( )⚙ D4148 perf: add a command to benchmark linelog edits

This is an archive of the discontinued Mercurial Phabricator instance.

Differential D4148

perf: add a command to benchmark linelog edits
ClosedPublic

Authored by quark on Aug 6 2018, 9:59 PM.

Details

Reviewers
durin42
Group Reviewers
hg-reviewers
Commits
rHG1601afbb573c: perf: add a command to benchmark linelog edits
Summary

The use pattern of creating a linelog is usually by calling "replacelines"
multiple times. Add a command to benchmark it.

Diff Detail

Repository
rHG Mercurial
Lint
Lint Skipped
Unit
Unit Tests Skipped

Event Timeline

quark created this revision.Aug 6 2018, 9:59 PM
quark updated this revision to Diff 10061.Aug 7 2018, 12:01 PM
indygreg added a subscriber: indygreg.Aug 7 2018, 7:33 PM
indygreg added inline comments.
contrib/perf.py
904–912

I'm a bit concerned about the use of random data in a benchmark. Since every run will be different, the benchmark may not be stable. And if the random source is slow, we could be spending time waiting on entropy.

Do you think it is worthwhile to precompute the arguments to replacelines() and have the benchmark simply iterate through a list of arguments and call replacelines() repeatedly?

quark added inline comments.Aug 7 2018, 8:15 PM
contrib/perf.py
904–912

I think random.seed(0) makes it stable?

Good point about pre-calculate randints. Will change.

quark updated this revision to Diff 10065.Aug 7 2018, 8:22 PM
durin42 accepted this revision.Aug 9 2018, 1:53 PM
durin42 added a subscriber: durin42.

I'm confident the random.seed(0) gives us the stability between runs we're concerned about.

This revision is now accepted and ready to land.Aug 9 2018, 1:53 PM
Closed by commit rHG1601afbb573c: perf: add a command to benchmark linelog edits (authored by quark). · Explain WhyAug 9 2018, 2:06 PM
This revision was automatically updated to reflect the committed changes.

Revision Contents
Changeset List

PathPackages
Mcontrib/perf.py (32 lines)
Mtests/test-contrib-perf.t (3 lines)

Diff 10065

contrib/perf.py

fm.end() fm.end()
@command('perflookup', formatteropts) @command('perflookup', formatteropts)
def perflookup(ui, repo, rev, **opts): def perflookup(ui, repo, rev, **opts):
timer, fm = gettimer(ui, opts) timer, fm = gettimer(ui, opts)
timer(lambda: len(repo.lookup(rev))) timer(lambda: len(repo.lookup(rev)))
fm.end() fm.end()
@command('perflinelogedits',
[('n', 'edits', 10000, 'number of edits'),
('', 'max-hunk-lines', 10, 'max lines in a hunk'),
], norepo=True)
def perflinelogedits(ui, **opts):
from mercurial import linelog
edits = opts['edits']
maxhunklines = opts['max_hunk_lines']
maxb1 = 100000
random.seed(0)
randint = random.randint
currentlines = 0
arglist = []
for rev in xrange(edits):
a1 = randint(0, currentlines)
a2 = randint(a1, min(currentlines, a1 + maxhunklines))
b1 = randint(0, maxb1)
b2 = randint(b1, b1 + maxhunklines)
currentlines += (b2 - b1) - (a2 - a1)
indygregUnsubmitted
Not Done

I'm a bit concerned about the use of random data in a benchmark. Since every run will be different, the benchmark may not be stable. And if the random source is slow, we could be spending time waiting on entropy.

Do you think it is worthwhile to precompute the arguments to replacelines() and have the benchmark simply iterate through a list of arguments and call replacelines() repeatedly?

indygreg: I'm a bit concerned about the use of random data in a benchmark. Since every run will be…
quarkAuthorUnsubmitted
Not Done

I think random.seed(0) makes it stable?

Good point about pre-calculate randints. Will change.

quark: I think `random.seed(0)` makes it stable? Good point about pre-calculate randints. Will change.
arglist.append((rev, a1, a2, b1, b2))
def d():
ll = linelog.linelog()
for args in arglist:
ll.replacelines(*args)
timer, fm = gettimer(ui, opts)
timer(d)
fm.end()
@command('perfrevrange', formatteropts) @command('perfrevrange', formatteropts)
def perfrevrange(ui, repo, *specs, **opts): def perfrevrange(ui, repo, *specs, **opts):
timer, fm = gettimer(ui, opts) timer, fm = gettimer(ui, opts)
revrange = scmutil.revrange revrange = scmutil.revrange
timer(lambda: len(revrange(repo, specs))) timer(lambda: len(revrange(repo, specs)))
fm.end() fm.end()
@command('perfnodelookup', formatteropts) @command('perfnodelookup', formatteropts)

tests/test-contrib-perf.t

perffncacheencode perffncacheencode
(no help text available) (no help text available)
perffncacheload perffncacheload
(no help text available) (no help text available)
perffncachewrite perffncachewrite
(no help text available) (no help text available)
perfheads (no help text available) perfheads (no help text available)
perfindex (no help text available) perfindex (no help text available)
perflinelogedits
(no help text available)
perfloadmarkers perfloadmarkers
benchmark the time to parse the on-disk markers for a repo benchmark the time to parse the on-disk markers for a repo
perflog (no help text available) perflog (no help text available)
perflookup (no help text available) perflookup (no help text available)
perflrucachedict perflrucachedict
(no help text available) (no help text available)
perfmanifest benchmark the time to read a manifest from disk and return a perfmanifest benchmark the time to read a manifest from disk and return a
usable usable
$ hg debugrebuildfncache $ hg debugrebuildfncache
fncache already up to date fncache already up to date
$ hg perffncachewrite $ hg perffncachewrite
$ hg debugrebuildfncache $ hg debugrebuildfncache
fncache already up to date fncache already up to date
#endif #endif
$ hg perfheads $ hg perfheads
$ hg perfindex $ hg perfindex
$ hg perflinelogedits -n 1
$ hg perfloadmarkers $ hg perfloadmarkers
$ hg perflog $ hg perflog
$ hg perflookup 2 $ hg perflookup 2
$ hg perflrucache $ hg perflrucache
$ hg perfmanifest 2 $ hg perfmanifest 2
$ hg perfmergecalculate -r 3 $ hg perfmergecalculate -r 3
$ hg perfmoonwalk $ hg perfmoonwalk
$ hg perfnodelookup 2 $ hg perfnodelookup 2