util: lower water mark when removing nodes after cost limit reached
See the inline comment for the reasoning here. This is a pretty
common strategy for garbage collectors, other cache-like primtives.
The performance impact is substantial:
$ hg perflrucachedict --size 4 --gets 1000000 --sets 1000000 --mixed 1000000 --costlimit 100
! inserts w/ cost limit
! wall 1.659181 comb 1.650000 user 1.650000 sys 0.000000 (best of 7)
! wall 1.722122 comb 1.720000 user 1.720000 sys 0.000000 (best of 6)
! mixed w/ cost limit
! wall 1.139955 comb 1.140000 user 1.140000 sys 0.000000 (best of 9)
! wall 1.182513 comb 1.180000 user 1.180000 sys 0.000000 (best of 9)
$ hg perflrucachedict --size 1000 --gets 1000000 --sets 1000000 --mixed 1000000 --costlimit 10000
! inserts
! wall 0.679546 comb 0.680000 user 0.680000 sys 0.000000 (best of 15)
! sets
! wall 0.825147 comb 0.830000 user 0.830000 sys 0.000000 (best of 13)
! inserts w/ cost limit
! wall 25.105273 comb 25.080000 user 25.080000 sys 0.000000 (best of 3)
! wall 1.724397 comb 1.720000 user 1.720000 sys 0.000000 (best of 6)
! mixed
! wall 0.807096 comb 0.810000 user 0.810000 sys 0.000000 (best of 13)
! mixed w/ cost limit
! wall 12.104470 comb 12.070000 user 12.070000 sys 0.000000 (best of 3)
! wall 1.190563 comb 1.190000 user 1.190000 sys 0.000000 (best of 9)
$ hg perflrucachedict --size 1000 --gets 1000000 --sets 1000000 --mixed 1000000 --costlimit 10000 --mixedgetfreq 90
! inserts
! wall 0.711177 comb 0.710000 user 0.710000 sys 0.000000 (best of 14)
! sets
! wall 0.846992 comb 0.850000 user 0.850000 sys 0.000000 (best of 12)
! inserts w/ cost limit
! wall 25.963028 comb 25.960000 user 25.960000 sys 0.000000 (best of 3)
! wall 2.184311 comb 2.180000 user 2.180000 sys 0.000000 (best of 5)
! mixed
! wall 0.728256 comb 0.730000 user 0.730000 sys 0.000000 (best of 14)
! mixed w/ cost limit
! wall 3.174256 comb 3.170000 user 3.170000 sys 0.000000 (best of 4)
! wall 0.773186 comb 0.770000 user 0.770000 sys 0.000000 (best of 13)
$ hg perflrucachedict --size 100000 --gets 1000000 --sets 1000000 --mixed 1000000 --mixedgetfreq 90 --costlimit 5000000
! gets
! wall 1.191368 comb 1.190000 user 1.190000 sys 0.000000 (best of 9)
! wall 1.195304 comb 1.190000 user 1.190000 sys 0.000000 (best of 9)
! inserts
! wall 0.950995 comb 0.950000 user 0.950000 sys 0.000000 (best of 11)
! inserts w/ cost limit
! wall 1.589732 comb 1.590000 user 1.590000 sys 0.000000 (best of 7)
! sets
! wall 1.094941 comb 1.100000 user 1.090000 sys 0.010000 (best of 9)
! mixed
! wall 0.936420 comb 0.940000 user 0.930000 sys 0.010000 (best of 10)
! mixed w/ cost limit
! wall 0.882780 comb 0.870000 user 0.870000 sys 0.000000 (best of 11)
This puts us ~2x slower than caches without cost accounting. And for
read-heavy workloads (the prime use cases for caches), performance is
nearly identical.
In the worst case (pure write workloads with cost accounting enabled),
we're looking at ~1.5us per insert on large caches. That seems "fast
enough."
Differential Revision: https://phab.mercurial-scm.org/D4505
$ 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 >> $TESTTMP/dirstaterace.py << EOF
> from mercurial import (
> context,
> extensions,
> )
> def extsetup():
> extensions.wrapfunction(context.workingctx, '_checklookup', overridechecklookup)
> def overridechecklookup(orig, self, files):
> # make an update that changes the dirstate from underneath
> self._repo.ui.system(r"sh '$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