 2.544530e+05 2.221270e+05 5.271330e+03 2.491500e+03 8.436680e+03 5.032080e+03 2.436050e+03 3.724590e+03
+1332496830416667 2.625880e+05 2.199500e+05 5.994620e+03 7.892740e+02 9.029650e+03 3.515740e+03 1.953570e+03 4.014520e+03
+1332496830425000 2.656100e+05 2.233330e+05 4.391410e+03 2.400960e+03 8.146460e+03 3.536960e+03 5.302320e+02 3.133920e+03
+1332496830433333 2.574700e+05 2.269770e+05 2.975320e+03 4.633530e+03 7.278560e+03 4.640100e+03 -5.015020e+01 2.024960e+03
+1332496830441667 2.506870e+05 2.263310e+05 4.517860e+03 3.183800e+03 8.072600e+03 5.281660e+03 1.605140e+03 2.335140e+03
+1332496830450000 2.555630e+05 2.244950e+05 5.551000e+03 1.101300e+03 8.461490e+03 4.725700e+03 2.726670e+03 3.480540e+03
+1332496830458333 2.613350e+05 2.246450e+05 4.764680e+03 1.557020e+03 7.833350e+03 3.524810e+03 1.577410e+03 4.038620e+03
+1332496830466667 2.602690e+05 2.240080e+05 3.558030e+03 2.987610e+03 7.362440e+03 3.279230e+03 5.624420e+02 3.786550e+03
+1332496830475000 2.574350e+05 2.217770e+05 4.972600e+03 2.166880e+03 8.481440e+03 3.328720e+03 1.037130e+03 3.271370e+03
+1332496830483333 2.610460e+05 2.215500e+05 5.816180e+03 5.902170e+02 9.120930e+03 3.895400e+03 2.382670e+03 2.824170e+03
+1332496830491667 2.627660e+05 2.244730e+05 4.835050e+03 1.785770e+03 7.880760e+03 4.745620e+03 2.443660e+03 3.229550e+03
+1332496830500000 2.565090e+05 2.264130e+05 3.758870e+03 3.461200e+03 6.743770e+03 4.928960e+03 1.536620e+03 3.546690e+03
+1332496830508333 2.507930e+05 2.243720e+05 5.218490e+03 2.865260e+03 7.803960e+03 4.351090e+03 1.333820e+03 3.680490e+03
+1332496830516667 2.563190e+05 2.220660e+05 6.403970e+03 7.323450e+02 9.627760e+03 3.089300e+03 1.516780e+03 3.653690e+03
+1332496830525000 2.633430e+05 2.232350e+05 5.200430e+03 1.388580e+03 9.372850e+03 3.371230e+03 1.450390e+03 2.678910e+03
+1332496830533333 2.609030e+05 2.251100e+05 3.722580e+03 3.246660e+03 7.876540e+03 4.716810e+03 1.498440e+03 2.116520e+03
+1332496830541667 2.544160e+05 2.237690e+05 4.841650e+03 2.956400e+03 8.115920e+03 5.392360e+03 2.142810e+03 2.652320e+03
+1332496830550000 2.566980e+05 2.221720e+05 6.471230e+03 9.703960e+02 8.834980e+03 4.816840e+03 2.376630e+03 3.605860e+03
+1332496830558333 2.618410e+05 2.235370e+05 5.500740e+03 1.189660e+03 8.365730e+03 4.016470e+03 1.042270e+03 3.821200e+03
+1332496830566667 2.595030e+05 2.258400e+05 3.827930e+03 3.088840e+03 7.676140e+03 3.978310e+03 -3.570070e+02 3.016420e+03
+1332496830575000 2.534570e+05 2.246360e+05 4.914610e+03 3.097450e+03 8.224900e+03 4.321440e+03 1.713740e+02 2.412360e+03
+1332496830583333 2.560290e+05 2.222210e+05 6.841800e+03 1.028500e+03 9.252300e+03 4.387570e+03 2.418140e+03 2.510100e+03
+1332496830591667 2.628400e+05 2.225500e+05 6.210250e+03 1.410730e+03 8.538900e+03 4.152580e+03 3.009300e+03 3.219760e+03
+1332496830600000 2.616330e+05 2.250650e+05 4.284530e+03 3.357210e+03 7.282170e+03 3.823590e+03 1.402840e+03 3.644670e+03
+1332496830608333 2.545910e+05 2.251090e+05 4.693160e+03 3.647740e+03 7.745160e+03 3.686380e+03 4.901610e+02 3.448860e+03
+1332496830616667 2.547800e+05 2.235990e+05 6.527380e+03 1.569870e+03 9.438430e+03 3.456580e+03 1.162520e+03 3.252010e+03
+1332496830625000 2.606390e+05 2.241070e+05 6.531050e+03 1.633050e+03 9.283720e+03 4.174020e+03 2.089550e+03 2.775750e+03
+1332496830633333 2.611080e+05 2.254720e+05 4.968260e+03 3.527850e+03 7.692870e+03 5.137100e+03 2.207390e+03 2.436660e+03
+1332496830641667 2.557750e+05 2.237080e+05 4.963450e+03 4.017370e+03 7.701420e+03 5.269650e+03 2.284400e+03 2.842080e+03
+1332496830650000 2.573980e+05 2.209470e+05 6.767500e+03 1.645710e+03 9.107070e+03 4.000180e+03 2.548860e+03 3.624770e+03
+1332496830658333 2.649240e+05 2.215590e+05 6.471460e+03 1.110330e+03 9.459650e+03 3.108170e+03 1.696970e+03 3.893440e+03
+1332496830666667 2.653390e+05 2.257330e+05 4.348800e+03 3.459510e+03 8.475300e+03 4.031240e+03 5.733470e+02 2.910270e+03
+1332496830675000 2.568140e+05 2.269950e+05 3.479540e+03 4.949790e+03 7.499910e+03 5.624710e+03 7.516560e+02 2.347710e+03
+1332496830683333 2.533160e+05 2.251610e+05 5.147060e+03 3.218430e+03 8.460160e+03 5.869300e+03 2.336320e+03 2.987960e+03
+1332496830691667 2.593600e+05 2.231010e+05 5.549120e+03 1.869950e+03 8.740760e+03 4.668940e+03 2.457910e+03 3.758820e+03
+1332496830700000 2.620120e+05 2.240160e+05 4.173610e+03 3.004130e+03 8.157040e+03 3.704730e+03 9.879640e+02 3.652750e+03
+1332496830708333 2.571760e+05 2.244200e+05 3.517300e+03 4.118750e+03 7.822240e+03 3.718230e+03 3.726490e+01 2.953680e+03
+1332496830716667 2.551460e+05 2.233220e+05 4.923980e+03 2.330680e+03 9.095910e+03 3.792400e+03 1.013070e+03 2.711240e+03
+1332496830725000 2.605240e+05 2.236510e+05 5.413630e+03 1.146210e+03 8.817170e+03 4.419650e+03 2.446650e+03 2.832050e+03
+1332496830733333 2.620980e+05 2.257520e+05 4.262980e+03 2.270970e+03 7.135480e+03 5.067120e+03 2.294680e+03 3.376620e+03
+1332496830741667 2.568890e+05 2.253790e+05 3.606460e+03 3.568190e+03 6.552650e+03 4.970270e+03 1.516380e+03 3.662570e+03
+1332496830750000 2.539480e+05 2.226310e+05 5.511700e+03 2.066300e+03 7.952660e+03 4.019910e+03 1.513140e+03 3.752630e+03
+1332496830758333 2.597990e+05 2.220670e+05 5.873500e+03 6.085840e+02 9.253780e+03 2.870740e+03 1.348240e+03 3.344200e+03
+1332496830766667 2.625470e+05 2.249010e+05 4.346080e+03 1.928100e+03 8.590970e+03 3.455460e+03 9.043910e+02 2.379270e+03
+1332496830775000 2.561370e+05 2.267610e+05 3.423560e+03 3.379080e+03 7.471150e+03 4.894170e+03 1.153540e+03 2.031410e+03
+1332496830783333 2.503260e+05 2.250130e+05 5.519980e+03 2.423970e+03 7.991760e+03 5.117950e+03 2.098790e+03 3.099240e+03
+1332496830791667 2.554540e+05 2.229920e+05 6.547950e+03 4.964960e+02 8.751340e+03 3.900560e+03 2.132290e+03 4.076810e+03
+1332496830800000 2.612860e+05 2.234890e+05 5.152850e+03 1.501510e+03 8.425610e+03 2.888030e+03 7.761140e+02 3.786360e+03
+1332496830808333 2.589690e+05 2.240690e+05 3.832610e+03 3.001980e+03 7.979260e+03 3.182310e+03 5.271600e+01 2.874800e+03
+1332496830816667 2.549460e+05 2.220350e+05 5.317880e+03 2.139800e+03 9.103140e+03 3.955610e+03 1.235170e+03 2.394150e+03
+1332496830825000 2.586760e+05 2.212050e+05 6.594910e+03 5.053440e+02 9.423360e+03 4.562470e+03 2.913740e+03 2.892350e+03
+1332496830833333 2.621250e+05 2.235660e+05 5.116750e+03 1.773600e+03 8.082200e+03 4.776370e+03 2.386390e+03 3.659730e+03
+1332496830841667 2.578350e+05 2.259180e+05 3.714300e+03 3.477080e+03 7.205370e+03 4.554610e+03 7.115390e+02 3.878420e+03
+1332496830850000 2.536600e+05 2.243710e+05 5.022450e+03 2.592430e+03 8.277200e+03 4.119370e+03 4.865080e+02 3.666740e+03
+1332496830858333 2.595030e+05 2.220610e+05 6.589950e+03 6.599360e+02 9.596920e+03 3.598100e+03 1.702490e+03 3.036600e+03
+1332496830866667 2.654950e+05 2.228430e+05 5.541850e+03 1.728430e+03 8.459960e+03 4.492000e+03 2.231970e+03 2.430620e+03
+1332496830875000 2.609290e+05 2.249960e+05 4.000950e+03 3.745990e+03 6.983790e+03 5.430860e+03 1.855260e+03 2.533380e+03
+1332496830883333 2.527160e+05 2.243350e+05 5.086560e+03 3.401150e+03 7.597970e+03 5.196120e+03 1.755720e+03 3.079760e+03
+1332496830891667 2.541100e+05 2.231110e+05 6.822190e+03 1.229080e+03 9.164340e+03 3.761230e+03 1.679390e+03 3.584880e+03
+1332496830900000 2.599690e+05 2.246930e+05 6.183950e+03 1.538500e+03 9.222080e+03 3.139170e+03 9.499020e+02 3.180800e+03
+1332496830908333 2.590780e+05 2.269130e+05 4.388890e+03 3.694820e+03 8.195020e+03 3.933000e+03 4.260800e+02 2.388450e+03
+1332496830916667 2.545630e+05 2.247600e+05 5.168440e+03 4.020940e+03 8.450270e+03 4.758910e+03 1.458900e+03 2.286430e+03
+1332496830925000 2.580590e+05 2.212170e+05 6.883460e+03 1.649530e+03 9.232780e+03 4.457650e+03 3.057820e+03 3.031950e+03
+1332496830933333 2.646670e+05 2.211770e+05 6.218510e+03 1.645730e+03 8.657180e+03 3.663500e+03 2.528280e+03 3.978340e+03
+1332496830941667 2.629250e+05 2.243820e+05 4.627500e+03 3.635930e+03 7.892800e+03 3.431320e+03 6.045090e+02 3.901370e+03
+1332496830950000 2.547080e+05 2.254480e+05 4.408250e+03 4.461040e+03 8.197170e+03 3.953750e+03 -4.453460e+01 3.154870e+03
+1332496830958333 2.537020e+05 2.246350e+05 5.825770e+03 2.577050e+03 9.590050e+03 4.569250e+03 1.460270e+03 2.785170e+03
+1332496830966667 2.602060e+05 2.241400e+05 5.387980e+03 1.951160e+03 8.789510e+03 5.131660e+03 2.706380e+03 2.972480e+03
+1332496830975000 2.612400e+05 2.247370e+05 3.860810e+03 3.418310e+03 7.414530e+03 5.284520e+03 2.271380e+03 3.183150e+03
+1332496830983333 2.561400e+05 2.232520e+05 3.850010e+03 3.957140e+03 7.262650e+03 4.964640e+03 1.499510e+03 3.453130e+03
+1332496830991667 2.561160e+05 2.213490e+05 5.594480e+03 2.054400e+03 8.835130e+03 3.662010e+03 1.485510e+03 3.613010e+03

tests/data/extract-3

@@ -0,0 +1 @@
+1332496830008333 2.595670e+05 2.226980e+05 6.207600e+03 6.786720e+02 9.380230e+03 4.575580e+03 2.830610e+03 2.688630e+03

tests/data/extract-4

@@ -0,0 +1,2 @@
+1332496830008333 2.595670e+05 2.226980e+05 6.207600e+03 6.786720e+02 9.380230e+03 4.575580e+03 2.830610e+03 2.688630e+03
+1332496830016667 2.630730e+05 2.233040e+05 4.961640e+03 2.197120e+03 7.687310e+03 4.861860e+03 2.732780e+03 3.008540e+03

tests/data/extract-5

@@ -0,0 +1,124 @@
+# path: /newton/prep
+# layout: float32_8
+# start: Fri, 23 Mar 2012 10:00:30.000000 +0000
+# end: Fri, 23 Mar 2012 10:00:31.000000 +0000
+2.517740e+05 2.242410e+05 5.688100e+03 1.915530e+03 9.329220e+03 4.183710e+03 1.212350e+03 2.641790e+03
+2.595670e+05 2.226980e+05 6.207600e+03 6.786720e+02 9.380230e+03 4.575580e+03 2.830610e+03 2.688630e+03
+2.630730e+05 2.233040e+05 4.961640e+03 2.197120e+03 7.687310e+03 4.861860e+03 2.732780e+03 3.008540e+03
+2.576140e+05 2.233230e+05 5.003660e+03 3.525140e+03 7.165310e+03 4.685620e+03 1.715380e+03 3.440480e+03
+2.557800e+05 2.219150e+05 6.357310e+03 2.145290e+03 8.426970e+03 3.775350e+03 1.475390e+03 3.797240e+03
+2.601660e+05 2.230080e+05 6.702590e+03 1.484960e+03 9.288100e+03 3.330830e+03 1.228500e+03 3.214320e+03
+2.612310e+05 2.264260e+05 4.980060e+03 2.982380e+03 8.499630e+03 4.267670e+03 9.940890e+02 2.292890e+03
+2.551170e+05 2.266420e+05 4.584410e+03 4.656440e+03 7.860150e+03 5.317310e+03 1.473600e+03 2.111690e+03
+2.533000e+05 2.235540e+05 6.455090e+03 3.036650e+03 8.869750e+03 4.986310e+03 2.607360e+03 2.839590e+03
+2.610610e+05 2.212630e+05 6.951980e+03 1.500240e+03 9.386100e+03 3.791680e+03 2.677010e+03 3.980630e+03
+2.665030e+05 2.231980e+05 5.189610e+03 2.594560e+03 8.571530e+03 3.175000e+03 9.198400e+02 3.792010e+03
+2.606920e+05 2.251840e+05 3.782480e+03 4.642880e+03 7.662960e+03 3.917790e+03 -2.510970e+02 2.907060e+03
+2.539630e+05 2.250810e+05 5.123530e+03 3.839550e+03 8.669030e+03 4.877820e+03 9.437240e+02 2.527450e+03
+2.565550e+05 2.241690e+05 5.930600e+03 2.298540e+03 8.906710e+03 5.331680e+03 2.549910e+03 3.053560e+03
+2.608890e+05 2.250100e+05 4.681130e+03 2.971870e+03 7.900040e+03 4.874080e+03 2.322430e+03 3.649120e+03
+2.579440e+05 2.249230e+05 3.291140e+03 4.357090e+03 7.131590e+03 4.385560e+03 1.077050e+03 3.664040e+03
+2.550090e+05 2.230180e+05 4.584820e+03 2.864000e+03 8.469490e+03 3.625580e+03 9.855570e+02 3.504230e+03
+2.601140e+05 2.219470e+05 5.676190e+03 1.210340e+03 9.393780e+03 3.390240e+03 1.654020e+03 3.018700e+03
+2.642770e+05 2.244380e+05 4.446620e+03 2.176720e+03 8.142090e+03 4.584880e+03 2.327830e+03 2.615800e+03
+2.592210e+05 2.264710e+05 2.734440e+03 4.182760e+03 6.389550e+03 5.540520e+03 1.958880e+03 2.720120e+03
+2.526500e+05 2.248310e+05 4.163640e+03 2.989990e+03 7.179200e+03 5.213060e+03 1.929550e+03 3.457660e+03
+2.570830e+05 2.220480e+05 5.759040e+03 7.024410e+02 8.566550e+03 3.552020e+03 1.832940e+03 3.956190e+03
+2.631300e+05 2.229670e+05 5.141140e+03 1.166120e+03 8.666960e+03 2.720370e+03 9.713740e+02 3.479730e+03
+2.602360e+05 2.252650e+05 3.425140e+03 3.339080e+03 7.853610e+03 3.674950e+03 5.259080e+02 2.443310e+03
+2.535030e+05 2.245270e+05 4.398130e+03 2.927430e+03 8.110280e+03 4.842470e+03 1.513870e+03 2.467100e+03
+2.561260e+05 2.226930e+05 6.043530e+03 6.562240e+02 8.797560e+03 4.832410e+03 2.832370e+03 3.426140e+03
+2.616770e+05 2.236080e+05 5.830460e+03 1.033910e+03 8.123940e+03 3.980690e+03 1.927960e+03 4.092720e+03
+2.594570e+05 2.255360e+05 4.015570e+03 2.995990e+03 7.135440e+03 3.713550e+03 3.072200e+02 3.849430e+03
+2.533520e+05 2.242160e+05 4.650560e+03 3.196620e+03 8.131280e+03 3.586160e+03 7.083230e+01 3.074180e+03
+2.561240e+05 2.215130e+05 6.100480e+03 8.219800e+02 9.757540e+03 3.474510e+03 1.647520e+03 2.559860e+03
+2.630240e+05 2.215590e+05 5.789960e+03 6.994170e+02 9.129740e+03 4.153080e+03 2.829250e+03 2.677270e+03
+2.617200e+05 2.240150e+05 4.358500e+03 2.645360e+03 7.414110e+03 4.810670e+03 2.225990e+03 3.185990e+03
+2.547560e+05 2.242400e+05 4.857380e+03 3.229680e+03 7.539310e+03 4.769140e+03 1.507130e+03 3.668260e+03
+2.568890e+05 2.226580e+05 6.473420e+03 1.214110e+03 9.010760e+03 3.848730e+03 1.303840e+03 3.778500e+03
+2.642080e+05 2.233160e+05 5.700450e+03 1.116560e+03 9.087610e+03 3.846680e+03 1.293590e+03 2.891560e+03
+2.633100e+05 2.257190e+05 3.936120e+03 3.252360e+03 7.552850e+03 4.897860e+03 1.156630e+03 2.037160e+03
+2.550790e+05 2.250860e+05 4.536450e+03 3.960110e+03 7.454590e+03 5.479070e+03 1.596360e+03 2.190800e+03
+2.544870e+05 2.225080e+05 6.635860e+03 1.758850e+03 8.732970e+03 4.466970e+03 2.650360e+03 3.139310e+03
+2.612410e+05 2.224320e+05 6.702270e+03 1.085130e+03 8.989230e+03 3.112990e+03 1.933560e+03 3.828410e+03
+2.621190e+05 2.255870e+05 4.714950e+03 2.892360e+03 8.107820e+03 2.961310e+03 2.399780e+02 3.273720e+03
+2.549990e+05 2.265140e+05 4.532090e+03 4.126900e+03 8.200130e+03 3.872590e+03 5.608900e+01 2.370580e+03
+2.542890e+05 2.240330e+05 6.538810e+03 2.251440e+03 9.419430e+03 4.564450e+03 2.077810e+03 2.508170e+03
+2.618900e+05 2.219600e+05 6.846090e+03 1.475270e+03 9.125590e+03 4.598290e+03 3.299220e+03 3.475420e+03
+2.645020e+05 2.230850e+05 5.066380e+03 3.270560e+03 7.933170e+03 4.173710e+03 1.908910e+03 3.867460e+03
+2.578890e+05 2.236560e+05 4.201660e+03 4.473640e+03 7.688340e+03 4.161580e+03 6.875790e+02 3.653690e+03
+2.542700e+05 2.231510e+05 5.715140e+03 2.752140e+03 9.273320e+03 3.772950e+03 8.964040e+02 3.256060e+03
+2.582570e+05 2.242170e+05 6.114310e+03 1.856860e+03 9.604320e+03 4.200490e+03 1.764380e+03 2.939220e+03
+2.600200e+05 2.268680e+05 4.237530e+03 3.605880e+03 8.066220e+03 5.430250e+03 2.138580e+03 2.696710e+03
+2.550830e+05 2.259240e+05 3.350310e+03 4.853070e+03 7.045820e+03 5.925200e+03 1.893610e+03 2.897340e+03
+2.544530e+05 2.221270e+05 5.271330e+03 2.491500e+03 8.436680e+03 5.032080e+03 2.436050e+03 3.724590e+03
+2.625880e+05 2.199500e+05 5.994620e+03 7.892740e+02 9.029650e+03 3.515740e+03 1.953570e+03 4.014520e+03
+2.656100e+05 2.233330e+05 4.391410e+03 2.400960e+03 8.146460e+03 3.536960e+03 5.302320e+02 3.133920e+03
+2.574700e+05 2.269770e+05 2.975320e+03 4.633530e+03 7.278560e+03 4.640100e+03 -5.015020e+01 2.024960e+03
+2.506870e+05 2.263310e+05 4.517860e+03 3.183800e+03 8.072600e+03 5.281660e+03 1.605140e+03 2.335140e+03
+2.555630e+05 2.244950e+05 5.551000e+03 1.101300e+03 8.461490e+03 4.725700e+03 2.726670e+03 3.480540e+03
+2.613350e+05 2.246450e+05 4.764680e+03 1.557020e+03 7.833350e+03 3.524810e+03 1.577410e+03 4.038620e+03
+2.602690e+05 2.240080e+05 3.558030e+03 2.987610e+03 7.362440e+03 3.279230e+03 5.624420e+02 3.786550e+03
+2.574350e+05 2.217770e+05 4.972600e+03 2.166880e+03 8.481440e+03 3.328720e+03 1.037130e+03 3.271370e+03
+2.610460e+05 2.215500e+05 5.816180e+03 5.902170e+02 9.120930e+03 3.895400e+03 2.382670e+03 2.824170e+03
+2.627660e+05 2.244730e+05 4.835050e+03 1.785770e+03 7.880760e+03 4.745620e+03 2.443660e+03 3.229550e+03
+2.565090e+05 2.264130e+05 3.758870e+03 3.461200e+03 6.743770e+03 4.928960e+03 1.536620e+03 3.546690e+03
+2.507930e+05 2.243720e+05 5.218490e+03 2.865260e+03 7.803960e+03 4.351090e+03 1.333820e+03 3.680490e+03
+2.563190e+05 2.220660e+05 6.403970e+03 7.323450e+02 9.627760e+03 3.089300e+03 1.516780e+03 3.653690e+03
+2.633430e+05 2.232350e+05 5.200430e+03 1.388580e+03 9.372850e+03 3.371230e+03 1.450390e+03 2.678910e+03
+2.609030e+05 2.251100e+05 3.722580e+03 3.246660e+03 7.876540e+03 4.716810e+03 1.498440e+03 2.116520e+03
+2.544160e+05 2.237690e+05 4.841650e+03 2.956400e+03 8.115920e+03 5.392360e+03 2.142810e+03 2.652320e+03
+2.566980e+05 2.221720e+05 6.471230e+03 9.703960e+02 8.834980e+03 4.816840e+03 2.376630e+03 3.605860e+03
+2.618410e+05 2.235370e+05 5.500740e+03 1.189660e+03 8.365730e+03 4.016470e+03 1.042270e+03 3.821200e+03
+2.595030e+05 2.258400e+05 3.827930e+03 3.088840e+03 7.676140e+03 3.978310e+03 -3.570070e+02 3.016420e+03
+2.534570e+05 2.246360e+05 4.914610e+03 3.097450e+03 8.224900e+03 4.321440e+03 1.713740e+02 2.412360e+03
+2.560290e+05 2.222210e+05 6.841800e+03 1.028500e+03 9.252300e+03 4.387570e+03 2.418140e+03 2.510100e+03
+2.628400e+05 2.225500e+05 6.210250e+03 1.410730e+03 8.538900e+03 4.152580e+03 3.009300e+03 3.219760e+03
+2.616330e+05 2.250650e+05 4.284530e+03 3.357210e+03 7.282170e+03 3.823590e+03 1.402840e+03 3.644670e+03
+2.545910e+05 2.251090e+05 4.693160e+03 3.647740e+03 7.745160e+03 3.686380e+03 4.901610e+02 3.448860e+03
+2.547800e+05 2.235990e+05 6.527380e+03 1.569870e+03 9.438430e+03 3.456580e+03 1.162520e+03 3.252010e+03
+2.606390e+05 2.241070e+05 6.531050e+03 1.633050e+03 9.283720e+03 4.174020e+03 2.089550e+03 2.775750e+03
+2.611080e+05 2.254720e+05 4.968260e+03 3.527850e+03 7.692870e+03 5.137100e+03 2.207390e+03 2.436660e+03
+2.557750e+05 2.237080e+05 4.963450e+03 4.017370e+03 7.701420e+03 5.269650e+03 2.284400e+03 2.842080e+03
+2.573980e+05 2.209470e+05 6.767500e+03 1.645710e+03 9.107070e+03 4.000180e+03 2.548860e+03 3.624770e+03
+2.649240e+05 2.215590e+05 6.471460e+03 1.110330e+03 9.459650e+03 3.108170e+03 1.696970e+03 3.893440e+03
+2.653390e+05 2.257330e+05 4.348800e+03 3.459510e+03 8.475300e+03 4.031240e+03 5.733470e+02 2.910270e+03
+2.568140e+05 2.269950e+05 3.479540e+03 4.949790e+03 7.499910e+03 5.624710e+03 7.516560e+02 2.347710e+03
+2.533160e+05 2.251610e+05 5.147060e+03 3.218430e+03 8.460160e+03 5.869300e+03 2.336320e+03 2.987960e+03
+2.593600e+05 2.231010e+05 5.549120e+03 1.869950e+03 8.740760e+03 4.668940e+03 2.457910e+03 3.758820e+03
+2.620120e+05 2.240160e+05 4.173610e+03 3.004130e+03 8.157040e+03 3.704730e+03 9.879640e+02 3.652750e+03
+2.571760e+05 2.244200e+05 3.517300e+03 4.118750e+03 7.822240e+03 3.718230e+03 3.726490e+01 2.953680e+03
+2.551460e+05 2.233220e+05 4.923980e+03 2.330680e+03 9.095910e+03 3.792400e+03 1.013070e+03 2.711240e+03
+2.605240e+05 2.236510e+05 5.413630e+03 1.146210e+03 8.817170e+03 4.419650e+03 2.446650e+03 2.832050e+03
+2.620980e+05 2.257520e+05 4.262980e+03 2.270970e+03 7.135480e+03 5.067120e+03 2.294680e+03 3.376620e+03
+2.568890e+05 2.253790e+05 3.606460e+03 3.568190e+03 6.552650e+03 4.970270e+03 1.516380e+03 3.662570e+03
+2.539480e+05 2.226310e+05 5.511700e+03 2.066300e+03 7.952660e+03 4.019910e+03 1.513140e+03 3.752630e+03
+2.597990e+05 2.220670e+05 5.873500e+03 6.085840e+02 9.253780e+03 2.870740e+03 1.348240e+03 3.344200e+03
+2.625470e+05 2.249010e+05 4.346080e+03 1.928100e+03 8.590970e+03 3.455460e+03 9.043910e+02 2.379270e+03
+2.561370e+05 2.267610e+05 3.423560e+03 3.379080e+03 7.471150e+03 4.894170e+03 1.153540e+03 2.031410e+03
+2.503260e+05 2.250130e+05 5.519980e+03 2.423970e+03 7.991760e+03 5.117950e+03 2.098790e+03 3.099240e+03
+2.554540e+05 2.229920e+05 6.547950e+03 4.964960e+02 8.751340e+03 3.900560e+03 2.132290e+03 4.076810e+03
+2.612860e+05 2.234890e+05 5.152850e+03 1.501510e+03 8.425610e+03 2.888030e+03 7.761140e+02 3.786360e+03
+2.589690e+05 2.240690e+05 3.832610e+03 3.001980e+03 7.979260e+03 3.182310e+03 5.271600e+01 2.874800e+03
+2.549460e+05 2.220350e+05 5.317880e+03 2.139800e+03 9.103140e+03 3.955610e+03 1.235170e+03 2.394150e+03
+2.586760e+05 2.212050e+05 6.594910e+03 5.053440e+02 9.423360e+03 4.562470e+03 2.913740e+03 2.892350e+03
+2.621250e+05 2.235660e+05 5.116750e+03 1.773600e+03 8.082200e+03 4.776370e+03 2.386390e+03 3.659730e+03
+2.578350e+05 2.259180e+05 3.714300e+03 3.477080e+03 7.205370e+03 4.554610e+03 7.115390e+02 3.878420e+03
+2.536600e+05 2.243710e+05 5.022450e+03 2.592430e+03 8.277200e+03 4.119370e+03 4.865080e+02 3.666740e+03
+2.595030e+05 2.220610e+05 6.589950e+03 6.599360e+02 9.596920e+03 3.598100e+03 1.702490e+03 3.036600e+03
+2.654950e+05 2.228430e+05 5.541850e+03 1.728430e+03 8.459960e+03 4.492000e+03 2.231970e+03 2.430620e+03
+2.609290e+05 2.249960e+05 4.000950e+03 3.745990e+03 6.983790e+03 5.430860e+03 1.855260e+03 2.533380e+03
+2.527160e+05 2.243350e+05 5.086560e+03 3.401150e+03 7.597970e+03 5.196120e+03 1.755720e+03 3.079760e+03
+2.541100e+05 2.231110e+05 6.822190e+03 1.229080e+03 9.164340e+03 3.761230e+03 1.679390e+03 3.584880e+03
+2.599690e+05 2.246930e+05 6.183950e+03 1.538500e+03 9.222080e+03 3.139170e+03 9.499020e+02 3.180800e+03
+2.590780e+05 2.269130e+05 4.388890e+03 3.694820e+03 8.195020e+03 3.933000e+03 4.260800e+02 2.388450e+03
+2.545630e+05 2.247600e+05 5.168440e+03 4.020940e+03 8.450270e+03 4.758910e+03 1.458900e+03 2.286430e+03
+2.580590e+05 2.212170e+05 6.883460e+03 1.649530e+03 9.232780e+03 4.457650e+03 3.057820e+03 3.031950e+03
+2.646670e+05 2.211770e+05 6.218510e+03 1.645730e+03 8.657180e+03 3.663500e+03 2.528280e+03 3.978340e+03
+2.629250e+05 2.243820e+05 4.627500e+03 3.635930e+03 7.892800e+03 3.431320e+03 6.045090e+02 3.901370e+03
+2.547080e+05 2.254480e+05 4.408250e+03 4.461040e+03 8.197170e+03 3.953750e+03 -4.453460e+01 3.154870e+03
+2.537020e+05 2.246350e+05 5.825770e+03 2.577050e+03 9.590050e+03 4.569250e+03 1.460270e+03 2.785170e+03
+2.602060e+05 2.241400e+05 5.387980e+03 1.951160e+03 8.789510e+03 5.131660e+03 2.706380e+03 2.972480e+03
+2.612400e+05 2.247370e+05 3.860810e+03 3.418310e+03 7.414530e+03 5.284520e+03 2.271380e+03 3.183150e+03
+2.561400e+05 2.232520e+05 3.850010e+03 3.957140e+03 7.262650e+03 4.964640e+03 1.499510e+03 3.453130e+03
+2.561160e+05 2.213490e+05 5.594480e+03 2.054400e+03 8.835130e+03 3.662010e+03 1.485510e+03 3.613010e+03

tests/data/extract-6

@@ -0,0 +1,120 @@
+2.517740e+05 2.242410e+05 5.688100e+03 1.915530e+03 9.329220e+03 4.183710e+03 1.212350e+03 2.641790e+03
+2.595670e+05 2.226980e+05 6.207600e+03 6.786720e+02 9.380230e+03 4.575580e+03 2.830610e+03 2.688630e+03
+2.630730e+05 2.233040e+05 4.961640e+03 2.197120e+03 7.687310e+03 4.861860e+03 2.732780e+03 3.008540e+03
+2.576140e+05 2.233230e+05 5.003660e+03 3.525140e+03 7.165310e+03 4.685620e+03 1.715380e+03 3.440480e+03
+2.557800e+05 2.219150e+05 6.357310e+03 2.145290e+03 8.426970e+03 3.775350e+03 1.475390e+03 3.797240e+03
+2.601660e+05 2.230080e+05 6.702590e+03 1.484960e+03 9.288100e+03 3.330830e+03 1.228500e+03 3.214320e+03
+2.612310e+05 2.264260e+05 4.980060e+03 2.982380e+03 8.499630e+03 4.267670e+03 9.940890e+02 2.292890e+03
+2.551170e+05 2.266420e+05 4.584410e+03 4.656440e+03 7.860150e+03 5.317310e+03 1.473600e+03 2.111690e+03
+2.533000e+05 2.235540e+05 6.455090e+03 3.036650e+03 8.869750e+03 4.986310e+03 2.607360e+03 2.839590e+03
+2.610610e+05 2.212630e+05 6.951980e+03 1.500240e+03 9.386100e+03 3.791680e+03 2.677010e+03 3.980630e+03
+2.665030e+05 2.231980e+05 5.189610e+03 2.594560e+03 8.571530e+03 3.175000e+03 9.198400e+02 3.792010e+03
+2.606920e+05 2.251840e+05 3.782480e+03 4.642880e+03 7.662960e+03 3.917790e+03 -2.510970e+02 2.907060e+03
+2.539630e+05 2.250810e+05 5.123530e+03 3.839550e+03 8.669030e+03 4.877820e+03 9.437240e+02 2.527450e+03
+2.565550e+05 2.241690e+05 5.930600e+03 2.298540e+03 8.906710e+03 5.331680e+03 2.549910e+03 3.053560e+03
+2.608890e+05 2.250100e+05 4.681130e+03 2.971870e+03 7.900040e+03 4.874080e+03 2.322430e+03 3.649120e+03
+2.579440e+05 2.249230e+05 3.291140e+03 4.357090e+03 7.131590e+03 4.385560e+03 1.077050e+03 3.664040e+03
+2.550090e+05 2.230180e+05 4.584820e+03 2.864000e+03 8.469490e+03 3.625580e+03 9.855570e+02 3.504230e+03
+2.601140e+05 2.219470e+05 5.676190e+03 1.210340e+03 9.393780e+03 3.390240e+03 1.654020e+03 3.018700e+03
+2.642770e+05 2.244380e+05 4.446620e+03 2.176720e+03 8.142090e+03 4.584880e+03 2.327830e+03 2.615800e+03
+2.592210e+05 2.264710e+05 2.734440e+03 4.182760e+03 6.389550e+03 5.540520e+03 1.958880e+03 2.720120e+03
+2.526500e+05 2.248310e+05 4.163640e+03 2.989990e+03 7.179200e+03 5.213060e+03 1.929550e+03 3.457660e+03
+2.570830e+05 2.220480e+05 5.759040e+03 7.024410e+02 8.566550e+03 3.552020e+03 1.832940e+03 3.956190e+03
+2.631300e+05 2.229670e+05 5.141140e+03 1.166120e+03 8.666960e+03 2.720370e+03 9.713740e+02 3.479730e+03
+2.602360e+05 2.252650e+05 3.425140e+03 3.339080e+03 7.853610e+03 3.674950e+03 5.259080e+02 2.443310e+03
+2.535030e+05 2.245270e+05 4.398130e+03 2.927430e+03 8.110280e+03 4.842470e+03 1.513870e+03 2.467100e+03
+2.561260e+05 2.226930e+05 6.043530e+03 6.562240e+02 8.797560e+03 4.832410e+03 2.832370e+03 3.426140e+03
+2.616770e+05 2.236080e+05 5.830460e+03 1.033910e+03 8.123940e+03 3.980690e+03 1.927960e+03 4.092720e+03
+2.594570e+05 2.255360e+05 4.015570e+03 2.995990e+03 7.135440e+03 3.713550e+03 3.072200e+02 3.849430e+03
+2.533520e+05 2.242160e+05 4.650560e+03 3.196620e+03 8.131280e+03 3.586160e+03 7.083230e+01 3.074180e+03
+2.561240e+05 2.215130e+05 6.100480e+03 8.219800e+02 9.757540e+03 3.474510e+03 1.647520e+03 2.559860e+03
+2.630240e+05 2.215590e+05 5.789960e+03 6.994170e+02 9.129740e+03 4.153080e+03 2.829250e+03 2.677270e+03
+2.617200e+05 2.240150e+05 4.358500e+03 2.645360e+03 7.414110e+03 4.810670e+03 2.225990e+03 3.185990e+03
+2.547560e+05 2.242400e+05 4.857380e+03 3.229680e+03 7.539310e+03 4.769140e+03 1.507130e+03 3.668260e+03
+2.568890e+05 2.226580e+05 6.473420e+03 1.214110e+03 9.010760e+03 3.848730e+03 1.303840e+03 3.778500e+03
+2.642080e+05 2.233160e+05 5.700450e+03 1.116560e+03 9.087610e+03 3.846680e+03 1.293590e+03 2.891560e+03
+2.633100e+05 2.257190e+05 3.936120e+03 3.252360e+03 7.552850e+03 4.897860e+03 1.156630e+03 2.037160e+03
+2.550790e+05 2.250860e+05 4.536450e+03 3.960110e+03 7.454590e+03 5.479070e+03 1.596360e+03 2.190800e+03
+2.544870e+05 2.225080e+05 6.635860e+03 1.758850e+03 8.732970e+03 4.466970e+03 2.650360e+03 3.139310e+03
+2.612410e+05 2.224320e+05 6.702270e+03 1.085130e+03 8.989230e+03 3.112990e+03 1.933560e+03 3.828410e+03
+2.621190e+05 2.255870e+05 4.714950e+03 2.892360e+03 8.107820e+03 2.961310e+03 2.399780e+02 3.273720e+03
+2.549990e+05 2.265140e+05 4.532090e+03 4.126900e+03 8.200130e+03 3.872590e+03 5.608900e+01 2.370580e+03
+2.542890e+05 2.240330e+05 6.538810e+03 2.251440e+03 9.419430e+03 4.564450e+03 2.077810e+03 2.508170e+03
+2.618900e+05 2.219600e+05 6.846090e+03 1.475270e+03 9.125590e+03 4.598290e+03 3.299220e+03 3.475420e+03
+2.645020e+05 2.230850e+05 5.066380e+03 3.270560e+03 7.933170e+03 4.173710e+03 1.908910e+03 3.867460e+03
+2.578890e+05 2.236560e+05 4.201660e+03 4.473640e+03 7.688340e+03 4.161580e+03 6.875790e+02 3.653690e+03
+2.542700e+05 2.231510e+05 5.715140e+03 2.752140e+03 9.273320e+03 3.772950e+03 8.964040e+02 3.256060e+03
+2.582570e+05 2.242170e+05 6.114310e+03 1.856860e+03 9.604320e+03 4.200490e+03 1.764380e+03 2.939220e+03
+2.600200e+05 2.268680e+05 4.237530e+03 3.605880e+03 8.066220e+03 5.430250e+03 2.138580e+03 2.696710e+03
+2.550830e+05 2.259240e+05 3.350310e+03 4.853070e+03 7.045820e+03 5.925200e+03 1.893610e+03 2.897340e+03
+2.544530e+05 2.221270e+05 5.271330e+03 2.491500e+03 8.436680e+03 5.032080e+03 2.436050e+03 3.724590e+03
+2.625880e+05 2.199500e+05 5.994620e+03 7.892740e+02 9.029650e+03 3.515740e+03 1.953570e+03 4.014520e+03
+2.656100e+05 2.233330e+05 4.391410e+03 2.400960e+03 8.146460e+03 3.536960e+03 5.302320e+02 3.133920e+03
+2.574700e+05 2.269770e+05 2.975320e+03 4.633530e+03 7.278560e+03 4.640100e+03 -5.015020e+01 2.024960e+03
+2.506870e+05 2.263310e+05 4.517860e+03 3.183800e+03 8.072600e+03 5.281660e+03 1.605140e+03 2.335140e+03
+2.555630e+05 2.244950e+05 5.551000e+03 1.101300e+03 8.461490e+03 4.725700e+03 2.726670e+03 3.480540e+03
+2.613350e+05 2.246450e+05 4.764680e+03 1.557020e+03 7.833350e+03 3.524810e+03 1.577410e+03 4.038620e+03
+2.602690e+05 2.240080e+05 3.558030e+03 2.987610e+03 7.362440e+03 3.279230e+03 5.624420e+02 3.786550e+03
+2.574350e+05 2.217770e+05 4.972600e+03 2.166880e+03 8.481440e+03 3.328720e+03 1.037130e+03 3.271370e+03
+2.610460e+05 2.215500e+05 5.816180e+03 5.902170e+02 9.120930e+03 3.895400e+03 2.382670e+03 2.824170e+03
+2.627660e+05 2.244730e+05 4.835050e+03 1.785770e+03 7.880760e+03 4.745620e+03 2.443660e+03 3.229550e+03
+2.565090e+05 2.264130e+05 3.758870e+03 3.461200e+03 6.743770e+03 4.928960e+03 1.536620e+03 3.546690e+03
+2.507930e+05 2.243720e+05 5.218490e+03 2.865260e+03 7.803960e+03 4.351090e+03 1.333820e+03 3.680490e+03
+2.563190e+05 2.220660e+05 6.403970e+03 7.323450e+02 9.627760e+03 3.089300e+03 1.516780e+03 3.653690e+03
+2.633430e+05 2.232350e+05 5.200430e+03 1.388580e+03 9.372850e+03 3.371230e+03 1.450390e+03 2.678910e+03
+2.609030e+05 2.251100e+05 3.722580e+03 3.246660e+03 7.876540e+03 4.716810e+03 1.498440e+03 2.116520e+03
+2.544160e+05 2.237690e+05 4.841650e+03 2.956400e+03 8.115920e+03 5.392360e+03 2.142810e+03 2.652320e+03
+2.566980e+05 2.221720e+05 6.471230e+03 9.703960e+02 8.834980e+03 4.816840e+03 2.376630e+03 3.605860e+03
+2.618410e+05 2.235370e+05 5.500740e+03 1.189660e+03 8.365730e+03 4.016470e+03 1.042270e+03 3.821200e+03
+2.595030e+05 2.258400e+05 3.827930e+03 3.088840e+03 7.676140e+03 3.978310e+03 -3.570070e+02 3.016420e+03
+2.534570e+05 2.246360e+05 4.914610e+03 3.097450e+03 8.224900e+03 4.321440e+03 1.713740e+02 2.412360e+03
+2.560290e+05 2.222210e+05 6.841800e+03 1.028500e+03 9.252300e+03 4.387570e+03 2.418140e+03 2.510100e+03
+2.628400e+05 2.225500e+05 6.210250e+03 1.410730e+03 8.538900e+03 4.152580e+03 3.009300e+03 3.219760e+03
+2.616330e+05 2.250650e+05 4.284530e+03 3.357210e+03 7.282170e+03 3.823590e+03 1.402840e+03 3.644670e+03
+2.545910e+05 2.251090e+05 4.693160e+03 3.647740e+03 7.745160e+03 3.686380e+03 4.901610e+02 3.448860e+03
+2.547800e+05 2.235990e+05 6.527380e+03 1.569870e+03 9.438430e+03 3.456580e+03 1.162520e+03 3.252010e+03
+2.606390e+05 2.241070e+05 6.531050e+03 1.633050e+03 9.283720e+03 4.174020e+03 2.089550e+03 2.775750e+03
+2.611080e+05 2.254720e+05 4.968260e+03 3.527850e+03 7.692870e+03 5.137100e+03 2.207390e+03 2.436660e+03
+2.557750e+05 2.237080e+05 4.963450e+03 4.017370e+03 7.701420e+03 5.269650e+03 2.284400e+03 2.842080e+03
+2.573980e+05 2.209470e+05 6.767500e+03 1.645710e+03 9.107070e+03 4.000180e+03 2.548860e+03 3.624770e+03
+2.649240e+05 2.215590e+05 6.471460e+03 1.110330e+03 9.459650e+03 3.108170e+03 1.696970e+03 3.893440e+03
+2.653390e+05 2.257330e+05 4.348800e+03 3.459510e+03 8.475300e+03 4.031240e+03 5.733470e+02 2.910270e+03
+2.568140e+05 2.269950e+05 3.479540e+03 4.949790e+03 7.499910e+03 5.624710e+03 7.516560e+02 2.347710e+03
+2.533160e+05 2.251610e+05 5.147060e+03 3.218430e+03 8.460160e+03 5.869300e+03 2.336320e+03 2.987960e+03
+2.593600e+05 2.231010e+05 5.549120e+03 1.869950e+03 8.740760e+03 4.668940e+03 2.457910e+03 3.758820e+03
+2.620120e+05 2.240160e+05 4.173610e+03 3.004130e+03 8.157040e+03 3.704730e+03 9.879640e+02 3.652750e+03
+2.571760e+05 2.244200e+05 3.517300e+03 4.118750e+03 7.822240e+03 3.718230e+03 3.726490e+01 2.953680e+03
+2.551460e+05 2.233220e+05 4.923980e+03 2.330680e+03 9.095910e+03 3.792400e+03 1.013070e+03 2.711240e+03
+2.605240e+05 2.236510e+05 5.413630e+03 1.146210e+03 8.817170e+03 4.419650e+03 2.446650e+03 2.832050e+03
+2.620980e+05 2.257520e+05 4.262980e+03 2.270970e+03 7.135480e+03 5.067120e+03 2.294680e+03 3.376620e+03
+2.568890e+05 2.253790e+05 3.606460e+03 3.568190e+03 6.552650e+03 4.970270e+03 1.516380e+03 3.662570e+03
+2.539480e+05 2.226310e+05 5.511700e+03 2.066300e+03 7.952660e+03 4.019910e+03 1.513140e+03 3.752630e+03
+2.597990e+05 2.220670e+05 5.873500e+03 6.085840e+02 9.253780e+03 2.870740e+03 1.348240e+03 3.344200e+03
+2.625470e+05 2.249010e+05 4.346080e+03 1.928100e+03 8.590970e+03 3.455460e+03 9.043910e+02 2.379270e+03
+2.561370e+05 2.267610e+05 3.423560e+03 3.379080e+03 7.471150e+03 4.894170e+03 1.153540e+03 2.031410e+03
+2.503260e+05 2.250130e+05 5.519980e+03 2.423970e+03 7.991760e+03 5.117950e+03 2.098790e+03 3.099240e+03
+2.554540e+05 2.229920e+05 6.547950e+03 4.964960e+02 8.751340e+03 3.900560e+03 2.132290e+03 4.076810e+03
+2.612860e+05 2.234890e+05 5.152850e+03 1.501510e+03 8.425610e+03 2.888030e+03 7.761140e+02 3.786360e+03
+2.589690e+05 2.240690e+05 3.832610e+03 3.001980e+03 7.979260e+03 3.182310e+03 5.271600e+01 2.874800e+03
+2.549460e+05 2.220350e+05 5.317880e+03 2.139800e+03 9.103140e+03 3.955610e+03 1.235170e+03 2.394150e+03
+2.586760e+05 2.212050e+05 6.594910e+03 5.053440e+02 9.423360e+03 4.562470e+03 2.913740e+03 2.892350e+03
+2.621250e+05 2.235660e+05 5.116750e+03 1.773600e+03 8.082200e+03 4.776370e+03 2.386390e+03 3.659730e+03
+2.578350e+05 2.259180e+05 3.714300e+03 3.477080e+03 7.205370e+03 4.554610e+03 7.115390e+02 3.878420e+03
+2.536600e+05 2.243710e+05 5.022450e+03 2.592430e+03 8.277200e+03 4.119370e+03 4.865080e+02 3.666740e+03
+2.595030e+05 2.220610e+05 6.589950e+03 6.599360e+02 9.596920e+03 3.598100e+03 1.702490e+03 3.036600e+03
+2.654950e+05 2.228430e+05 5.541850e+03 1.728430e+03 8.459960e+03 4.492000e+03 2.231970e+03 2.430620e+03
+2.609290e+05 2.249960e+05 4.000950e+03 3.745990e+03 6.983790e+03 5.430860e+03 1.855260e+03 2.533380e+03
+2.527160e+05 2.243350e+05 5.086560e+03 3.401150e+03 7.597970e+03 5.196120e+03 1.755720e+03 3.079760e+03
+2.541100e+05 2.231110e+05 6.822190e+03 1.229080e+03 9.164340e+03 3.761230e+03 1.679390e+03 3.584880e+03
+2.599690e+05 2.246930e+05 6.183950e+03 1.538500e+03 9.222080e+03 3.139170e+03 9.499020e+02 3.180800e+03
+2.590780e+05 2.269130e+05 4.388890e+03 3.694820e+03 8.195020e+03 3.933000e+03 4.260800e+02 2.388450e+03
+2.545630e+05 2.247600e+05 5.168440e+03 4.020940e+03 8.450270e+03 4.758910e+03 1.458900e+03 2.286430e+03
+2.580590e+05 2.212170e+05 6.883460e+03 1.649530e+03 9.232780e+03 4.457650e+03 3.057820e+03 3.031950e+03
+2.646670e+05 2.211770e+05 6.218510e+03 1.645730e+03 8.657180e+03 3.663500e+03 2.528280e+03 3.978340e+03
+2.629250e+05 2.243820e+05 4.627500e+03 3.635930e+03 7.892800e+03 3.431320e+03 6.045090e+02 3.901370e+03
+2.547080e+05 2.254480e+05 4.408250e+03 4.461040e+03 8.197170e+03 3.953750e+03 -4.453460e+01 3.154870e+03
+2.537020e+05 2.246350e+05 5.825770e+03 2.577050e+03 9.590050e+03 4.569250e+03 1.460270e+03 2.785170e+03
+2.602060e+05 2.241400e+05 5.387980e+03 1.951160e+03 8.789510e+03 5.131660e+03 2.706380e+03 2.972480e+03
+2.612400e+05 2.247370e+05 3.860810e+03 3.418310e+03 7.414530e+03 5.284520e+03 2.271380e+03 3.183150e+03
+2.561400e+05 2.232520e+05 3.850010e+03 3.957140e+03 7.262650e+03 4.964640e+03 1.499510e+03 3.453130e+03
+2.561160e+05 2.213490e+05 5.594480e+03 2.054400e+03 8.835130e+03 3.662010e+03 1.485510e+03 3.613010e+03

tests/data/extract-7

@@ -0,0 +1,124 @@
+# path: /newton/prep
+# layout: float32_8
+# start: 1332496830000000
+# end: 1332496830999000
+1332496830000000 2.517740e+05 2.242410e+05 5.688100e+03 1.915530e+03 9.329220e+03 4.183710e+03 1.212350e+03 2.641790e+03
+1332496830008333 2.595670e+05 2.226980e+05 6.207600e+03 6.786720e+02 9.380230e+03 4.575580e+03 2.830610e+03 2.688630e+03
+1332496830016667 2.630730e+05 2.233040e+05 4.961640e+03 2.197120e+03 7.687310e+03 4.861860e+03 2.732780e+03 3.008540e+03
+1332496830025000 2.576140e+05 2.233230e+05 5.003660e+03 3.525140e+03 7.165310e+03 4.685620e+03 1.715380e+03 3.440480e+03
+1332496830033333 2.557800e+05 2.219150e+05 6.357310e+03 2.145290e+03 8.426970e+03 3.775350e+03 1.475390e+03 3.797240e+03
+1332496830041667 2.601660e+05 2.230080e+05 6.702590e+03 1.484960e+03 9.288100e+03 3.330830e+03 1.228500e+03 3.214320e+03
+1332496830050000 2.612310e+05 2.264260e+05 4.980060e+03 2.982380e+03 8.499630e+03 4.267670e+03 9.940890e+02 2.292890e+03
+1332496830058333 2.551170e+05 2.266420e+05 4.584410e+03 4.656440e+03 7.860150e+03 5.317310e+03 1.473600e+03 2.111690e+03
+1332496830066667 2.533000e+05 2.235540e+05 6.455090e+03 3.036650e+03 8.869750e+03 4.986310e+03 2.607360e+03 2.839590e+03
+1332496830075000 2.610610e+05 2.212630e+05 6.951980e+03 1.500240e+03 9.386100e+03 3.791680e+03 2.677010e+03 3.980630e+03
+1332496830083333 2.665030e+05 2.231980e+05 5.189610e+03 2.594560e+03 8.571530e+03 3.175000e+03 9.198400e+02 3.792010e+03
+1332496830091667 2.606920e+05 2.251840e+05 3.782480e+03 4.642880e+03 7.662960e+03 3.917790e+03 -2.510970e+02 2.907060e+03
+1332496830100000 2.539630e+05 2.250810e+05 5.123530e+03 3.839550e+03 8.669030e+03 4.877820e+03 9.437240e+02 2.527450e+03
+1332496830108333 2.565550e+05 2.241690e+05 5.930600e+03 2.298540e+03 8.906710e+03 5.331680e+03 2.549910e+03 3.053560e+03
+1332496830116667 2.608890e+05 2.250100e+05 4.681130e+03 2.971870e+03 7.900040e+03 4.874080e+03 2.322430e+03 3.649120e+03
+1332496830125000 2.579440e+05 2.249230e+05 3.291140e+03 4.357090e+03 7.131590e+03 4.385560e+03 1.077050e+03 3.664040e+03
+1332496830133333 2.550090e+05 2.230180e+05 4.584820e+03 2.864000e+03 8.469490e+03 3.625580e+03 9.855570e+02 3.504230e+03
+1332496830141667 2.601140e+05 2.219470e+05 5.676190e+03 1.210340e+03 9.393780e+03 3.390240e+03 1.654020e+03 3.018700e+03
+1332496830150000 2.642770e+05 2.244380e+05 4.446620e+03 2.176720e+03 8.142090e+03 4.584880e+03 2.327830e+03 2.615800e+03
+1332496830158333 2.592210e+05 2.264710e+05 2.734440e+03 4.182760e+03 6.389550e+03 5.540520e+03 1.958880e+03 2.720120e+03
+1332496830166667 2.526500e+05 2.248310e+05 4.163640e+03 2.989990e+03 7.179200e+03 5.213060e+03 1.929550e+03 3.457660e+03
+1332496830175000 2.570830e+05 2.220480e+05 5.759040e+03 7.024410e+02 8.566550e+03 3.552020e+03 1.832940e+03 3.956190e+03
+1332496830183333 2.631300e+05 2.229670e+05 5.141140e+03 1.166120e+03 8.666960e+03 2.720370e+03 9.713740e+02 3.479730e+03
+1332496830191667 2.602360e+05 2.252650e+05 3.425140e+03 3.339080e+03 7.853610e+03 3.674950e+03 5.259080e+02 2.443310e+03
+1332496830200000 2.535030e+05 2.245270e+05 4.398130e+03 2.927430e+03 8.110280e+03 4.842470e+03 1.513870e+03 2.467100e+03
+1332496830208333 2.561260e+05 2.226930e+05 6.043530e+03 6.562240e+02 8.797560e+03 4.832410e+03 2.832370e+03 3.426140e+03
+1332496830216667 2.616770e+05 2.236080e+05 5.830460e+03 1.033910e+03 8.123940e+03 3.980690e+03 1.927960e+03 4.092720e+03
+1332496830225000 2.594570e+05 2.255360e+05 4.015570e+03 2.995990e+03 7.135440e+03 3.713550e+03 3.072200e+02 3.849430e+03
+1332496830233333 2.533520e+05 2.242160e+05 4.650560e+03 3.196620e+03 8.131280e+03 3.586160e+03 7.083230e+01 3.074180e+03
+1332496830241667 2.561240e+05 2.215130e+05 6.100480e+03 8.219800e+02 9.757540e+03 3.474510e+03 1.647520e+03 2.559860e+03
+1332496830250000 2.630240e+05 2.215590e+05 5.789960e+03 6.994170e+02 9.129740e+03 4.153080e+03 2.829250e+03 2.677270e+03
+1332496830258333 2.617200e+05 2.240150e+05 4.358500e+03 2.645360e+03 7.414110e+03 4.810670e+03 2.225990e+03 3.185990e+03
+1332496830266667 2.547560e+05 2.242400e+05 4.857380e+03 3.229680e+03 7.539310e+03 4.769140e+03 1.507130e+03 3.668260e+03
+1332496830275000 2.568890e+05 2.226580e+05 6.473420e+03 1.214110e+03 9.010760e+03 3.848730e+03 1.303840e+03 3.778500e+03
+1332496830283333 2.642080e+05 2.233160e+05 5.700450e+03 1.116560e+03 9.087610e+03 3.846680e+03 1.293590e+03 2.891560e+03
+1332496830291667 2.633100e+05 2.257190e+05 3.936120e+03 3.252360e+03 7.552850e+03 4.897860e+03 1.156630e+03 2.037160e+03
+1332496830300000 2.550790e+05 2.250860e+05 4.536450e+03 3.960110e+03 7.454590e+03 5.479070e+03 1.596360e+03 2.190800e+03
+1332496830308333 2.544870e+05 2.225080e+05 6.635860e+03 1.758850e+03 8.732970e+03 4.466970e+03 2.650360e+03 3.139310e+03
+1332496830316667 2.612410e+05 2.224320e+05 6.702270e+03 1.085130e+03 8.989230e+03 3.112990e+03 1.933560e+03 3.828410e+03
+1332496830325000 2.621190e+05 2.255870e+05 4.714950e+03 2.892360e+03 8.107820e+03 2.961310e+03 2.399780e+02 3.273720e+03
+1332496830333333 2.549990e+05 2.265140e+05 4.532090e+03 4.126900e+03 8.200130e+03 3.872590e+03 5.608900e+01 2.370580e+03
+1332496830341667 2.542890e+05 2.240330e+05 6.538810e+03 2.251440e+03 9.419430e+03 4.564450e+03 2.077810e+03 2.508170e+03
+1332496830350000 2.618900e+05 2.219600e+05 6.846090e+03 1.475270e+03 9.125590e+03 4.598290e+03 3.299220e+03 3.475420e+03
+1332496830358333 2.645020e+05 2.230850e+05 5.066380e+03 3.270560e+03 7.933170e+03 4.173710e+03 1.908910e+03 3.867460e+03
+1332496830366667 2.578890e+05 2.236560e+05 4.201660e+03 4.473640e+03 7.688340e+03 4.161580e+03 6.875790e+02 3.653690e+03
+1332496830375000 2.542700e+05 2.231510e+05 5.715140e+03 2.752140e+03 9.273320e+03 3.772950e+03 8.964040e+02 3.256060e+03
+1332496830383333 2.582570e+05 2.242170e+05 6.114310e+03 1.856860e+03 9.604320e+03 4.200490e+03 1.764380e+03 2.939220e+03
+1332496830391667 2.600200e+05 2.268680e+05 4.237530e+03 3.605880e+03 8.066220e+03 5.430250e+03 2.138580e+03 2.696710e+03
+1332496830400000 2.550830e+05 2.259240e+05 3.350310e+03 4.853070e+03 7.045820e+03 5.925200e+03 1.893610e+03 2.897340e+03
+1332496830408333 2.544530e+05 2.221270e+05 5.271330e+03 2.491500e+03 8.436680e+03 5.032080e+03 2.436050e+03 3.724590e+03
+1332496830416667 2.625880e+05 2.199500e+05 5.994620e+03 7.892740e+02 9.029650e+03 3.515740e+03 1.953570e+03 4.014520e+03
+1332496830425000 2.656100e+05 2.233330e+05 4.391410e+03 2.400960e+03 8.146460e+03 3.536960e+03 5.302320e+02 3.133920e+03
+1332496830433333 2.574700e+05 2.269770e+05 2.975320e+03 4.633530e+03 7.278560e+03 4.640100e+03 -5.015020e+01 2.024960e+03
+1332496830441667 2.506870e+05 2.263310e+05 4.517860e+03 3.183800e+03 8.072600e+03 5.281660e+03 1.605140e+03 2.335140e+03
+1332496830450000 2.555630e+05 2.244950e+05 5.551000e+03 1.101300e+03 8.461490e+03 4.725700e+03 2.726670e+03 3.480540e+03
+1332496830458333 2.613350e+05 2.246450e+05 4.764680e+03 1.557020e+03 7.833350e+03 3.524810e+03 1.577410e+03 4.038620e+03
+1332496830466667 2.602690e+05 2.240080e+05 3.558030e+03 2.987610e+03 7.362440e+03 3.279230e+03 5.624420e+02 3.786550e+03
+1332496830475000 2.574350e+05 2.217770e+05 4.972600e+03 2.166880e+03 8.481440e+03 3.328720e+03 1.037130e+03 3.271370e+03
+1332496830483333 2.610460e+05 2.215500e+05 5.816180e+03 5.902170e+02 9.120930e+03 3.895400e+03 2.382670e+03 2.824170e+03
+1332496830491667 2.627660e+05 2.244730e+05 4.835050e+03 1.785770e+03 7.880760e+03 4.745620e+03 2.443660e+03 3.229550e+03
+1332496830500000 2.565090e+05 2.264130e+05 3.758870e+03 3.461200e+03 6.743770e+03 4.928960e+03 1.536620e+03 3.546690e+03
+1332496830508333 2.507930e+05 2.243720e+05 5.218490e+03 2.865260e+03 7.803960e+03 4.351090e+03 1.333820e+03 3.680490e+03
+1332496830516667 2.563190e+05 2.220660e+05 6.403970e+03 7.323450e+02 9.627760e+03 3.089300e+03 1.516780e+03 3.653690e+03
+1332496830525000 2.633430e+05 2.232350e+05 5.200430e+03 1.388580e+03 9.372850e+03 3.371230e+03 1.450390e+03 2.678910e+03
+1332496830533333 2.609030e+05 2.251100e+05 3.722580e+03 3.246660e+03 7.876540e+03 4.716810e+03 1.498440e+03 2.116520e+03
+1332496830541667 2.544160e+05 2.237690e+05 4.841650e+03 2.956400e+03 8.115920e+03 5.392360e+03 2.142810e+03 2.652320e+03
+1332496830550000 2.566980e+05 2.221720e+05 6.471230e+03 9.703960e+02 8.834980e+03 4.816840e+03 2.376630e+03 3.605860e+03
+1332496830558333 2.618410e+05 2.235370e+05 5.500740e+03 1.189660e+03 8.365730e+03 4.016470e+03 1.042270e+03 3.821200e+03
+1332496830566667 2.595030e+05 2.258400e+05 3.827930e+03 3.088840e+03 7.676140e+03 3.978310e+03 -3.570070e+02 3.016420e+03
+1332496830575000 2.534570e+05 2.246360e+05 4.914610e+03 3.097450e+03 8.224900e+03 4.321440e+03 1.713740e+02 2.412360e+03
+1332496830583333 2.560290e+05 2.222210e+05 6.841800e+03 1.028500e+03 9.252300e+03 4.387570e+03 2.418140e+03 2.510100e+03
+1332496830591667 2.628400e+05 2.225500e+05 6.210250e+03 1.410730e+03 8.538900e+03 4.152580e+03 3.009300e+03 3.219760e+03
+1332496830600000 2.616330e+05 2.250650e+05 4.284530e+03 3.357210e+03 7.282170e+03 3.823590e+03 1.402840e+03 3.644670e+03
+1332496830608333 2.545910e+05 2.251090e+05 4.693160e+03 3.647740e+03 7.745160e+03 3.686380e+03 4.901610e+02 3.448860e+03
+1332496830616667 2.547800e+05 2.235990e+05 6.527380e+03 1.569870e+03 9.438430e+03 3.456580e+03 1.162520e+03 3.252010e+03
+1332496830625000 2.606390e+05 2.241070e+05 6.531050e+03 1.633050e+03 9.283720e+03 4.174020e+03 2.089550e+03 2.775750e+03
+1332496830633333 2.611080e+05 2.254720e+05 4.968260e+03 3.527850e+03 7.692870e+03 5.137100e+03 2.207390e+03 2.436660e+03
+1332496830641667 2.557750e+05 2.237080e+05 4.963450e+03 4.017370e+03 7.701420e+03 5.269650e+03 2.284400e+03 2.842080e+03
+1332496830650000 2.573980e+05 2.209470e+05 6.767500e+03 1.645710e+03 9.107070e+03 4.000180e+03 2.548860e+03 3.624770e+03
+1332496830658333 2.649240e+05 2.215590e+05 6.471460e+03 1.110330e+03 9.459650e+03 3.108170e+03 1.696970e+03 3.893440e+03
+1332496830666667 2.653390e+05 2.257330e+05 4.348800e+03 3.459510e+03 8.475300e+03 4.031240e+03 5.733470e+02 2.910270e+03
+1332496830675000 2.568140e+05 2.269950e+05 3.479540e+03 4.949790e+03 7.499910e+03 5.624710e+03 7.516560e+02 2.347710e+03
+1332496830683333 2.533160e+05 2.251610e+05 5.147060e+03 3.218430e+03 8.460160e+03 5.869300e+03 2.336320e+03 2.987960e+03
+1332496830691667 2.593600e+05 2.231010e+05 5.549120e+03 1.869950e+03 8.740760e+03 4.668940e+03 2.457910e+03 3.758820e+03
+1332496830700000 2.620120e+05 2.240160e+05 4.173610e+03 3.004130e+03 8.157040e+03 3.704730e+03 9.879640e+02 3.652750e+03
+1332496830708333 2.571760e+05 2.244200e+05 3.517300e+03 4.118750e+03 7.822240e+03 3.718230e+03 3.726490e+01 2.953680e+03
+1332496830716667 2.551460e+05 2.233220e+05 4.923980e+03 2.330680e+03 9.095910e+03 3.792400e+03 1.013070e+03 2.711240e+03
+1332496830725000 2.605240e+05 2.236510e+05 5.413630e+03 1.146210e+03 8.817170e+03 4.419650e+03 2.446650e+03 2.832050e+03
+1332496830733333 2.620980e+05 2.257520e+05 4.262980e+03 2.270970e+03 7.135480e+03 5.067120e+03 2.294680e+03 3.376620e+03
+1332496830741667 2.568890e+05 2.253790e+05 3.606460e+03 3.568190e+03 6.552650e+03 4.970270e+03 1.516380e+03 3.662570e+03
+1332496830750000 2.539480e+05 2.226310e+05 5.511700e+03 2.066300e+03 7.952660e+03 4.019910e+03 1.513140e+03 3.752630e+03
+1332496830758333 2.597990e+05 2.220670e+05 5.873500e+03 6.085840e+02 9.253780e+03 2.870740e+03 1.348240e+03 3.344200e+03
+1332496830766667 2.625470e+05 2.249010e+05 4.346080e+03 1.928100e+03 8.590970e+03 3.455460e+03 9.043910e+02 2.379270e+03
+1332496830775000 2.561370e+05 2.267610e+05 3.423560e+03 3.379080e+03 7.471150e+03 4.894170e+03 1.153540e+03 2.031410e+03
+1332496830783333 2.503260e+05 2.250130e+05 5.519980e+03 2.423970e+03 7.991760e+03 5.117950e+03 2.098790e+03 3.099240e+03
+1332496830791667 2.554540e+05 2.229920e+05 6.547950e+03 4.964960e+02 8.751340e+03 3.900560e+03 2.132290e+03 4.076810e+03
+1332496830800000 2.612860e+05 2.234890e+05 5.152850e+03 1.501510e+03 8.425610e+03 2.888030e+03 7.761140e+02 3.786360e+03
+1332496830808333 2.589690e+05 2.240690e+05 3.832610e+03 3.001980e+03 7.979260e+03 3.182310e+03 5.271600e+01 2.874800e+03
+1332496830816667 2.549460e+05 2.220350e+05 5.317880e+03 2.139800e+03 9.103140e+03 3.955610e+03 1.235170e+03 2.394150e+03
+1332496830825000 2.586760e+05 2.212050e+05 6.594910e+03 5.053440e+02 9.423360e+03 4.562470e+03 2.913740e+03 2.892350e+03
+1332496830833333 2.621250e+05 2.235660e+05 5.116750e+03 1.773600e+03 8.082200e+03 4.776370e+03 2.386390e+03 3.659730e+03
+1332496830841667 2.578350e+05 2.259180e+05 3.714300e+03 3.477080e+03 7.205370e+03 4.554610e+03 7.115390e+02 3.878420e+03
+1332496830850000 2.536600e+05 2.243710e+05 5.022450e+03 2.592430e+03 8.277200e+03 4.119370e+03 4.865080e+02 3.666740e+03
+1332496830858333 2.595030e+05 2.220610e+05 6.589950e+03 6.599360e+02 9.596920e+03 3.598100e+03 1.702490e+03 3.036600e+03
+1332496830866667 2.654950e+05 2.228430e+05 5.541850e+03 1.728430e+03 8.459960e+03 4.492000e+03 2.231970e+03 2.430620e+03
+1332496830875000 2.609290e+05 2.249960e+05 4.000950e+03 3.745990e+03 6.983790e+03 5.430860e+03 1.855260e+03 2.533380e+03
+1332496830883333 2.527160e+05 2.243350e+05 5.086560e+03 3.401150e+03 7.597970e+03 5.196120e+03 1.755720e+03 3.079760e+03
+1332496830891667 2.541100e+05 2.231110e+05 6.822190e+03 1.229080e+03 9.164340e+03 3.761230e+03 1.679390e+03 3.584880e+03
+1332496830900000 2.599690e+05 2.246930e+05 6.183950e+03 1.538500e+03 9.222080e+03 3.139170e+03 9.499020e+02 3.180800e+03
+1332496830908333 2.590780e+05 2.269130e+05 4.388890e+03 3.694820e+03 8.195020e+03 3.933000e+03 4.260800e+02 2.388450e+03
+1332496830916667 2.545630e+05 2.247600e+05 5.168440e+03 4.020940e+03 8.450270e+03 4.758910e+03 1.458900e+03 2.286430e+03
+1332496830925000 2.580590e+05 2.212170e+05 6.883460e+03 1.649530e+03 9.232780e+03 4.457650e+03 3.057820e+03 3.031950e+03
+1332496830933333 2.646670e+05 2.211770e+05 6.218510e+03 1.645730e+03 8.657180e+03 3.663500e+03 2.528280e+03 3.978340e+03
+1332496830941667 2.629250e+05 2.243820e+05 4.627500e+03 3.635930e+03 7.892800e+03 3.431320e+03 6.045090e+02 3.901370e+03
+1332496830950000 2.547080e+05 2.254480e+05 4.408250e+03 4.461040e+03 8.197170e+03 3.953750e+03 -4.453460e+01 3.154870e+03
+1332496830958333 2.537020e+05 2.246350e+05 5.825770e+03 2.577050e+03 9.590050e+03 4.569250e+03 1.460270e+03 2.785170e+03
+1332496830966667 2.602060e+05 2.241400e+05 5.387980e+03 1.951160e+03 8.789510e+03 5.131660e+03 2.706380e+03 2.972480e+03
+1332496830975000 2.612400e+05 2.247370e+05 3.860810e+03 3.418310e+03 7.414530e+03 5.284520e+03 2.271380e+03 3.183150e+03
+1332496830983333 2.561400e+05 2.232520e+05 3.850010e+03 3.957140e+03 7.262650e+03 4.964640e+03 1.499510e+03 3.453130e+03
+1332496830991667 2.561160e+05 2.213490e+05 5.594480e+03 2.054400e+03 8.835130e+03 3.662010e+03 1.485510e+03 3.613010e+03

tests/data/extract-8

@@ -0,0 +1,28 @@
+# interval-start 1332496919900000
+1332496919900000 2.523050e+05 2.254020e+05 4.779410e+03 3.638030e+03 8.138070e+03 4.334460e+03 1.083780e+03 3.743730e+03
+1332496919908333 2.551190e+05 2.237870e+05 5.965640e+03 2.076350e+03 9.468790e+03 3.693880e+03 1.247860e+03 3.393680e+03
+1332496919916667 2.616370e+05 2.247980e+05 4.848970e+03 2.315620e+03 9.323300e+03 4.225460e+03 1.805780e+03 2.593050e+03
+1332496919925000 2.606460e+05 2.251300e+05 3.061360e+03 3.951840e+03 7.662910e+03 5.341410e+03 1.986520e+03 2.276780e+03
+1332496919933333 2.559710e+05 2.235030e+05 4.096030e+03 3.296970e+03 7.827080e+03 5.452120e+03 2.492520e+03 2.929450e+03
+1332496919941667 2.579260e+05 2.217080e+05 5.472320e+03 1.555700e+03 8.495760e+03 4.491140e+03 2.379780e+03 3.741710e+03
+1332496919950000 2.610180e+05 2.242350e+05 4.669770e+03 1.876190e+03 8.366680e+03 3.677510e+03 9.021690e+02 3.549040e+03
+1332496919958333 2.569150e+05 2.274650e+05 2.785070e+03 3.751930e+03 7.440320e+03 3.964860e+03 -3.227860e+02 2.460890e+03
+1332496919966667 2.509510e+05 2.262000e+05 3.772710e+03 3.131950e+03 8.159860e+03 4.539860e+03 7.375190e+02 2.126750e+03
+1332496919975000 2.556710e+05 2.223720e+05 5.826200e+03 8.715560e+02 9.120240e+03 4.545110e+03 2.804310e+03 2.721000e+03
+1332496919983333 2.649730e+05 2.214860e+05 5.839130e+03 4.659180e+02 8.628300e+03 3.934870e+03 2.972490e+03 3.773730e+03
+1332496919991667 2.652170e+05 2.233920e+05 3.718770e+03 2.834970e+03 7.209900e+03 3.460260e+03 1.324930e+03 4.075960e+03
+# interval-end 1332496919991668
+# interval-start 1332496920000000
+1332496920000000 2.564370e+05 2.244300e+05 4.011610e+03 3.475340e+03 7.495890e+03 3.388940e+03 2.613970e+02 3.731260e+03
+1332496920008333 2.539630e+05 2.241670e+05 5.621070e+03 1.548010e+03 9.165170e+03 3.522930e+03 1.058930e+03 2.996960e+03
+1332496920016667 2.585080e+05 2.249300e+05 6.011400e+03 8.188660e+02 9.039950e+03 4.482440e+03 2.490390e+03 2.679340e+03
+1332496920025000 2.596270e+05 2.260220e+05 4.474500e+03 2.423020e+03 7.414190e+03 5.071970e+03 2.439380e+03 2.962960e+03
+1332496920033333 2.551870e+05 2.246320e+05 4.738570e+03 3.398040e+03 7.395120e+03 4.726450e+03 1.839030e+03 3.393530e+03
+1332496920041667 2.571020e+05 2.216230e+05 6.144130e+03 1.441090e+03 8.756480e+03 3.495320e+03 1.869940e+03 3.752530e+03
+1332496920050000 2.636530e+05 2.217700e+05 6.221770e+03 7.389620e+02 9.547600e+03 2.666820e+03 1.462660e+03 3.332570e+03
+1332496920058333 2.636130e+05 2.252560e+05 4.477120e+03 2.437450e+03 8.510210e+03 3.855630e+03 9.594420e+02 2.387180e+03
+1332496920066667 2.553500e+05 2.262640e+05 4.283720e+03 3.923940e+03 7.912470e+03 5.466520e+03 1.284990e+03 2.093720e+03
+1332496920075000 2.527270e+05 2.246090e+05 5.851930e+03 2.491980e+03 8.540630e+03 5.623050e+03 2.339780e+03 3.007140e+03
+1332496920083333 2.584750e+05 2.235780e+05 5.924870e+03 1.394480e+03 8.779620e+03 4.544180e+03 2.132030e+03 3.849760e+03
+1332496920091667 2.615630e+05 2.246090e+05 4.336140e+03 2.455750e+03 8.055380e+03 3.469110e+03 6.278730e+02 3.664200e+03
+# interval-end 1332496920100000