Attempt at fixing up more Unicode issues with metadata.

We were still calculating the maximum number of rows correctly,
so the extra data was really extra and would get re-written to the
beginning of the subsequent file.

The only case in which this would lead to database issues is if the
very last file was lengthened incorrectly, and the "nrows" calculation
would therefore be wrong when the database was reopened.  Still, even
in that case, it should just leave a small gap in the data, not cause
any errors.

README.txt

@@ -10,6 +10,9 @@ Prerequisites:
   sudo apt-get install python-cherrypy3 python-decorator python-simplejson
   sudo apt-get install python-requests python-dateutil python-tz python-psutil
 
+  # Other dependencies (required by some modules)
+  sudo apt-get install python-numpy
+
   # Tools for running tests
   sudo apt-get install python-nose python-coverage
 

docs/design.md

@@ -389,3 +389,52 @@ Possible solutions:
     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 `cherrpy.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.

extras/fix-oversize-files.py

@@ -0,0 +1,50 @@
+#!/usr/bin/python
+
+import os
+import sys
+import cPickle as 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)

nilmdb/client/client.py

@@ -6,7 +6,6 @@ import nilmdb.utils
 import nilmdb.client.httpclient
 from nilmdb.client.errors import ClientError
 
-import re
 import time
 import simplejson as json
 import contextlib
@@ -66,12 +65,8 @@ class Client(object):
             params["layout"] = layout
         if extended:
             params["extended"] = 1
-        def sort_streams_nicely(x):
-            """Human-friendly sort (/stream/2 before /stream/10)"""
-            num = lambda t: int(t) if t.isdigit() else t
-            key = lambda k: [ num(c) for c in re.split('([0-9]+)', k[0]) ]
-            return sorted(x, key = key)
-        return sort_streams_nicely(self.http.get("stream/list", params))
+        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):
         params = { "path": path }
@@ -122,15 +117,19 @@ class Client(object):
             params["start"] = timestamp_to_string(start)
         if end is not None:
             params["end"] = timestamp_to_string(end)
-        return self.http.post("stream/remove", params)
+        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 be provided
-        as single lines, and is aggregated and sent to the server in larger
-        chunks as necessary.  Data lines must match the database layout for
-        the given path, and end with a newline.
+        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:
@@ -142,15 +141,17 @@ class Client(object):
         This may make multiple requests to the server, if the data is
         large enough or enough time has passed between insertions.
         """
-        ctx = StreamInserter(self.http, path, start, end)
+        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.  See stream_insert_context for details on the
-        'start' and 'end' parameters."""
+        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, basestring):
                 ctx.insert(data)
@@ -159,11 +160,28 @@ class Client(object):
                     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, binary = binary)
+
     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
+        If 'diffpath' is not None, yields only interval ranges that are
         present in 'path' but not in 'diffpath'.
         """
         params = {
@@ -184,16 +202,16 @@ class Client(object):
         lines of ASCII-formatted data that matches the database
         layout for the given path.
 
-        Specify count = True to return a count of matching data points
+        If 'count' is True, return a count of matching data points
         rather than the actual data.  The output format is unchanged.
 
-        Specify markup = True to include comments in the returned data
+        If 'markup' is True, include comments in the returned data
         that indicate interval starts and ends.
 
-        Specify binary = True to return chunks of raw binary data,
-        rather than lines of ASCII-formatted data.  Raw binary data
-        is always little-endian and matches the database types
-        (including a uint64 timestamp).
+        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,
@@ -257,13 +275,13 @@ class StreamInserter(object):
     _max_data = 2 * 1024 * 1024
     _max_data_after_send = 64 * 1024
 
-    def __init__(self, http, path, start = None, end = None):
-        """'http' is the httpclient object.  'path' is the database
+    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."""
+        contiguous interval and may be None."""
         self.last_response = None
 
-        self._http = http
+        self._client = client
         self._path = path
 
         # Start and end for the overall contiguous interval we're
@@ -276,6 +294,15 @@ class StreamInserter(object):
         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'."""
@@ -422,7 +449,7 @@ class StreamInserter(object):
             self._interval_start = end_ts
 
         # Double check endpoints
-        if start_ts is None or end_ts is None:
+        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)
@@ -431,9 +458,7 @@ class StreamInserter(object):
             raise ClientError("have data to send, but no start/end times")
 
         # Send it
-        params = { "path": self._path,
-                   "start": timestamp_to_string(start_ts),
-                   "end": timestamp_to_string(end_ts) }
-        self.last_response = self._http.put("stream/insert", block, params)
+        self.last_response = self._client.stream_insert_block(
+            self._path, block, start_ts, end_ts, binary = False)
 
         return

nilmdb/client/httpclient.py

@@ -105,9 +105,13 @@ class HTTPClient(object):
         else:
             return self._req("POST", url, None, params)
 
-    def put(self, url, data, params = None):
+    def put(self, url, data, params = None, binary = False):
         """Simple PUT (parameters in URL, data in body)"""
-        return self._req("PUT", url, params, data)
+        if binary:
+            h = { 'Content-type': 'application/octet-stream' }
+        else:
+            h = { 'Content-type': 'text/plain; charset=utf-8' }
+        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,
@@ -133,5 +137,14 @@ class HTTPClient(object):
         """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

@@ -31,6 +31,14 @@ 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):
@@ -44,12 +52,7 @@ class NumpyClient(nilmdb.client.client.Client):
         and returned in a flat 2D array.  Otherwise, data is returned
         as a structured dtype in a 1D array.
         """
-        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]
-        dtype = layout_to_dtype(layout)
+        dtype = self._get_dtype(path, layout)
 
         def to_numpy(data):
             a = numpy.fromstring(data, dtype)
@@ -75,3 +78,181 @@ class NumpyClient(nilmdb.client.client.Client):
 
         if total_len:
             yield to_numpy("".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 type(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 len(array) == 0:
+                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/cmdline.py

@@ -10,6 +10,7 @@ import sys
 import os
 import argparse
 from argparse import ArgumentDefaultsHelpFormatter as def_form
+import signal
 
 try: # pragma: no cover
     import argcomplete
@@ -71,15 +72,27 @@ class Complete(object): # pragma: no cover
         path = parsed_args.path
         if not path:
             return []
-        return ( self.escape(k + '=' + v)
-                 for (k,v) in client.stream_get_metadata(path).iteritems()
-                 if k.startswith(prefix) )
-
+        results = []
+        # prefix comes in as UTF-8, but results need to be Unicode,
+        # weird.  Still doesn't work in all cases, but that's bugs in
+        # argcomplete.
+        prefix = nilmdb.utils.unicode.decode(prefix)
+        for (k,v) in client.stream_get_metadata(path).iteritems():
+            kv = self.escape(k + '=' + v)
+            if kv.startswith(prefix):
+                results.append(kv)
+        return results
 
 class Cmdline(object):
 
     def __init__(self, argv = None):
         self.argv = argv or sys.argv[1:]
+        try:
+            # Assume command line arguments are encoded with stdin's encoding,
+            # and reverse it.  Won't be needed in Python 3, but for now..
+            self.argv = [ x.decode(sys.stdin.encoding) for x in self.argv ]
+        except Exception: # pragma: no cover
+            pass
         self.client = None
         self.def_url = os.environ.get("NILMDB_URL", "http://localhost/nilmdb/")
         self.subcmd = {}
@@ -126,6 +139,13 @@ class Cmdline(object):
         sys.exit(-1)
 
     def run(self):
+        # Set SIGPIPE to its default handler -- we don't need Python
+        # to catch it for us.
+        try:
+            signal.signal(signal.SIGPIPE, signal.SIG_DFL)
+        except ValueError: # pragma: no cover
+            pass
+
         # Clear cached timezone, so that we can pick up timezone changes
         # while running this from the test suite.
         datetime_tz._localtz = None

nilmdb/cmdline/destroy.py

@@ -1,5 +1,6 @@
 from nilmdb.utils.printf import *
 import nilmdb.client
+import fnmatch
 
 from argparse import ArgumentDefaultsHelpFormatter as def_form
 
@@ -10,25 +11,39 @@ def setup(self, sub):
                          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",
-                       help="Path of the stream to delete, e.g. /foo/bar",
+    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"""
-    if self.args.remove:
+    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:
-            count = self.client.stream_remove(self.args.path)
+            if self.args.remove:
+                count = self.client.stream_remove(path)
+            self.client.stream_destroy(path)
         except nilmdb.client.ClientError as e:
-            self.die("error removing data: %s", str(e))
-    try:
-        self.client.stream_destroy(self.args.path)
-    except nilmdb.client.ClientError as e:
-        self.die("error destroying stream: %s", str(e))
+            self.die("error destroying stream: %s", str(e))

nilmdb/cmdline/extract.py

@@ -1,6 +1,7 @@
 from __future__ import print_function
 from nilmdb.utils.printf import *
 import nilmdb.client
+import sys
 
 def setup(self, sub):
     cmd = sub.add_parser("extract", help="Extract data",
@@ -24,6 +25,8 @@ def setup(self, sub):
                        ).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",
@@ -42,6 +45,11 @@ 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:
@@ -60,16 +68,23 @@ def cmd_extract(self):
         printf("# end: %s\n", time_string(self.args.end))
 
     printed = False
+    if self.args.binary:
+        printer = sys.stdout.write
+    else:
+        printer = print
+    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):
-        if self.args.bare and not 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 = ' '.join(dataline.split(' ')[1:])
-        print(dataline)
+        printer(dataline)
         printed = True
     if not printed:
         if self.args.annotate:

nilmdb/cmdline/info.py

@@ -21,5 +21,8 @@ def cmd_info(self):
     printf("Server URL: %s\n", self.client.geturl())
     dbinfo = self.client.dbinfo()
     printf("Server database path: %s\n", dbinfo["path"])
-    printf("Server database size: %s\n", human_size(dbinfo["size"]))
-    printf("Server database free space: %s\n", human_size(dbinfo["free"]))
+    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/list.py

@@ -10,22 +10,16 @@ def setup(self, sub):
                          formatter_class = def_form,
                          description="""
                          List streams available in the database,
-                         optionally filtering by layout or path.  Wildcards
-                         are accepted.
+                         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("-p", "--path", metavar="PATH", default="*",
-                       help="Match only this path (-p can be omitted)",
+    group.add_argument("path", metavar="PATH", default=["*"], nargs='*',
                        ).completer = self.complete.path
-    group.add_argument("path_positional", default="*",
-                       nargs="?", help=argparse.SUPPRESS,
-                       ).completer = self.complete.path
-    group.add_argument("-l", "--layout", default="*",
-                       help="Match only this stream layout",
-                       ).completer = self.complete.layout
 
     group = cmd.add_argument_group("Interval info")
     group.add_argument("-E", "--ext", action="store_true",
@@ -49,20 +43,12 @@ def setup(self, sub):
     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")
 
     return cmd
 
 def cmd_list_verify(self):
-    # A hidden "path_positional" argument lets the user leave off the
-    # "-p" when specifying the path.  Handle it here.
-    got_opt = self.args.path != "*"
-    got_pos = self.args.path_positional != "*"
-    if got_pos:
-        if got_opt:
-            self.parser.error("too many paths specified")
-        else:
-            self.args.path = self.args.path_positional
-
     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")
@@ -80,29 +66,33 @@ def cmd_list(self):
     else:
         time_string = nilmdb.utils.time.timestamp_to_human
 
-    for stream in streams:
-        (path, layout, int_min, int_max, rows, time) = stream[:6]
-        if not (fnmatch.fnmatch(path, self.args.path) and
-                fnmatch.fnmatch(layout, self.args.layout)):
-            continue
+    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
 
-        printf("%s %s\n", path, layout)
-
-        if self.args.ext:
-            if int_min is None or int_max is None:
-                printf("  interval extents: (no data)\n")
+            if self.args.layout:
+                printf("%s %s\n", path, layout)
             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))
+                printf("%s\n", path)
 
-        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")
+            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

@@ -41,10 +41,10 @@ def cmd_metadata(self):
     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
+            keyvals = map(nilmdb.utils.unicode.decode, self.args.set)
             handler = self.client.stream_set_metadata
         else:
-            keyvals = self.args.update
+            keyvals = map(nilmdb.utils.unicode.decode, self.args.update)
             handler = self.client.stream_update_metadata
 
         # Extract key=value pairs
@@ -62,7 +62,9 @@ def cmd_metadata(self):
             self.die("error setting/updating metadata: %s", str(e))
     elif self.args.delete is not None:
         # Delete (by setting values to empty strings)
-        keys = self.args.delete or None
+        keys = None
+        if self.args.delete:
+            keys = map(nilmdb.utils.unicode.decode, self.args.delete)
         try:
             data = self.client.stream_get_metadata(self.args.path, keys)
             for key in data:
@@ -72,7 +74,9 @@ def cmd_metadata(self):
             self.die("error deleting metadata: %s", str(e))
     else:
         # Get (or unspecified)
-        keys = self.args.get or None
+        keys = None
+        if self.args.get:
+            keys = map(nilmdb.utils.unicode.decode, self.args.get)
         try:
             data = self.client.stream_get_metadata(self.args.path, keys)
         except nilmdb.client.ClientError as e:
@@ -81,4 +85,6 @@ def cmd_metadata(self):
             # Print nonexistant keys as having empty value
             if value is None:
                 value = ""
-            printf("%s=%s\n", key, value)
+            printf("%s=%s\n",
+                   nilmdb.utils.unicode.encode(key),
+                   nilmdb.utils.unicode.encode(value))

nilmdb/cmdline/remove.py

@@ -1,17 +1,19 @@
 from nilmdb.utils.printf import *
 import nilmdb.client
+import fnmatch
 
 def setup(self, sub):
     cmd = sub.add_parser("remove", help="Remove data",
                          description="""
                          Remove all data from a specified time range within a
-                         stream.
+                         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",
-                       help="Path of stream, e.g. /foo/bar",
+    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,
@@ -23,18 +25,31 @@ def setup(self, sub):
                        ).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:
-        count = self.client.stream_remove(self.args.path,
-                                          self.args.start, self.args.end)
+        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))
 
-    if self.args.count:
-        printf("%d\n", count)
-
     return 0

nilmdb/server/bulkdata.py

@@ -19,8 +19,8 @@ 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 = 16
-fd_cache_size = 16
+table_cache_size = 32
+fd_cache_size = 8
 
 @nilmdb.utils.must_close(wrap_verify = False)
 class BulkData(object):
@@ -413,12 +413,16 @@ class Table(object):
         return rocket.Rocket(self.layout,
                              os.path.join(self.root, subdir, filename))
 
-    def append_string(self, data, start, end):
+    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.
 
+        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."""
@@ -437,17 +441,26 @@ class Table(object):
                 # 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
-                     ) = f.append_string(count, data, data_offset, linenum,
-                                         start, end, last_timestamp)
+                     ) = appender(count, data, data_offset, linenum,
+                                  start, end, last_timestamp)
                 except rocket.ParseError as e:
                     (linenum, colnum, errtype, obj) = e.args
-                    where = "line %d, column %d: " % (linenum, colnum)
+                    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]
-                        badptr = ' ' * (colnum - 1) + '^'
-                    except IndexError: # pragma: no cover
+                        bad += '\n' + ' ' * (colnum - 1) + '^'
+                    except IndexError:
                         bad = ""
                     if errtype == rocket.ERR_NON_MONOTONIC:
                         err = "timestamp is not monotonically increasing"
@@ -463,7 +476,7 @@ class Table(object):
                     else:
                         err = str(obj)
                     raise ValueError("error parsing input data: " +
-                                     where + err + "\n" + bad + "\n" + badptr)
+                                     where + err + "\n" + bad)
                 tot_rows += added_rows
         except Exception:
             # Some failure, so try to roll things back by truncating or

nilmdb/server/nilmdb.py

@@ -176,7 +176,7 @@ class NilmDB(object):
             raise NilmDBError("start must precede end")
         return (start, end)
 
-    @nilmdb.utils.lru_cache(size = 16)
+    @nilmdb.utils.lru_cache(size = 64)
     def _get_intervals(self, stream_id):
         """
         Return a mutable IntervalSet corresponding to the given stream ID.
@@ -475,12 +475,16 @@ class NilmDB(object):
             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):
+    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)
@@ -494,7 +498,7 @@ class NilmDB(object):
         # there are any parse errors.
         table = self.data.getnode(path)
         row_start = table.nrows
-        table.append_string(data, start, end)
+        table.append_data(data, start, end, binary)
         row_end = table.nrows
 
         # Insert the record into the sql database.
@@ -671,6 +675,7 @@ class NilmDB(object):
 
             # Count how many were removed
             removed += row_end - row_start
+            remaining -= row_end - row_start
 
             if restart is not None:
                 break

nilmdb/server/rocket.c

@@ -28,7 +28,7 @@ const static char __long_ok[1 - 2*!(sizeof(int64_t) ==
 
 /* Somewhat arbitrary, just so we can use fixed sizes for strings
    etc. */
-static const int MAX_LAYOUT_COUNT = 128;
+static const int MAX_LAYOUT_COUNT = 1024;
 
 /* Error object and constants */
 static PyObject *ParseError;
@@ -419,6 +419,68 @@ extra_data_on_line:
 			 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;
+	timestamp_t start;
+	timestamp_t end;
+	timestamp_t last_timestamp;
+
+	if (!PyArg_ParseTuple(args, "it#iilll:append_binary",
+                              &count, &data, &data_len, &offset,
+                              &linenum, &start, &end, &last_timestamp))
+		return NULL;
+
+        /* 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) != 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,
+                          last_timestamp, linenum);
+	return o;
+}
+
 /****
  * Extract to string
  */
@@ -484,7 +546,7 @@ static PyObject *Rocket_extract_string(Rocket *self, PyObject *args)
 			/* read and format in a loop */			\
 			for (i = 0; i < self->layout_count; i++) {	\
 				if (fread(&disktype, bytes,		\
-					  1, self->file) < 0)		\
+					  1, self->file) != 1)		\
 					goto err;			\
 				disktype = letoh(disktype);		\
 				ret = sprintf(&str[len], " " fmt,	\
@@ -611,11 +673,13 @@ static PyMemberDef Rocket_members[] = {
 };
 
 static PyMethodDef Rocket_methods[] = {
-	{ "close", (PyCFunction)Rocket_close, METH_NOARGS,
+	{ "close",
+          (PyCFunction)Rocket_close, METH_NOARGS,
 	  "close(self)\n\n"
 	  "Close file handle" },
 
-	{ "append_string", (PyCFunction)Rocket_append_string, METH_VARARGS,
+	{ "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"
@@ -630,12 +694,36 @@ static PyMethodDef Rocket_methods[] = {
 	  "Raises ParseError if timestamps are non-monotonic, outside\n"
 	  "the start/end interval etc.\n"
 	  "\n"
-          "On success, return a tuple with three values:\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" },
+          "  last_timestamp: last timestamp we parsed\n"
+          "  linenum: current line number" },
 
-	{ "extract_string", (PyCFunction)Rocket_extract_string, METH_VARARGS,
+	{ "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" },

nilmdb/server/server.py

@@ -167,9 +167,13 @@ class Root(NilmApp):
         """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": nilmdb.utils.du(path),
-                 "free": psutil.disk_usage(path).free }
+                 "size": dbsize,
+                 "other": usage.used - dbsize,
+                 "reserved": usage.total - usage.used - usage.free,
+                 "free": usage.free }
 
 class Stream(NilmApp):
     """Stream-specific operations"""
@@ -305,10 +309,15 @@ class Stream(NilmApp):
     @cherrypy.tools.json_out()
     @exception_to_httperror(NilmDBError, ValueError)
     @cherrypy.tools.CORS_allow(methods = ["PUT"])
-    def insert(self, path, start, end):
+    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).
         """
         # Important that we always read the input before throwing any
         # errors, to keep lengths happy for persistent connections.
@@ -316,6 +325,14 @@ class Stream(NilmApp):
         # 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)
+
         # Check path and get layout
         if len(self.db.stream_list(path = path)) != 1:
             raise cherrypy.HTTPError("404", "No such stream: " + path)
@@ -325,7 +342,7 @@ class Stream(NilmApp):
 
         # 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)
+        self.db.stream_insert(path, start, end, body, binary)
 
         # Done
         return
@@ -334,24 +351,34 @@ class Stream(NilmApp):
     # /stream/remove?path=/newton/prep&start=1234567890.0&end=1234567899.0
     @cherrypy.expose
     @cherrypy.tools.json_in()
-    @cherrypy.tools.json_out()
-    @exception_to_httperror(NilmDBError)
     @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.
+        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)
-        total_removed = 0
-        while True:
-            (removed, restart) = self.db.stream_remove(path, start, end)
-            total_removed += removed
-            if restart is None:
-                break
-            start = restart
-        return total_removed
+
+        if len(self.db.stream_list(path = path)) != 1:
+            raise cherrypy.HTTPError("404", "No such stream: " + path)
+
+        @workaround_cp_bug_1200
+        def content(start, end):
+            # Note: disable chunked responses to see tracebacks from here.
+            while True:
+                (removed, restart) = self.db.stream_remove(path, start, end)
+                yield json.dumps(removed) + "\r\n"
+                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
@@ -398,7 +425,6 @@ class Stream(NilmApp):
     # /stream/extract?path=/newton/prep&start=1234567890.0&end=1234567899.0
     @cherrypy.expose
     @chunked_response
-    @response_type("text/plain")
     def extract(self, path, start = None, end = None,
                 count = False, markup = False, binary = False):
         """
@@ -414,8 +440,8 @@ class Stream(NilmApp):
 
         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 a
-        uint64 timestamp).
+        little-endian and matches the database types (including an
+        int64 timestamp).
         """
         (start, end) = self._get_times(start, end)
 
@@ -424,11 +450,13 @@ class Stream(NilmApp):
             raise cherrypy.HTTPError("404", "No such stream: " + path)
 
         if binary:
-            cherrypy.response.headers['Content-Type'] = (
-                "application/octet-stream")
+            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
 
         @workaround_cp_bug_1200
         def content(start, end):

nilmdb/utils/__init__.py

@@ -13,3 +13,5 @@ import nilmdb.utils.time
 import nilmdb.utils.iterator
 import nilmdb.utils.interval
 import nilmdb.utils.lock
+import nilmdb.utils.sort
+import nilmdb.utils.unicode

nilmdb/utils/diskusage.py

@@ -1,4 +1,5 @@
 import os
+import errno
 from math import log
 
 def human_size(num):
@@ -16,10 +17,17 @@ def human_size(num):
         return '1 byte'
 
 def du(path):
-    """Like du -sb, returns total size of path in bytes."""
-    size = os.path.getsize(path)
-    if os.path.isdir(path):
-        for thisfile in os.listdir(path):
-            filepath = os.path.join(path, thisfile)
-            size += du(filepath)
-    return size
+    """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:
+        size = os.path.getsize(path)
+        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: # pragma: no cover
+        if e.errno != errno.ENOENT:
+            raise
+        return 0

nilmdb/utils/sort.py

@@ -0,0 +1,18 @@
+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/time.py

@@ -6,7 +6,7 @@ import time
 
 # Range
 min_timestamp = (-2**63)
-max_timestamp = (2**62 - 1)
+max_timestamp = (2**63 - 1)
 
 # Smallest representable step
 epsilon = 1
@@ -32,6 +32,10 @@ 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")
 

nilmdb/utils/unicode.py

@@ -0,0 +1,22 @@
+def encode(u):
+    """Try to encode something from Unicode to a string using the
+    default encoding.  If it fails, try encoding as UTF-8."""
+    if not isinstance(u, unicode):
+        return u
+    try:
+        return u.encode()
+    except UnicodeEncodeError:
+        return u.encode("utf-8")
+
+def decode(s):
+    """Try to decode someting from string to Unicode using the
+    default encoding.  If it fails, try decoding as UTF-8."""
+    if isinstance(s, unicode):
+        return s
+    try:
+        return s.decode()
+    except UnicodeDecodeError:
+        try:
+            return s.decode("utf-8")
+        except UnicodeDecodeError:
+            return s # best we can do

setup.py

@@ -114,7 +114,6 @@ setup(name='nilmdb',
       install_requires = [ 'decorator',
                            'cherrypy >= 3.2',
                            'simplejson',
-                           'pycurl',
                            'python-dateutil',
                            'pytz',
                            'psutil >= 0.3.0',

tests/test_bulkdata.py

@@ -69,9 +69,9 @@ class TestBulkData(object):
         raw = []
         for i in range(1000):
             raw.append("%d 1 2 3 4 5 6 7 8\n" % (10000 + i))
-        node.append_string("".join(raw[0:1]), 0, 50000)
-        node.append_string("".join(raw[1:100]), 0, 50000)
-        node.append_string("".join(raw[100:]), 0, 50000)
+        node.append_data("".join(raw[0:1]), 0, 50000)
+        node.append_data("".join(raw[1:100]), 0, 50000)
+        node.append_data("".join(raw[100:]), 0, 50000)
 
         misc_slices = [ 0, 100, slice(None), slice(0), slice(10),
                         slice(5,10), slice(3,None), slice(3,-3),
@@ -85,7 +85,7 @@ class TestBulkData(object):
         # Extract misc slices while appending, to make sure the
         # data isn't being added in the middle of the file
         for s in [2, slice(1,5), 2, slice(1,5)]:
-            node.append_string("0 0 0 0 0 0 0 0 0\n", 0, 50000)
+            node.append_data("0 0 0 0 0 0 0 0 0\n", 0, 50000)
             raw.append("0 0 0 0 0 0 0 0 0\n")
             eq_(get_node_slice(s), raw[s])
 

tests/test_client.py

@@ -105,16 +105,19 @@ class TestClient(object):
                 client.http.post("/stream/list")
         client = nilmdb.client.Client(url = testurl)
 
-        # Create three streams
+        # Create four streams
         client.stream_create("/newton/prep", "float32_8")
         client.stream_create("/newton/raw", "uint16_6")
-        client.stream_create("/newton/zzz/rawnotch", "uint16_9")
+        client.stream_create("/newton/zzz/rawnotch2", "uint16_9")
+        client.stream_create("/newton/zzz/rawnotch11", "uint16_9")
 
-        # Verify we got 3 streams
+        # Verify we got 4 streams in the right order
         eq_(client.stream_list(), [ ["/newton/prep", "float32_8"],
                                     ["/newton/raw", "uint16_6"],
-                                    ["/newton/zzz/rawnotch", "uint16_9"]
+                                    ["/newton/zzz/rawnotch2", "uint16_9"],
+                                    ["/newton/zzz/rawnotch11", "uint16_9"]
                                     ])
+
         # Match just one type or one path
         eq_(client.stream_list(layout="uint16_6"),
             [ ["/newton/raw", "uint16_6"] ])
@@ -239,6 +242,22 @@ class TestClient(object):
         in_("400 Bad Request", str(e.exception))
         in_("start must precede end", str(e.exception))
 
+        # Good content type
+        with assert_raises(ClientError) as e:
+            client.http.put("stream/insert", "",
+                            { "path": "xxxx", "start": 0, "end": 1,
+                              "binary": 1 },
+                            binary = True)
+        in_("No such stream", str(e.exception))
+
+        # Bad content type
+        with assert_raises(ClientError) as e:
+            client.http.put("stream/insert", "",
+                            { "path": "xxxx", "start": 0, "end": 1,
+                              "binary": 1 },
+                            binary = False)
+        in_("Content type must be application/octet-stream", str(e.exception))
+
         # Specify start/end (starts too late)
         data = timestamper.TimestamperRate(testfile, start, 120)
         with assert_raises(ClientError) as e:
@@ -311,6 +330,10 @@ class TestClient(object):
                      2525.169921875, 8350.83984375, 3724.699951171875,
                      1355.3399658203125, 2039.0))
 
+        # Just get some coverage
+        with assert_raises(ClientError) as e:
+            client.http.post("/stream/remove", { "path": "/none" })
+
         client.close()
 
     def test_client_06_generators(self):
@@ -383,6 +406,17 @@ class TestClient(object):
             raise AssertionError("/stream/extract is not text/plain:\n" +
                                  headers())
 
+        x = http.get("stream/extract",
+                            { "path": "/newton/prep",
+                              "start": "123",
+                              "end": "124",
+                              "binary": "1" })
+        if "transfer-encoding: chunked" not in headers():
+            warnings.warn("Non-chunked HTTP response for /stream/extract")
+        if "content-type: application/octet-stream" not in headers():
+            raise AssertionError("/stream/extract is not binary:\n" +
+                                 headers())
+
         client.close()
 
     def test_client_08_unicode(self):
@@ -459,72 +493,75 @@ class TestClient(object):
             # override _max_data to trigger frequent server updates
             ctx._max_data = 15
 
-            ctx.insert("100 1\n")
+            ctx.insert("1000 1\n")
 
-            ctx.insert("101 ")
-            ctx.insert("1\n102 1")
+            ctx.insert("1010 ")
+            ctx.insert("1\n1020 1")
             ctx.insert("")
-            ctx.insert("\n103 1\n")
+            ctx.insert("\n1030 1\n")
 
-            ctx.insert("104 1\n")
+            ctx.insert("1040 1\n")
             ctx.insert("# hello\n")
             ctx.insert("   # hello\n")
-            ctx.insert("  105 1\n")
+            ctx.insert("  1050 1\n")
             ctx.finalize()
 
-            ctx.insert("107 1\n")
-            ctx.update_end(108)
+            ctx.insert("1070 1\n")
+            ctx.update_end(1080)
             ctx.finalize()
-            ctx.update_start(109)
-            ctx.insert("110 1\n")
-            ctx.insert("111 1\n")
+            ctx.update_start(1090)
+            ctx.insert("1100 1\n")
+            ctx.insert("1110 1\n")
             ctx.send()
-            ctx.insert("112 1\n")
-            ctx.insert("113 1\n")
-            ctx.insert("114 1\n")
-            ctx.update_end(116)
-            ctx.insert("115 1\n")
-            ctx.update_end(117)
-            ctx.insert("116 1\n")
-            ctx.update_end(118)
-            ctx.insert("117 1" +
+            ctx.insert("1120 1\n")
+            ctx.insert("1130 1\n")
+            ctx.insert("1140 1\n")
+            ctx.update_end(1160)
+            ctx.insert("1150 1\n")
+            ctx.update_end(1170)
+            ctx.insert("1160 1\n")
+            ctx.update_end(1180)
+            ctx.insert("1170 1" +
                        " # this is super long" * 100 +
                        "\n")
             ctx.finalize()
             ctx.insert("# this is super long" * 100)
 
         with assert_raises(ClientError):
-            with client.stream_insert_context("/context/test", 100, 200) as ctx:
-                ctx.insert("118 1\n")
+            with client.stream_insert_context("/context/test",
+                                              1000, 2000) as ctx:
+                ctx.insert("1180 1\n")
 
         with assert_raises(ClientError):
-            with client.stream_insert_context("/context/test", 200, 300) as ctx:
-                ctx.insert("118 1\n")
+            with client.stream_insert_context("/context/test",
+                                              2000, 3000) as ctx:
+                ctx.insert("1180 1\n")
 
         with assert_raises(ClientError):
             with client.stream_insert_context("/context/test") as ctx:
                 ctx.insert("bogus data\n")
 
-        with client.stream_insert_context("/context/test", 200, 300) as ctx:
+        with client.stream_insert_context("/context/test", 2000, 3000) as ctx:
             # make sure our override wasn't permanent
             ne_(ctx._max_data, 15)
-            ctx.insert("225 1\n")
+            ctx.insert("2250 1\n")
             ctx.finalize()
 
         with assert_raises(ClientError):
-            with client.stream_insert_context("/context/test", 300, 400) as ctx:
-                ctx.insert("301 1\n")
-                ctx.insert("302 2\n")
-                ctx.insert("303 3\n")
-                ctx.insert("304 4\n")
-                ctx.insert("304 4\n") # non-monotonic after a few lines
+            with client.stream_insert_context("/context/test",
+                                              3000, 4000) as ctx:
+                ctx.insert("3010 1\n")
+                ctx.insert("3020 2\n")
+                ctx.insert("3030 3\n")
+                ctx.insert("3040 4\n")
+                ctx.insert("3040 4\n") # non-monotonic after a few lines
                 ctx.finalize()
 
         eq_(list(client.stream_intervals("/context/test")),
-            [ [ 100, 106 ],
-              [ 107, 108 ],
-              [ 109, 118 ],
-              [ 200, 300 ] ])
+            [ [ 1000, 1051 ],
+              [ 1070, 1080 ],
+              [ 1090, 1180 ],
+              [ 2000, 3000 ] ])
 
         # destroy stream (try without removing data first)
         with assert_raises(ClientError):
@@ -583,8 +620,12 @@ class TestClient(object):
         with client.stream_insert_context("/empty/test", end = 950):
             pass
 
+        # Equal start and end is OK as long as there's no data
+        with client.stream_insert_context("/empty/test", start=9, end=9):
+            pass
+
         # Try various things that might cause problems
-        with client.stream_insert_context("/empty/test", 1000, 1050):
+        with client.stream_insert_context("/empty/test", 1000, 1050) as ctx:
             ctx.finalize() # inserts [1000, 1050]
             ctx.finalize() # nothing
             ctx.finalize() # nothing

tests/test_cmdline.py

@@ -88,7 +88,7 @@ class TestCmdline(object):
                 sys.exit(0)
             except SystemExit as e:
                 exitcode = e.code
-        captured = outfile.getvalue()
+        captured = nilmdb.utils.unicode.decode(outfile.getvalue())
         self.captured = captured
         self.exitcode = exitcode
 
@@ -245,8 +245,10 @@ class TestCmdline(object):
         self.contain("Client version: " + nilmdb.__version__)
         self.contain("Server version: " + test_server.version)
         self.contain("Server database path")
-        self.contain("Server database size")
-        self.contain("Server database free space")
+        self.contain("Server disk space used by NilmDB")
+        self.contain("Server disk space used by other")
+        self.contain("Server disk space reserved")
+        self.contain("Server disk space free")
 
     def test_04_createlist(self):
         # Basic stream tests, like those in test_client.
@@ -300,38 +302,19 @@ class TestCmdline(object):
 
         # Verify we got those 3 streams and they're returned in
         # alphabetical order.
-        self.ok("list")
+        self.ok("list -l")
         self.match("/newton/prep float32_8\n"
                    "/newton/raw uint16_6\n"
                    "/newton/zzz/rawnotch uint16_9\n")
 
         # Match just one type or one path.  Also check
         # that --path is optional
-        self.ok("list --path /newton/raw")
-        self.match("/newton/raw uint16_6\n")
-
-        self.ok("list /newton/raw")
-        self.match("/newton/raw uint16_6\n")
-
-        self.fail("list -p /newton/raw /newton/raw")
-        self.contain("too many paths")
-
-        self.ok("list --layout uint16_6")
+        self.ok("list --layout /newton/raw")
         self.match("/newton/raw uint16_6\n")
 
         # Wildcard matches
-        self.ok("list --layout uint16*")
-        self.match("/newton/raw uint16_6\n"
-                   "/newton/zzz/rawnotch uint16_9\n")
-
-        self.ok("list --path *zzz* --layout uint16*")
-        self.match("/newton/zzz/rawnotch uint16_9\n")
-
-        self.ok("list *zzz* --layout uint16*")
-        self.match("/newton/zzz/rawnotch uint16_9\n")
-
-        self.ok("list --path *zzz* --layout float32*")
-        self.match("")
+        self.ok("list *zzz*")
+        self.match("/newton/zzz/rawnotch\n")
 
         # reversed range
         self.fail("list /newton/prep --start 2020-01-01 --end 2000-01-01")
@@ -497,28 +480,28 @@ class TestCmdline(object):
         self.ok("list --detail")
         lines_(self.captured, 8)
 
-        self.ok("list --detail --path *prep")
+        self.ok("list --detail *prep")
         lines_(self.captured, 4)
 
-        self.ok("list --detail --path *prep --start='23 Mar 2012 10:02'")
+        self.ok("list --detail *prep --start='23 Mar 2012 10:02'")
         lines_(self.captured, 3)
 
-        self.ok("list --detail --path *prep --start='23 Mar 2012 10:05'")
+        self.ok("list --detail *prep --start='23 Mar 2012 10:05'")
         lines_(self.captured, 2)
 
-        self.ok("list --detail --path *prep --start='23 Mar 2012 10:05:15'")
+        self.ok("list --detail *prep --start='23 Mar 2012 10:05:15'")
         lines_(self.captured, 2)
         self.contain("10:05:15.000")
 
-        self.ok("list --detail --path *prep --start='23 Mar 2012 10:05:15.50'")
+        self.ok("list --detail *prep --start='23 Mar 2012 10:05:15.50'")
         lines_(self.captured, 2)
         self.contain("10:05:15.500")
 
-        self.ok("list --detail --path *prep --start='23 Mar 2012 19:05:15.50'")
+        self.ok("list --detail *prep --start='23 Mar 2012 19:05:15.50'")
         lines_(self.captured, 2)
         self.contain("no intervals")
 
-        self.ok("list --detail --path *prep --start='23 Mar 2012 10:05:15.50'"
+        self.ok("list --detail *prep --start='23 Mar 2012 10:05:15.50'"
                 + " --end='23 Mar 2012 10:05:15.51'")
         lines_(self.captured, 2)
         self.contain("10:05:15.500")
@@ -527,15 +510,15 @@ class TestCmdline(object):
         lines_(self.captured, 8)
 
         # Verify the "raw timestamp" output
-        self.ok("list --detail --path *prep --timestamp-raw "
+        self.ok("list --detail *prep --timestamp-raw "
                 "--start='23 Mar 2012 10:05:15.50'")
         lines_(self.captured, 2)
         self.contain("[ 1332497115500000 -> 1332497160000000 ]")
 
         # bad time
-        self.fail("list --detail --path *prep -T --start='9332497115.612'")
+        self.fail("list --detail *prep -T --start='9332497115.612'")
         # good time
-        self.ok("list --detail --path *prep -T --start='1332497115.612'")
+        self.ok("list --detail *prep -T --start='1332497115.612'")
         lines_(self.captured, 2)
         self.contain("[ 1332497115612000 -> 1332497160000000 ]")
 
@@ -615,11 +598,19 @@ class TestCmdline(object):
         test(8, "10:01:59.9", "10:02:00.1", extra="-m")
 
         # all data put in by tests
-        self.ok("extract -a /newton/prep --start 2000-01-01 --end 2020-01-01")
+        self.ok("extract -a /newton/prep --start min --end max")
         lines_(self.captured, 43204)
         self.ok("extract -c /newton/prep --start 2000-01-01 --end 2020-01-01")
         self.match("43200\n")
 
+        # test binary mode
+        self.fail("extract -c -B /newton/prep -s min -e max")
+        self.contain("binary cannot be combined")
+        self.fail("extract -m -B /newton/prep -s min -e max")
+        self.contain("binary cannot be combined")
+        self.ok("extract -B /newton/prep -s min -e max")
+        eq_(len(self.captured), 43200 * (8 + 8*4))
+
         # markup for 3 intervals, plus extra markup lines whenever we had
         # a "restart" from the nilmdb.stream_extract function
         self.ok("extract -m /newton/prep --start 2000-01-01 --end 2020-01-01")
@@ -639,7 +630,7 @@ class TestCmdline(object):
 
         # Try nonexistent stream
         self.fail("remove /no/such/foo --start 2000-01-01 --end 2020-01-01")
-        self.contain("No stream at path")
+        self.contain("no stream matched path")
 
         # empty or backward ranges return errors
         self.fail("remove /newton/prep --start 2020-01-01 --end 2000-01-01")
@@ -667,9 +658,14 @@ class TestCmdline(object):
                 "--start '23 Mar 2022 20:00:30' " +
                 "--end '23 Mar 2022 20:00:31'")
         self.match("0\n")
+        self.ok("remove -c /newton/prep /newton/pre* " +
+                "--start '23 Mar 2022 20:00:30' " +
+                "--end '23 Mar 2022 20:00:31'")
+        self.match("Removing from /newton/prep\n0\n" +
+                   "Removing from /newton/prep\n0\n")
 
         # Make sure we have the data we expect
-        self.ok("list --detail /newton/prep")
+        self.ok("list -l --detail /newton/prep")
         self.match("/newton/prep float32_8\n" +
                    "  [ Fri, 23 Mar 2012 10:00:00.000000 +0000"
                    " -> Fri, 23 Mar 2012 10:01:59.991668 +0000 ]\n"
@@ -704,7 +700,7 @@ class TestCmdline(object):
         self.match("24000\n")
 
         # See the missing chunks in list output
-        self.ok("list --detail /newton/prep")
+        self.ok("list --layout --detail /newton/prep")
         self.match("/newton/prep float32_8\n" +
                    "  [ Fri, 23 Mar 2012 10:00:00.000000 +0000"
                    " -> Fri, 23 Mar 2012 10:00:05.000000 +0000 ]\n"
@@ -718,7 +714,7 @@ class TestCmdline(object):
         # Remove all data, verify it's missing
         self.ok("remove /newton/prep --start 2000-01-01 --end 2020-01-01")
         self.match("")  # no count requested this time
-        self.ok("list --detail /newton/prep")
+        self.ok("list -l --detail /newton/prep")
         self.match("/newton/prep float32_8\n" +
                    "  (no intervals)\n")
 
@@ -736,16 +732,16 @@ class TestCmdline(object):
         self.contain("too few arguments")
 
         self.fail("destroy /no/such/stream")
-        self.contain("No stream at path")
+        self.contain("no stream matched path")
 
         self.fail("destroy -R /no/such/stream")
-        self.contain("No stream at path")
+        self.contain("no stream matched path")
 
         self.fail("destroy asdfasdf")
-        self.contain("No stream at path")
+        self.contain("no stream matched path")
 
         # From previous tests, we have:
-        self.ok("list")
+        self.ok("list -l")
         self.match("/newton/prep float32_8\n"
                    "/newton/raw uint16_6\n"
                    "/newton/zzz/rawnotch uint16_9\n")
@@ -761,13 +757,13 @@ class TestCmdline(object):
         lines_(self.captured, 7)
 
         # Destroy for real
-        self.ok("destroy -R /newton/prep")
-        self.ok("list")
+        self.ok("destroy -R /n*/prep")
+        self.ok("list -l")
         self.match("/newton/raw uint16_6\n"
                    "/newton/zzz/rawnotch uint16_9\n")
 
         self.ok("destroy /newton/zzz/rawnotch")
-        self.ok("list")
+        self.ok("list -l")
         self.match("/newton/raw uint16_6\n")
 
         self.ok("destroy /newton/raw")
@@ -786,18 +782,17 @@ class TestCmdline(object):
             self.ok("list")
             self.contain(path)
             # Make sure it was created empty
-            self.ok("list --detail --path " + path)
+            self.ok("list --detail " + path)
             self.contain("(no intervals)")
 
     def test_12_unicode(self):
         # Unicode paths.
         self.ok("destroy /newton/asdf/qwer")
-        self.ok("destroy /newton/prep")
-        self.ok("destroy /newton/raw")
+        self.ok("destroy /newton/prep /newton/raw")
         self.ok("destroy /newton/zzz")
 
         self.ok(u"create /düsseldorf/raw uint16_6")
-        self.ok("list --detail")
+        self.ok("list -l --detail")
         self.contain(u"/düsseldorf/raw uint16_6")
         self.contain("(no intervals)")
 
@@ -883,7 +878,7 @@ class TestCmdline(object):
         du_before = nilmdb.utils.diskusage.du(testdb)
 
         # Make sure we have the data we expect
-        self.ok("list --detail")
+        self.ok("list -l --detail")
         self.match("/newton/prep float32_8\n" +
                    "  [ Fri, 23 Mar 2012 10:00:00.000000 +0000"
                    " -> Fri, 23 Mar 2012 10:01:59.991668 +0000 ]\n"
@@ -919,7 +914,7 @@ class TestCmdline(object):
         self.match("3600\n")
 
         # See the missing chunks in list output
-        self.ok("list --detail")
+        self.ok("list -l --detail")
         self.match("/newton/prep float32_8\n" +
                    "  [ Fri, 23 Mar 2012 10:00:00.000000 +0000"
                    " -> Fri, 23 Mar 2012 10:00:05.000000 +0000 ]\n"
@@ -1043,7 +1038,7 @@ class TestCmdline(object):
             else:
                 raise AssertionError("data not found at " + seek)
             # Verify "list" output
-            self.ok("list")
+            self.ok("list -l")
             self.match("/" + "/".join(components) + " float32_8\n")
 
         # Lots of renames

tests/test_numpyclient.py

@@ -28,7 +28,10 @@ def setup_module():
     recursive_unlink(testdb)
 
     # Start web app on a custom port
-    test_db = nilmdb.utils.serializer_proxy(nilmdb.server.NilmDB)(testdb)
+    test_db = nilmdb.utils.serializer_proxy(nilmdb.server.NilmDB)(
+        testdb, bulkdata_args = { "file_size" : 16384,
+                                  "files_per_dir" : 3 } )
+
     test_server = nilmdb.server.Server(test_db, host = "127.0.0.1",
                                        port = 32180, stoppable = False,
                                        fast_shutdown = True,
@@ -106,3 +109,265 @@ class TestNumpyClient(object):
         assert(np.allclose(array, actual))
 
         client.close()
+
+    def test_numpyclient_03_insert(self):
+        client = nilmdb.client.numpyclient.NumpyClient(url = testurl)
+
+        # Limit _max_data just to get better coverage
+        old_max_data = nilmdb.client.numpyclient.StreamInserterNumpy._max_data
+        nilmdb.client.numpyclient.StreamInserterNumpy._max_data = 100000
+
+        client.stream_create("/test/1", "uint16_1")
+        client.stream_insert_numpy("/test/1",
+                                   np.array([[0, 1],
+                                             [1, 2],
+                                             [2, 3],
+                                             [3, 4]]))
+
+        # Wrong number of dimensions
+        with assert_raises(ValueError) as e:
+            client.stream_insert_numpy("/test/1",
+                                       np.array([[[0, 1],
+                                                  [1, 2]],
+                                                 [[3, 4],
+                                                  [4, 5]]]))
+        in_("wrong number of dimensions", str(e.exception))
+
+        # Wrong number of fields
+        with assert_raises(ValueError) as e:
+            client.stream_insert_numpy("/test/1",
+                                       np.array([[0, 1, 2],
+                                                 [1, 2, 3],
+                                                 [3, 4, 5],
+                                                 [4, 5, 6]]))
+        in_("wrong number of fields", str(e.exception))
+
+        # Unstructured
+        client.stream_create("/test/2", "float32_8")
+        client.stream_insert_numpy(
+            "/test/2",
+            client.stream_extract_numpy(
+                "/newton/prep", structured = False, maxrows = 1000))
+
+        # Structured, and specifying layout
+        client.stream_create("/test/3", "float32_8")
+        client.stream_insert_numpy(
+            path = "/test/3", layout = "float32_8",
+            data = client.stream_extract_numpy(
+                "/newton/prep", structured = True, maxrows = 1000))
+
+        # Structured, specifying wrong layout
+        client.stream_create("/test/4", "float32_8")
+        with assert_raises(ValueError) as e:
+            client.stream_insert_numpy(
+                "/test/4", layout = "uint16_1",
+                data = client.stream_extract_numpy(
+                    "/newton/prep", structured = True, maxrows = 1000))
+        in_("wrong dtype", str(e.exception))
+
+        # Unstructured, and specifying wrong layout
+        client.stream_create("/test/5", "float32_8")
+        with assert_raises(ClientError) as e:
+            client.stream_insert_numpy(
+                "/test/5", layout = "uint16_8",
+                data = client.stream_extract_numpy(
+                    "/newton/prep", structured = False, maxrows = 1000))
+        # timestamps will be screwy here, because data will be parsed wrong
+        in_("error parsing input data", str(e.exception))
+
+        # Make sure the /newton/prep copies are identical
+        a = np.vstack(client.stream_extract_numpy("/newton/prep"))
+        b = np.vstack(client.stream_extract_numpy("/test/2"))
+        c = np.vstack(client.stream_extract_numpy("/test/3"))
+        assert(np.array_equal(a,b))
+        assert(np.array_equal(a,c))
+
+        # Make sure none of the files are greater than 16384 bytes as
+        # we configured with the bulkdata_args above.
+        datapath = os.path.join(testdb, "data")
+        for (dirpath, dirnames, filenames) in os.walk(datapath):
+            for f in filenames:
+                fn = os.path.join(dirpath, f)
+                size = os.path.getsize(fn)
+                if size > 16384:
+                    raise AssertionError(sprintf("%s is too big: %d > %d\n",
+                                                 fn, size, 16384))
+
+        nilmdb.client.numpyclient.StreamInserterNumpy._max_data = old_max_data
+        client.close()
+
+    def test_numpyclient_04_context(self):
+        # Like test_client_context, but with Numpy data
+        client = nilmdb.client.numpyclient.NumpyClient(testurl)
+
+        client.stream_create("/context/test", "uint16_1")
+        with client.stream_insert_numpy_context("/context/test") as ctx:
+            # override _max_rows to trigger frequent server updates
+            ctx._max_rows = 2
+            ctx.insert([[1000, 1]])
+            ctx.insert([[1010, 1], [1020, 1], [1030, 1]])
+            ctx.insert([[1040, 1], [1050, 1]])
+            ctx.finalize()
+            ctx.insert([[1070, 1]])
+            ctx.update_end(1080)
+            ctx.finalize()
+            ctx.update_start(1090)
+            ctx.insert([[1100, 1]])
+            ctx.insert([[1110, 1]])
+            ctx.send()
+            ctx.insert([[1120, 1], [1130, 1], [1140, 1]])
+            ctx.update_end(1160)
+            ctx.insert([[1150, 1]])
+            ctx.update_end(1170)
+            ctx.insert([[1160, 1]])
+            ctx.update_end(1180)
+            ctx.insert([[1170, 123456789.0]])
+            ctx.finalize()
+            ctx.insert(np.zeros((0,2)))
+
+        with assert_raises(ClientError):
+            with client.stream_insert_numpy_context("/context/test",
+                                                    1000, 2000) as ctx:
+                ctx.insert([[1180, 1]])
+
+        with assert_raises(ClientError):
+            with client.stream_insert_numpy_context("/context/test",
+                                                    2000, 3000) as ctx:
+                ctx._max_rows = 2
+                ctx.insert([[3180, 1]])
+                ctx.insert([[3181, 1]])
+
+        with client.stream_insert_numpy_context("/context/test",
+                                                2000, 3000) as ctx:
+            # make sure our override wasn't permanent
+            ne_(ctx._max_rows, 2)
+            ctx.insert([[2250, 1]])
+            ctx.finalize()
+
+        with assert_raises(ClientError):
+            with client.stream_insert_numpy_context("/context/test",
+                                                    3000, 4000) as ctx:
+                ctx.insert([[3010, 1]])
+                ctx.insert([[3020, 2]])
+                ctx.insert([[3030, 3]])
+                ctx.insert([[3040, 4]])
+                ctx.insert([[3040, 4]]) # non-monotonic after a few lines
+                ctx.finalize()
+
+        eq_(list(client.stream_intervals("/context/test")),
+            [ [ 1000, 1051 ],
+              [ 1070, 1080 ],
+              [ 1090, 1180 ],
+              [ 2000, 3000 ] ])
+
+        client.stream_remove("/context/test")
+        client.stream_destroy("/context/test")
+        client.close()
+
+    def test_numpyclient_05_emptyintervals(self):
+        # Like test_client_emptyintervals, with insert_numpy_context
+        client = nilmdb.client.numpyclient.NumpyClient(testurl)
+        client.stream_create("/empty/test", "uint16_1")
+        def info():
+            result = []
+            for interval in list(client.stream_intervals("/empty/test")):
+                result.append((client.stream_count("/empty/test", *interval),
+                               interval))
+            return result
+        eq_(info(), [])
+
+        # Insert a region with just a few points
+        with client.stream_insert_numpy_context("/empty/test") as ctx:
+            ctx.update_start(100)
+            ctx.insert([[140, 1]])
+            ctx.insert([[150, 1]])
+            ctx.insert([[160, 1]])
+            ctx.update_end(200)
+            ctx.finalize()
+        eq_(info(), [(3, [100, 200])])
+
+        # Delete chunk, which will leave one data point and two intervals
+        client.stream_remove("/empty/test", 145, 175)
+        eq_(info(), [(1, [100, 145]),
+                     (0, [175, 200])])
+
+        # Try also creating a completely empty interval from scratch,
+        # in a few different ways.
+        client.stream_insert("/empty/test", "", 300, 350)
+        client.stream_insert("/empty/test", [], 400, 450)
+        with client.stream_insert_numpy_context("/empty/test", 500, 550):
+            pass
+
+        # If enough timestamps aren't provided, empty streams won't be created.
+        client.stream_insert("/empty/test", [])
+        with client.stream_insert_numpy_context("/empty/test"):
+            pass
+        client.stream_insert("/empty/test", [], start = 600)
+        with client.stream_insert_numpy_context("/empty/test", start = 700):
+            pass
+        client.stream_insert("/empty/test", [], end = 850)
+        with client.stream_insert_numpy_context("/empty/test", end = 950):
+            pass
+
+        # Equal start and end is OK as long as there's no data
+        with assert_raises(ClientError) as e:
+            with client.stream_insert_numpy_context("/empty/test",
+                                                    start=9, end=9) as ctx:
+                ctx.insert([[9, 9]])
+                ctx.finalize()
+        in_("have data to send, but invalid start/end times", str(e.exception))
+
+        with client.stream_insert_numpy_context("/empty/test",
+                                                start=9, end=9) as ctx:
+            pass
+
+        # reusing a context object is bad
+        with assert_raises(Exception) as e:
+            ctx.insert([[9, 9]])
+
+        # Try various things that might cause problems
+        with client.stream_insert_numpy_context("/empty/test",
+                                                1000, 1050) as ctx:
+            ctx.finalize() # inserts [1000, 1050]
+            ctx.finalize() # nothing
+            ctx.finalize() # nothing
+            ctx.insert([[1100, 1]])
+            ctx.finalize() # inserts [1100, 1101]
+            ctx.update_start(1199)
+            ctx.insert([[1200, 1]])
+            ctx.update_end(1250)
+            ctx.finalize() # inserts [1199, 1250]
+            ctx.update_start(1299)
+            ctx.finalize() # nothing
+            ctx.update_end(1350)
+            ctx.finalize() # nothing
+            ctx.update_start(1400)
+            ctx.insert(np.zeros((0,2)))
+            ctx.update_end(1450)
+            ctx.finalize()
+            ctx.update_start(1500)
+            ctx.insert(np.zeros((0,2)))
+            ctx.update_end(1550)
+            ctx.finalize()
+            ctx.insert(np.zeros((0,2)))
+            ctx.insert(np.zeros((0,2)))
+            ctx.insert(np.zeros((0,2)))
+            ctx.finalize()
+
+        # Check everything
+        eq_(info(), [(1, [100, 145]),
+                     (0, [175, 200]),
+                     (0, [300, 350]),
+                     (0, [400, 450]),
+                     (0, [500, 550]),
+                     (0, [1000, 1050]),
+                     (1, [1100, 1101]),
+                     (1, [1199, 1250]),
+                     (0, [1400, 1450]),
+                     (0, [1500, 1550]),
+                     ])
+
+        # Clean up
+        client.stream_remove("/empty/test")
+        client.stream_destroy("/empty/test")
+        client.close()