view tests/test-dirstate-race.t @ 40326:fed697fa1734

sqlitestore: file storage backend using SQLite

This commit provides an extension which uses SQLite to store file
data (as opposed to revlogs).

As the inline documentation describes, there are still several
aspects to the extension that are incomplete. But it's a start.
The extension does support basic clone, checkout, and commit
workflows, which makes it suitable for simple use cases.

One notable missing feature is support for "bundlerepos." This is
probably responsible for the most test failures when the extension
is activated as part of the test suite.

All revision data is stored in SQLite. Data is stored as zstd
compressed chunks (default if zstd is available), zlib compressed
chunks (default if zstd is not available), or raw chunks (if
configured or if a compressed delta is not smaller than the raw
delta). This makes things very similar to revlogs.

Unlike revlogs, the extension doesn't yet enforce a limit on delta
chain length. This is an obvious limitation and should be addressed.
This is somewhat mitigated by the use of zstd, which is much faster
than zlib to decompress.

There is a dedicated table for storing deltas. Deltas are stored
by the SHA-1 hash of their uncompressed content. The "fileindex" table
has columns that reference the delta for each revision and the base
delta that delta should be applied against. A recursive SQL query
is used to resolve the delta chain along with the delta data.

By storing deltas by hash, we are able to de-duplicate delta storage!
With revlogs, the same deltas in different revlogs would result in
duplicate storage of that delta. In this scheme, inserting the
duplicate delta is a no-op and delta chains simply reference the
existing delta.

When initially implementing this extension, I did not have
content-indexed deltas and deltas could be duplicated across files
(just like revlogs). When I implemented content-indexed deltas, the
size of the SQLite database for a full clone of mozilla-unified
dropped:

before: 2,554,261,504 bytes
after:  2,488,754,176 bytes

Surprisingly, this is still larger than the bytes size of revlog
files:

revlog files: 2,104,861,230 bytes
du -b:        2,254,381,614

I would have expected storage to be smaller since we're not limiting
delta chain length and since we're using zstd instead of zlib. I
suspect the SQLite indexes and per-column overhead account for the
bulk of the differences. (Keep in mind that revlog uses a 64-byte
packed struct for revision index data and deltas are stored without
padding. Aside from the 12 unused bytes in the 32 byte node field,
revlogs are pretty efficient.) Another source of overhead is file
name storage. With revlogs, file names are stored in the filesystem.
But with SQLite, we need to store file names in the database. This is
roughly equivalent to the size of the fncache file, which for the
mozilla-unified repository is ~34MB.

Since the SQLite database isn't append-only and since delta chains
can reference any delta, this opens some interesting possibilities.
For example, we could store deltas in reverse, such that fulltexts
are stored for newer revisions and deltas are applied to reconstruct
older revisions. This is likely a more optimal storage strategy for
version control, as new data tends to be more frequently accessed
than old data. We would obviously need wire protocol support for
transferring revision data from newest to oldest. And we would
probably need some kind of mechanism for "re-encoding" stores. But
it should be doable.

This extension is very much experimental quality. There are a handful
of features that don't work. It probably isn't suitable for day-to-day
use. But it could be used in limited cases (e.g. read-only checkouts
like in CI). And it is also a good proving ground for alternate
storage backends. As we continue to define interfaces for all things
storage, it will be useful to have a viable alternate storage backend
to see how things shake out in practice.

test-storage.py passes on Python 2 and introduces no new test failures on
Python 3. Having the storage-level unit tests has proved to be insanely
useful when developing this extension. Those tests caught numerous bugs
during development and I'm convinced this style of testing is the way
forward for ensuring alternate storage backends work as intended. Of
course, test coverage isn't close to what it needs to be. But it is
a start. And what coverage we have gives me confidence that basic store
functionality is implemented properly.

Differential Revision: https://phab.mercurial-scm.org/D4928

  $ hg init repo
  $ cd repo
  $ echo a > a
  $ hg add a
  $ hg commit -m test

Do we ever miss a sub-second change?:

  $ for i in 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20; do
  >     hg co -qC 0
  >     echo b > a
  >     hg st
  > done
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a
  M a

  $ echo test > b
  $ mkdir dir1
  $ echo test > dir1/c
  $ echo test > d

  $ echo test > e
#if execbit
A directory will typically have the execute bit -- make sure it doesn't get
confused with a file with the exec bit set
  $ chmod +x e
#endif

  $ hg add b dir1 d e
  adding dir1/c
  $ hg commit -m test2

  $ cat >> $TESTTMP/dirstaterace.py << EOF
  > from mercurial import (
  >     context,
  >     extensions,
  > )
  > def extsetup():
  >     extensions.wrapfunction(context.workingctx, '_checklookup', overridechecklookup)
  > def overridechecklookup(orig, self, files):
  >     # make an update that changes the dirstate from underneath
  >     self._repo.ui.system(br"sh '$TESTTMP/dirstaterace.sh'",
  >                          cwd=self._repo.root)
  >     return orig(self, files)
  > EOF

  $ hg debugrebuilddirstate
  $ hg debugdirstate
  n   0         -1 unset               a
  n   0         -1 unset               b
  n   0         -1 unset               d
  n   0         -1 unset               dir1/c
  n   0         -1 unset               e

XXX Note that this returns M for files that got replaced by directories. This is
definitely a bug, but the fix for that is hard and the next status run is fine
anyway.

  $ cat > $TESTTMP/dirstaterace.sh <<EOF
  > rm b && rm -r dir1 && rm d && mkdir d && rm e && mkdir e
  > EOF

  $ hg status --config extensions.dirstaterace=$TESTTMP/dirstaterace.py
  M d
  M e
  ! b
  ! dir1/c
  $ hg debugdirstate
  n 644          2 * a (glob)
  n   0         -1 unset               b
  n   0         -1 unset               d
  n   0         -1 unset               dir1/c
  n   0         -1 unset               e

  $ hg status
  ! b
  ! d
  ! dir1/c
  ! e

  $ rmdir d e
  $ hg update -C -q .

Test that dirstate changes aren't written out at the end of "hg
status", if .hg/dirstate is already changed simultaneously before
acquisition of wlock in workingctx._poststatusfixup().

This avoidance is important to keep consistency of dirstate in race
condition (see issue5584 for detail).

  $ hg parents -q
  1:* (glob)

  $ hg debugrebuilddirstate
  $ hg debugdirstate
  n   0         -1 unset               a
  n   0         -1 unset               b
  n   0         -1 unset               d
  n   0         -1 unset               dir1/c
  n   0         -1 unset               e

  $ cat > $TESTTMP/dirstaterace.sh <<EOF
  > # This script assumes timetable of typical issue5584 case below:
  > #
  > # 1. "hg status" loads .hg/dirstate
  > # 2. "hg status" confirms clean-ness of FILE
  > # 3. "hg update -C 0" updates the working directory simultaneously
  > #    (FILE is removed, and FILE is dropped from .hg/dirstate)
  > # 4. "hg status" acquires wlock
  > #    (.hg/dirstate is re-loaded = no FILE entry in dirstate)
  > # 5. "hg status" marks FILE in dirstate as clean
  > #    (FILE entry is added to in-memory dirstate)
  > # 6. "hg status" writes dirstate changes into .hg/dirstate
  > #    (FILE entry is written into .hg/dirstate)
  > #
  > # To reproduce similar situation easily and certainly, #2 and #3
  > # are swapped.  "hg cat" below ensures #2 on "hg status" side.
  > 
  > hg update -q -C 0
  > hg cat -r 1 b > b
  > EOF

"hg status" below should excludes "e", of which exec flag is set, for
portability of test scenario, because unsure but missing "e" is
treated differently in _checklookup() according to runtime platform.

- "missing(!)" on POSIX, "pctx[f].cmp(self[f])" raises ENOENT
- "modified(M)" on Windows, "self.flags(f) != pctx.flags(f)" is True

  $ hg status --config extensions.dirstaterace=$TESTTMP/dirstaterace.py --debug -X path:e
  skip updating dirstate: identity mismatch
  M a
  ! d
  ! dir1/c

  $ hg parents -q
  0:* (glob)
  $ hg files
  a
  $ hg debugdirstate
  n * * * a (glob)

  $ rm b

#if fsmonitor

Create fsmonitor state.

  $ hg status
  $ f --type .hg/fsmonitor.state
  .hg/fsmonitor.state: file

Test that invalidating fsmonitor state in the middle (which doesn't require the
wlock) causes the fsmonitor update to be skipped.
hg debugrebuilddirstate ensures that the dirstaterace hook will be called, but
it also invalidates the fsmonitor state. So back it up and restore it.

  $ mv .hg/fsmonitor.state .hg/fsmonitor.state.tmp
  $ hg debugrebuilddirstate
  $ mv .hg/fsmonitor.state.tmp .hg/fsmonitor.state

  $ cat > $TESTTMP/dirstaterace.sh <<EOF
  > rm .hg/fsmonitor.state
  > EOF

  $ hg status --config extensions.dirstaterace=$TESTTMP/dirstaterace.py --debug
  skip updating fsmonitor.state: identity mismatch
  $ f .hg/fsmonitor.state
  .hg/fsmonitor.state: file not found

#endif

Set up a rebase situation for issue5581.

  $ echo c2 > a
  $ echo c2 > b
  $ hg add b
  $ hg commit -m c2
  created new head
  $ echo c3 >> a
  $ hg commit -m c3
  $ hg update 2
  1 files updated, 0 files merged, 0 files removed, 0 files unresolved
  $ echo c4 >> a
  $ echo c4 >> b
  $ hg commit -m c4
  created new head

Configure a merge tool that runs status in the middle of the rebase. The goal of
the status call is to trigger a potential bug if fsmonitor's state is written
even though the wlock is held by another process. The output of 'hg status' in
the merge tool goes to /dev/null because we're more interested in the results of
'hg status' run after the rebase.

  $ cat >> $TESTTMP/mergetool-race.sh << EOF
  > echo "custom merge tool"
  > printf "c2\nc3\nc4\n" > \$1
  > hg --cwd "$TESTTMP/repo" status > /dev/null
  > echo "custom merge tool end"
  > EOF
  $ cat >> $HGRCPATH << EOF
  > [extensions]
  > rebase =
  > [merge-tools]
  > test.executable=sh
  > test.args=$TESTTMP/mergetool-race.sh \$output
  > EOF

  $ hg rebase -s . -d 3 --tool test
  rebasing 4:b08445fd6b2a "c4" (tip)
  merging a
  custom merge tool
  custom merge tool end
  saved backup bundle to $TESTTMP/repo/.hg/strip-backup/* (glob)

This hg status should be empty, whether or not fsmonitor is enabled (issue5581).

  $ hg status