Browse Source

fsck: add fix for file position (row) overlap in database

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
Jim Paris 5 months ago
1 changed files with 46 additions and 6 deletions
  1. +46

+ 46
- 6
nilmdb/fsck/ View File

@@ -238,9 +238,7 @@ class Fsck(object):
posiset += new
except IntervalError:
raise FsckError("%s: overlap in file offsets:\n"
"set: %s\nnew: %s",
path, str(posiset), str(new))
self.fix_row_overlap(sid, path, posiset, new)

# Check bulkdata
@@ -268,6 +266,48 @@ class Fsck(object):
path, str(e))

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 offests:\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"
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

@@ -438,8 +478,8 @@ class Fsck(object):

# 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\n")
err("data from this stream to a new stream using nilm-copy, then\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()
@@ -564,7 +604,7 @@ class Fsck(object):
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")
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