view tests/test-dirstate-race.t @ 47010:76ae43d5b1db stable

repoview: fix memory leak of filtered repo classes

The leak occurs in long-running server processes with
extensions, and is measured at 110kB per request.

Before this change, the contents of the `_filteredrepotypes`
cache are not properly garbage collected, despite it begin
a `WeakKeyDictionary`.

Extensions have a tendency to generate a new repository class
for each `localrepo` instantiation. Server processes based
on `hgwebdir_mod` will instantiate a new `localrepo` for each
HTTP request that involves a repository.

As a result, with a testing process that repeatedly opens a
repository with several extensions activated
(`topic` notably among them), we see a steady increase in
resident memory of 110kB per repository instantiation before this
change. This is also true, if we call `gc.collect()` at each
instantiation, like `hgwebdir_mod` does, or not.

The cause of the leak is that the *values* aren't weak references.
This change uses `weakref.ref` for the values, and this makes
in our measurements the resident size increase drop to 5kB per
repository instantiation, with no explicit call of `gc.collect()`
at all.
There is currently no reason to believe that this remaining leak
of 5kB is related to or even due to Mercurial core.

We've also seen evidence that `ui.ui` instances weren't properly
garbage collected before the change (with the change, they are).
This could explain why the figures are relatively high.

In theory, the collection of weak references could lead to
much more misses in the cache, so we measured the impact on
the original case that was motivation for introducing that cache
in 7e89bd0cfb86 (see also issue5043): `hg convert` of the
mozilla-central repository. The bad news here is that there is a
major memory leak there, both with and without the present changeset.
There were no more cache misses, and we could see no
more memory leak with this change: the resident size after importing
roughly 100000 changesets was at 12.4GB before, and 12.5GB after.
The small increase is mentioned for completeness only, and we
believe that it should be ignored, at least as long as the main
leak isn't fixed. At less than 1% of the main leak, even finding out
whether it is merely noise would be wasteful.

Original context where this was spotted and first mitigated:
  https://foss.heptapod.net/heptapod/heptapod/-/issues/466

The leak reduction was also obtained in Heptapod inner HTTP server,
which amounts to the same as `hgwebdir_mod` for these questions.

The measurements done with Python 3.9, similar figures seen with 3.8.
More work on our side would be needed to give measurements with 2.7,
because of testing server process does not support it.

  $ 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 >> $TESTTMP/dirstaterace.py << EOF
  > from mercurial import (
  >     context,
  >     extensions,
  > )
  > def extsetup(ui):
  >     extensions.wrapfunction(context.workingctx, '_checklookup', overridechecklookup)
  > def overridechecklookup(orig, self, files):
  >     # make an update that changes the dirstate from underneath
  >     self._repo.ui.system(br"sh '$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 tip "c4"
  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