This is an archive of the discontinued Mercurial Phabricator instance.

Differential D4131

dirstate: use visitchildrenset in traverse
ClosedPublic

Authored by spectral on Aug 6 2018, 2:56 PM.

Download Raw Diff

Details

Reviewers

None

Group Reviewers

hg-reviewers

Commits

rHGa3cabe9415e1: dirstate: use visitchildrenset in traverse

Summary

This speeds up hg status a fair amount when there is a very large directory
and narrow is in use.

Timing numbers according to command:

hyperfine --warmup 1 'hg status'

HGRCPATH points to a file with the following contents:

[extensions]
narrow =

mozilla-unified (called m-u below) was at revision #468856.

      regular hash: eb39298e432d
treemanifests hash: 0553b7f29eaf

large-dir-repo (called l-d-r below) was generated with the following script:

#!/bin/bash
hg init large-dir-repo
mkdir -p large-dir-repo/third_party/rust/log
touch large-dir-repo/third_party/rust/log/foo.txt
for i in $(seq 1 30000); do
    d=$(mktemp -d large-dir-repo/third_party/XXXXXXXXX)
    touch $d/file.txt
done
hg -R large-dir-repo ci -Am 'rev0' --user test --date '0 0'

for repos that use narrow, the narrowspec was this:

[includes]
rootfilesin:third_party/rust/log
[excludes]

This narrowspec was chosen due to the size of the third_party/rust directory;
this directory was *not* modified in revision #468856 in mozilla-unified.

Importantly, when using narrow, these repos had everything checked out (in the
case of large-dir-repo, that means all 30,001 directories), *before* adding the
narrowspec. This is to simulate the behavior when using a virtual filesystem
that shows everything for the user even if they haven't added it to the
narrowspec yet. This is not a supported configuration, and hg update will not
really do the "correct" thing, but non-mutating commands should behave
correctly.

There are two repos below that do not follow the setup above, 'citc1' and
'citc2', which are using a virtual filesystem and can not be reproduced
upstream; these numbers are here mostly to indicate that these performance
improvements are not hypothetical, and show the benefits we're hoping to achieve
on our real workloads. 'citc1' is closest to large-dir-repo with one of our
pathological cases, 'citc2' is an arbitrary repo and closer to "average".

I'm not claiming anything less than a 5% speed win as improvements due to this
change; these are probably eiter measurement artifacts or constant time
improvements. The numbers that aren't changing are shown primarily to prove that
this doesn't make anything worse in any case I plan on testing during this
series.

'before' is hg from commit c83ad576. 'N' indicates narrow in use, 'T' indicates
treemanifest in use.

hg status:
repo  | N | T | before (mean +- stdev) | after (mean +- stdev) | % of before
------+---+---+------------------------+-----------------------+------------
m-u   |   |   | 2.284 s +-  0.022 s    | 2.274 s +-  0.021 s   |  99.6%
m-u   |   | x | 2.289 s +-  0.008 s    | 2.284 s +-  0.028 s   |  99.8%
m-u   | x |   | 430.8 ms +-   3.1 ms   | 424.5 ms +-   3.2 ms  |  98.5%
m-u   | x | x | 429.8 ms +-   2.5 ms   | 425.8 ms +-   3.7 ms  |  99.1%
l-d-r |   |   | 681.3 ms +-   5.5 ms   | 689.6 ms +-   8.0 ms  | 101.2%
l-d-r |   | x | 666.8 ms +-  21.8 ms   | 672.5 ms +-  14.9 ms  | 100.9%
l-d-r | x |   | 282.6 ms +-   1.8 ms   | 203.0 ms +-   1.2 ms  |  71.8% <--
l-d-r | x | x | 275.2 ms +-   3.9 ms   | 199.3 ms +-   3.5 ms  |  72.4% <--
citc1 | x | x | 1.023 s +-  0.011 s    | 398.6 ms +-   9.2 ms  |  39.0% <--
citc2 | x | x | 297.9 ms +-   4.4 ms   | 289.6 ms +-   4.2 ms  |  97.2%

hg status --change .:
repo  | N | T | before (mean +- stdev) | after (mean +- stdev) | % of before
------+---+---+------------------------+-----------------------+------------
m-u   |   |   | 478.2 ms +-   2.0 ms   | 476.9 ms +-   3.7 ms  |  99.7%
m-u   |   | x | 169.5 ms +-   2.7 ms   | 169.5 ms +-   2.5 ms  | 100.0%
m-u   | x |   | 477.0 ms +-   2.4 ms   | 476.1 ms +-   1.4 ms  |  99.8%
m-u   | x | x | 124.7 ms +-   1.9 ms   | 124.2 ms +-   3.3 ms  |  99.6%
l-d-r |   |   | 97.4 ms +-   1.2 ms    | 96.5 ms +-   1.2 ms   |  99.1%
l-d-r |   | x | 4.778 s +-  0.018 s    | 4.774 s +-  0.011 s   |  99.9%
l-d-r | x |   | 99.9 ms +-   1.1 ms    | 98.8 ms +-   1.3 ms   |  98.9%
l-d-r | x | x | 848.7 ms +-   7.1 ms   | 849.4 ms +-   6.5 ms  | 100.1%
citc1 | x | x | 4.250 s +-  0.051 s    | 4.283 s +-  0.042 s   | 100.8%
citc2 | x | x | 341.5 ms +-   4.7 ms   | 341.5 ms +-   4.1 ms  | 100.0%

hg update $rev^; hg update $rev:
repo  | N | T | before (mean +- stdev) | after (mean +- stdev) | % of before
------+---+---+------------------------+-----------------------+------------
m-u   |   |   | 4.357 s +-  0.032 s    | 4.312 s +-  0.093 s   |  99.0%
m-u   |   | x | 3.599 s +-  0.061 s    | 3.592 s +-  0.071 s   |  99.8%
m-u   | x |   | 1.815 s +-  0.012 s    | 1.816 s +-  0.013 s   | 100.1%
m-u   | x | x | 1.110 s +-  0.009 s    | 1.106 s +-  0.005 s   |  99.6%
l-d-r |   |   | 527.1 ms +-   7.8 ms   | 523.3 ms +-   6.5 ms  |  99.3%
l-d-r |   | x | 8.835 s +-  0.067 s    | 8.825 s +-  0.064 s   |  99.9%
l-d-r | x |   | 313.0 ms +-   2.2 ms   | 312.1 ms +-   1.2 ms  |  99.7%
l-d-r | x | x | 1.780 s +-  0.011 s    | 1.799 s +-  0.013 s   | 101.1%
citc1 | x | x | 6.825 s +-  0.262 s    | 6.707 s +-  0.353 s   |  98.3%
citc2 | x | x | 776.4 ms +-   4.5 ms   | 781.3 ms +-   6.3 ms  | 100.6%

hg diff:
repo  | N | T | before (mean +- stdev) | after (mean +- stdev) | % of before
------+---+---+------------------------+-----------------------+------------
m-u   |   |   | 1.519 s +-  0.015 s    | 1.525 s +-  0.017 s   | 100.4%
m-u   |   | x | 1.512 s +-  0.010 s    | 1.517 s +-  0.027 s   | 100.3%
m-u   | x |   | 420.0 ms +-   3.2 ms   | 417.1 ms +-   1.9 ms  |  99.3%
m-u   | x | x | 415.0 ms +-   3.8 ms   | 415.7 ms +-   2.7 ms  | 100.2%
l-d-r |   |   | 220.8 ms +-   4.0 ms   | 220.8 ms +-   3.7 ms  | 100.0%
l-d-r |   | x | 216.6 ms +-   7.5 ms   | 211.4 ms +-   2.1 ms  |  97.6%
l-d-r | x |   | 111.9 ms +-   1.8 ms   | 112.0 ms +-   1.5 ms  | 100.1%
l-d-r | x | x | 111.4 ms +-   1.4 ms   | 110.2 ms +-   1.0 ms  |  98.9%
citc1 | x | x | 268.7 ms +-   2.3 ms   | 269.6 ms +-   2.8 ms  | 100.3%
citc2 | x | x | 273.5 ms +-   5.5 ms   | 273.9 ms +-   3.7 ms  | 100.1%

hg diff -c .:
repo  | N | T | before (mean +- stdev) | after (mean +- stdev) | % of before
------+---+---+--------------------------+-----------------------+----------
m-u   |   |   | 497.1 ms +-   1.4 ms   | 500.1 ms +-   2.4 ms  | 100.6%
m-u   |   | x | 195.3 ms +-  13.2 ms   | 191.6 ms +-   3.0 ms  |  98.1%
m-u   | x |   | 476.8 ms +-   1.9 ms   | 476.7 ms +-   2.3 ms  | 100.0%
m-u   | x | x | 122.8 ms +-   2.1 ms   | 122.9 ms +-   2.0 ms  | 100.1%
l-d-r |   |   | 99.3 ms +-   2.3 ms    | 98.8 ms +-   1.7 ms   |  99.5%
l-d-r |   | x | 4.875 s +-  0.041 s    | 4.847 s +-  0.038 s   |  99.4%
l-d-r | x |   | 98.5 ms +-   1.2 ms    | 98.9 ms +-   1.3 ms   | 100.4%
l-d-r | x | x | 864.6 ms +-   7.4 ms   | 855.4 ms +-   6.6 ms  |  98.9%
citc1 | x | x | 4.505 s +-  0.060 s    | 4.466 s +-  0.036 s   |  99.1%
citc2 | x | x | 368.0 ms +-   4.0 ms   | 365.5 ms +-   6.3 ms  |  99.3%

Diff Detail

Repository

rHG Mercurial

Lint

Automatic diff as part of commit; lint not applicable.

Unit

Automatic diff as part of commit; unit tests not applicable.

Event Timeline

spectral created this revision.Aug 6 2018, 2:56 PM

Herald added a reviewer: hg-reviewers. · View Herald TranscriptAug 6 2018, 2:56 PM

Herald added a subscriber: mercurial-devel. · View Herald Transcript

spectral added a parent revision: D4130: match: add visitchildrenset complement to visitdir.Aug 6 2018, 3:08 PM

spectral planned changes to this revision.Aug 6 2018, 3:56 PM

spectral edited the summary of this revision. (Show Details)Aug 6 2018, 5:17 PM

spectral requested review of this revision.Aug 6 2018, 8:19 PM

Queued the series, thanks.

wadd = work.append
while work:
    nd = work.pop()
if not match.visitdir(nd):

+ visitentries = match.visitchildrenset(nd)
+ if not visitentries:
continue
+ if visitentries == 'this' or visitentries == 'all':
+ visitentries = None
skip = None
if nd == '.':
    nd = ''
@@ -909,6 +912,8 @@
        continue
    raise
for f, kind, st in entries:
+ if visitentries and f not in visitentries:
+ continue

Correct me if I get it wrong:

visitentries is a list of direct sub-directories.
if there are any files to visit, visitentries should be either 'this' or 'all'.
so we don't have to check if f is a directory or not.

Closed by commit rHGa3cabe9415e1: dirstate: use visitchildrenset in traverse (authored by spectral). · Explain WhyAug 8 2018, 11:19 AM

This revision was automatically updated to reflect the committed changes.

In D4131#64347, @yuja wrote:
Queued the series, thanks.
wadd = work.append
while work:
    nd = work.pop()
if not match.visitdir(nd):

+ visitentries = match.visitchildrenset(nd)
+ if not visitentries:
continue
+ if visitentries == 'this' or visitentries == 'all':
+ visitentries = None
skip = None
if nd == '.':
    nd = ''
@@ -909,6 +912,8 @@
        continue
    raise
for f, kind, st in entries:
+ if visitentries and f not in visitentries:
+ continue
Correct me if I get it wrong:

visitentries is a list of direct sub-directories.

if there are any files to visit, visitentries should be either 'this' or 'all'.

so we don't have to check if f is a directory or not.

This is correct. I'll add a comment for future maintainers.

Revision Contents
Changeset List

			Path	Packages
M			mercurial/dirstate.py (7 lines)

Status	Author	Revision
Closed	spectral	D4131 dirstate: use visitchildrenset in traverse
Closed	spectral	D4130 match: add visitchildrenset complement to visitdir
Closed	spectral	D4129 includematcher: separate "parents" from "dirs"
Closed	spectral	D4128 match: add tests for visitdir functionality

Diff 10076

mercurial/dirstate.py

	skipstep3 = skipstep3 and not (work or dirsnotfound)			skipstep3 = skipstep3 and not (work or dirsnotfound)
	work = [d for d in work if not dirignore(d[0])]			work = [d for d in work if not dirignore(d[0])]

	# step 2: visit subdirectories			# step 2: visit subdirectories
	def traverse(work, alreadynormed):			def traverse(work, alreadynormed):
	wadd = work.append			wadd = work.append
	while work:			while work:
	nd = work.pop()			nd = work.pop()
	if not match.visitdir(nd):			visitentries = match.visitchildrenset(nd)
				if not visitentries:
	continue			continue
				if visitentries == 'this' or visitentries == 'all':
				visitentries = None
	skip = None			skip = None
	if nd == '.':			if nd == '.':
	nd = ''			nd = ''
	else:			else:
	skip = '.hg'			skip = '.hg'
	try:			try:
	entries = listdir(join(nd), stat=True, skip=skip)			entries = listdir(join(nd), stat=True, skip=skip)
	except OSError as inst:			except OSError as inst:
	if inst.errno in (errno.EACCES, errno.ENOENT):			if inst.errno in (errno.EACCES, errno.ENOENT):
	match.bad(self.pathto(nd),			match.bad(self.pathto(nd),
	encoding.strtolocal(inst.strerror))			encoding.strtolocal(inst.strerror))
	continue			continue
	raise			raise
	for f, kind, st in entries:			for f, kind, st in entries:
				if visitentries and f not in visitentries:
				continue
	if normalizefile:			if normalizefile:
	# even though f might be a directory, we're only			# even though f might be a directory, we're only
	# interested in comparing it to files currently in the			# interested in comparing it to files currently in the
	# dmap -- therefore normalizefile is enough			# dmap -- therefore normalizefile is enough
	nf = normalizefile(nd and (nd + "/" + f) or f, True,			nf = normalizefile(nd and (nd + "/" + f) or f, True,
	True)			True)
	else:			else:
	nf = nd and (nd + "/" + f) or f			nf = nd and (nd + "/" + f) or f

Diff	ID	Description	Created	Lint	Unit
Base		Base
Diff 1	10002		Aug 6 2018, 2:56 PM	★	★
Diff 2	10076	rHGa3cabe9415e161f701aeeed978e387747f5abcc8	Jul 31 2018, 7:47 PM	★	★