Mercurial > hg
view contrib/synthrepo.py @ 39506:b66ea3fc3a86
sparse-revlog: set max delta chain length to on thousand
The new snapshot system used in the sparse-revlog case gave us some small size
benefit so far. However its most important property is to gracefully handle
harder limit on delta chainlength.
Long delta chain has a very detrimental impact on read (and write) performance
in revlog. Being able to shorter them provide a great boost. However, shorting
delta used to result significantly lower compression ratio. The intermediate
snapshots effectively suppress most of this effect (even all in some case).
# Effect on the test repository
The repository we use for test is not "realistic" but can still show this in
action using an unreasonably low chain limit. Limiting the chain length show a
sizeable increase but stay under control: +6% for limit=15; +15% for limit=10.
Without the snapshot system the increase is significantly bigger: +45% for
limit=15; +80% for limit=10. Even slightly larger than without delta chain
limit, the resulting size is still smaller than before we started doing
snapshots.
Here is a table for comparison. *Since the repository is not branchy, the
initial sparse-revlog version does not bring much benefit compare to the
non-sparse one):
chain length limit | none | limit=15 | limit=10 |
without sparse-revlog | 62 818 987 | 112 664 615 | 131 222 574 |
without snapshot | 74 365 490 | 108 211 410 | 133 857 764 |
with snapshot | 59 230 936 | 63 002 924 | 68 415 329 |
# Effect On Real Life Repositories
The series provides significant benefits on all kind of repositories.
Using `hg debugupgraderepo -o redeltaparent --run`, we recomputed delta chain
for various repositories with different settings:
- delta chain length: unlimited or 1000 limit
- sparse-revlog: enabled or disabled
- this series: applied or not applied
We can observe multiple types of effect:
- On very branchy repositories:
* The delta chain limit as low impact on the repo size.
* Intermediate snapshot greatly reduces manifest size:
- pypy: -80%
- netbeans: -95%
* The delta chain limit is effective, without a size impact:
- netbeans average: 613 -> 282
- private #1 average: 1 068 -> 307
- On more linear repository:
* Intermediate snapshot limit the impact of delta chain limit:
- mozilla:
without the series: +360%
with the series: +25%
* The delta chain limit provides large improvement:
- mozilla's average chain length:
unlimited: 15 338
limited: 469
* Despite the chain length limit, the manifest size is reduced:
- mercurial: -25%
- mozilla: -30%
It is clear that the use of chains of intermediate snapshots provide large
benefits both in storage size and delta chains quality. We should now switch our
effort toward making sure the write performance are acceptable. Then,
`sparse-revlog` will be a suitable format for all new repository.
# Raw Statistic
* no-sparse: general delta repository not using sparse-revlog
* no-snapshot: sparse-revlog repository not using this series
* snapshot: sparse-revlog repository using this series
mercurial
Manifest Size:
limit | none | 1000
------------|-------------|------------
no-sparse | 8 021 373 | 8 199 366
no-snapshot | 8 103 561 | 8 259 719
snapshot | 6 137 116 | 6 126 433
Manifest Chain length data
limit || none || 1000 ||
value || average | max || average | max ||
------------||---------|---------||---------|---------||
no-sparse || 307 | 1456 || 279 | 1000 ||
no-snapshot || 312 | 1456 || 283 | 1000 ||
snapshot || 248 | 1208 || 241 | 1000 ||
Full Store Size
limit | none | 1000
------------|-------------|------------
no-sparse | 51 013 198 | 51 201 574
no-snapshot | 50 930 795 | 51 141 006
snapshot | 48 072 037 | 48 093 572
pypy
Manifest Size:
limit | none | 1000
------------|-------------|------------
no-sparse | 193 987 784 | 193 987 784
no-snapshot | 163 171 745 | 163 312 229
snapshot | 34 605 900 | 34 600 750
Manifest Chain length data
limit || none || 1000 ||
value || average | max || average | max ||
------------||---------|---------||---------|---------||
no-sparse || 101 | 692 || 101 | 692 ||
no-snapshot || 151 | 1307 || 148 | 1000 ||
snapshot || 128 | 1309 || 125 | 1000 ||
Full Store Size
limit | none | 1000
------------|-------------|------------
no-sparse | 495 931 473 | 495 931 473
no-snapshot | 465 441 017 | 465 581 501
snapshot | 355 467 301 | 355 472 451
Mozilla
Manifest Size:
limit | none | 1000
------------|----------------|---------------
no-sparse | 416 757 148 | 1 869 009 668
no-snapshot | 401 592 370 | 1 843 493 795
snapshot | 224 359 521 | 284 615 500
Manifest Chain length data
limit || none || 1000 ||
value || average | max || average | max ||
------------||---------|---------||---------|---------||
no-sparse || 15 333 | 58 980 || 468 | 1 000 ||
no-snapshot || 15 336 | 58 980 || 469 | 1 000 ||
snapshot || 15 338 | 58 983 || 469 | 1 000 ||
Full Store Size
limit | none | 1000
------------|----------------|---------------
no-sparse | 2 712 477 887 | 4 164 995 451
no-snapshot | 2 698 887 835 | 4 141 054 304
snapshot | 2 518 130 385 | 2 578 587 596
Netbeans
Manifest Size:
limit | none | 1000
------------|----------------|---------------
no-sparse | 4 766 794 101 | 4 870 642 687
no-snapshot | 4 334 806 082 | 4 428 681 309
snapshot | 232 659 666 | 240 330 665
Manifest Chain length data
limit || none || 1000 ||
value || average | max || average | max ||
------------||---------|---------||---------|---------||
no-sparse || 597 | 6802 || 254 | 1 000 ||
no-snapshot || 648 | 6 802 || 305 | 1 000 ||
snapshot || 613 | 6 804 || 282 | 1 000 ||
Full Store Size
limit | none | 1000
------------|----------------|---------------
no-sparse | 5 807 347 998 | 5 911 196 584
no-snapshot | 5 375 398 602 | 5 469 273 829
snapshot | 1 282 519 928 | 1 290 190 927
Private repo #1
Manifest Size:
limit | none | 1000
------------|-----------------|---------------
no-sparse | 41 389 010 840 | 41 398 162 091
no-snapshot | 9 737 319 435 | 10 223 773 150
snapshot | 744 215 807 | 747 961 822
Manifest Chain length data
limit || none || 1000 ||
value || average | max || average | max ||
------------||---------|---------||---------|---------||
no-sparse || 245 | 8 885 || 81 | 1 000 ||
no-snapshot || 1 225 | 8 885 || 336 | 1 000 ||
snapshot || 1 068 | 7 909 || 307 | 1 000 ||
Full Store Size
limit | none | 1000
------------|----------------|---------------
no-sparse | 49 646 065 126 | 49 655 216 377
no-snapshot | 17 924 862 856 | 18 411 316 571
snapshot | 9 009 024 710 | 9 012 770 725
Private repo #2
We currently have less data available for that repository.
* Before is a sparse-revlog repository without this series
* After is a sparse-revlog repository with this series + 1000 chain limit
Manifest Size:
Before: 1 531 485 040 bytes
After: 1 091 422 451 bytes
Manifest Chain:
Before: 2 218 avg; 6 575 Max
After: 442 avg; 1 000 Max
Full Store Size
Before: 15 203 955 615
after: 8 207 180 693
author | Boris Feld <boris.feld@octobus.net> |
---|---|
date | Fri, 07 Sep 2018 11:18:45 -0400 |
parents | 1c93e0237a24 |
children | 876494fd967d |
line wrap: on
line source
# synthrepo.py - repo synthesis # # Copyright 2012 Facebook # # This software may be used and distributed according to the terms of the # GNU General Public License version 2 or any later version. '''synthesize structurally interesting change history This extension is useful for creating a repository with properties that are statistically similar to an existing repository. During analysis, a simple probability table is constructed from the history of an existing repository. During synthesis, these properties are reconstructed. Properties that are analyzed and synthesized include the following: - Lines added or removed when an existing file is modified - Number and sizes of files added - Number of files removed - Line lengths - Topological distance to parent changeset(s) - Probability of a commit being a merge - Probability of a newly added file being added to a new directory - Interarrival time, and time zone, of commits - Number of files in each directory A few obvious properties that are not currently handled realistically: - Merges are treated as regular commits with two parents, which is not realistic - Modifications are not treated as operations on hunks of lines, but as insertions and deletions of randomly chosen single lines - Committer ID (always random) - Executability of files - Symlinks and binary files are ignored ''' from __future__ import absolute_import import bisect import collections import itertools import json import os import random import sys import time from mercurial.i18n import _ from mercurial.node import ( nullid, nullrev, short, ) from mercurial import ( context, diffutil, error, hg, patch, registrar, scmutil, ) from mercurial.utils import ( dateutil, ) # Note for extension authors: ONLY specify testedwith = 'ships-with-hg-core' for # extensions which SHIP WITH MERCURIAL. Non-mainline extensions should # be specifying the version(s) of Mercurial they are tested with, or # leave the attribute unspecified. testedwith = 'ships-with-hg-core' cmdtable = {} command = registrar.command(cmdtable) newfile = {'new fi', 'rename', 'copy f', 'copy t'} def zerodict(): return collections.defaultdict(lambda: 0) def roundto(x, k): if x > k * 2: return int(round(x / float(k)) * k) return int(round(x)) def parsegitdiff(lines): filename, mar, lineadd, lineremove = None, None, zerodict(), 0 binary = False for line in lines: start = line[:6] if start == 'diff -': if filename: yield filename, mar, lineadd, lineremove, binary mar, lineadd, lineremove, binary = 'm', zerodict(), 0, False filename = patch.gitre.match(line).group(1) elif start in newfile: mar = 'a' elif start == 'GIT bi': binary = True elif start == 'delete': mar = 'r' elif start: s = start[0] if s == '-' and not line.startswith('--- '): lineremove += 1 elif s == '+' and not line.startswith('+++ '): lineadd[roundto(len(line) - 1, 5)] += 1 if filename: yield filename, mar, lineadd, lineremove, binary @command('analyze', [('o', 'output', '', _('write output to given file'), _('FILE')), ('r', 'rev', [], _('analyze specified revisions'), _('REV'))], _('hg analyze'), optionalrepo=True) def analyze(ui, repo, *revs, **opts): '''create a simple model of a repository to use for later synthesis This command examines every changeset in the given range (or all of history if none are specified) and creates a simple statistical model of the history of the repository. It also measures the directory structure of the repository as checked out. The model is written out to a JSON file, and can be used by :hg:`synthesize` to create or augment a repository with synthetic commits that have a structure that is statistically similar to the analyzed repository. ''' root = repo.root if not root.endswith(os.path.sep): root += os.path.sep revs = list(revs) revs.extend(opts['rev']) if not revs: revs = [':'] output = opts['output'] if not output: output = os.path.basename(root) + '.json' if output == '-': fp = sys.stdout else: fp = open(output, 'w') # Always obtain file counts of each directory in the given root directory. def onerror(e): ui.warn(_('error walking directory structure: %s\n') % e) dirs = {} rootprefixlen = len(root) for dirpath, dirnames, filenames in os.walk(root, onerror=onerror): dirpathfromroot = dirpath[rootprefixlen:] dirs[dirpathfromroot] = len(filenames) if '.hg' in dirnames: dirnames.remove('.hg') lineschanged = zerodict() children = zerodict() p1distance = zerodict() p2distance = zerodict() linesinfilesadded = zerodict() fileschanged = zerodict() filesadded = zerodict() filesremoved = zerodict() linelengths = zerodict() interarrival = zerodict() parents = zerodict() dirsadded = zerodict() tzoffset = zerodict() # If a mercurial repo is available, also model the commit history. if repo: revs = scmutil.revrange(repo, revs) revs.sort() progress = ui.makeprogress(_('analyzing'), unit=_('changesets'), total=len(revs)) for i, rev in enumerate(revs): progress.update(i) ctx = repo[rev] pl = ctx.parents() pctx = pl[0] prev = pctx.rev() children[prev] += 1 p1distance[rev - prev] += 1 parents[len(pl)] += 1 tzoffset[ctx.date()[1]] += 1 if len(pl) > 1: p2distance[rev - pl[1].rev()] += 1 if prev == rev - 1: lastctx = pctx else: lastctx = repo[rev - 1] if lastctx.rev() != nullrev: timedelta = ctx.date()[0] - lastctx.date()[0] interarrival[roundto(timedelta, 300)] += 1 diffopts = diffutil.diffallopts(ui, {'git': True}) diff = sum((d.splitlines() for d in ctx.diff(pctx, opts=diffopts)), []) fileadds, diradds, fileremoves, filechanges = 0, 0, 0, 0 for filename, mar, lineadd, lineremove, isbin in parsegitdiff(diff): if isbin: continue added = sum(lineadd.itervalues(), 0) if mar == 'm': if added and lineremove: lineschanged[roundto(added, 5), roundto(lineremove, 5)] += 1 filechanges += 1 elif mar == 'a': fileadds += 1 if '/' in filename: filedir = filename.rsplit('/', 1)[0] if filedir not in pctx.dirs(): diradds += 1 linesinfilesadded[roundto(added, 5)] += 1 elif mar == 'r': fileremoves += 1 for length, count in lineadd.iteritems(): linelengths[length] += count fileschanged[filechanges] += 1 filesadded[fileadds] += 1 dirsadded[diradds] += 1 filesremoved[fileremoves] += 1 progress.complete() invchildren = zerodict() for rev, count in children.iteritems(): invchildren[count] += 1 if output != '-': ui.status(_('writing output to %s\n') % output) def pronk(d): return sorted(d.iteritems(), key=lambda x: x[1], reverse=True) json.dump({'revs': len(revs), 'initdirs': pronk(dirs), 'lineschanged': pronk(lineschanged), 'children': pronk(invchildren), 'fileschanged': pronk(fileschanged), 'filesadded': pronk(filesadded), 'linesinfilesadded': pronk(linesinfilesadded), 'dirsadded': pronk(dirsadded), 'filesremoved': pronk(filesremoved), 'linelengths': pronk(linelengths), 'parents': pronk(parents), 'p1distance': pronk(p1distance), 'p2distance': pronk(p2distance), 'interarrival': pronk(interarrival), 'tzoffset': pronk(tzoffset), }, fp) fp.close() @command('synthesize', [('c', 'count', 0, _('create given number of commits'), _('COUNT')), ('', 'dict', '', _('path to a dictionary of words'), _('FILE')), ('', 'initfiles', 0, _('initial file count to create'), _('COUNT'))], _('hg synthesize [OPTION].. DESCFILE')) def synthesize(ui, repo, descpath, **opts): '''synthesize commits based on a model of an existing repository The model must have been generated by :hg:`analyze`. Commits will be generated randomly according to the probabilities described in the model. If --initfiles is set, the repository will be seeded with the given number files following the modeled repository's directory structure. When synthesizing new content, commit descriptions, and user names, words will be chosen randomly from a dictionary that is presumed to contain one word per line. Use --dict to specify the path to an alternate dictionary to use. ''' try: fp = hg.openpath(ui, descpath) except Exception as err: raise error.Abort('%s: %s' % (descpath, err[0].strerror)) desc = json.load(fp) fp.close() def cdf(l): if not l: return [], [] vals, probs = zip(*sorted(l, key=lambda x: x[1], reverse=True)) t = float(sum(probs, 0)) s, cdfs = 0, [] for v in probs: s += v cdfs.append(s / t) return vals, cdfs lineschanged = cdf(desc['lineschanged']) fileschanged = cdf(desc['fileschanged']) filesadded = cdf(desc['filesadded']) dirsadded = cdf(desc['dirsadded']) filesremoved = cdf(desc['filesremoved']) linelengths = cdf(desc['linelengths']) parents = cdf(desc['parents']) p1distance = cdf(desc['p1distance']) p2distance = cdf(desc['p2distance']) interarrival = cdf(desc['interarrival']) linesinfilesadded = cdf(desc['linesinfilesadded']) tzoffset = cdf(desc['tzoffset']) dictfile = opts.get('dict') or '/usr/share/dict/words' try: fp = open(dictfile, 'rU') except IOError as err: raise error.Abort('%s: %s' % (dictfile, err.strerror)) words = fp.read().splitlines() fp.close() initdirs = {} if desc['initdirs']: for k, v in desc['initdirs']: initdirs[k.encode('utf-8').replace('.hg', '_hg')] = v initdirs = renamedirs(initdirs, words) initdirscdf = cdf(initdirs) def pick(cdf): return cdf[0][bisect.bisect_left(cdf[1], random.random())] def pickpath(): return os.path.join(pick(initdirscdf), random.choice(words)) def makeline(minimum=0): total = max(minimum, pick(linelengths)) c, l = 0, [] while c < total: w = random.choice(words) c += len(w) + 1 l.append(w) return ' '.join(l) wlock = repo.wlock() lock = repo.lock() nevertouch = {'.hgsub', '.hgignore', '.hgtags'} _synthesizing = _('synthesizing') _files = _('initial files') _changesets = _('changesets') # Synthesize a single initial revision adding files to the repo according # to the modeled directory structure. initcount = int(opts['initfiles']) if initcount and initdirs: pctx = repo[None].parents()[0] dirs = set(pctx.dirs()) files = {} def validpath(path): # Don't pick filenames which are already directory names. if path in dirs: return False # Don't pick directories which were used as file names. while path: if path in files: return False path = os.path.dirname(path) return True progress = ui.makeprogress(_synthesizing, unit=_files, total=initcount) for i in xrange(0, initcount): progress.update(i) path = pickpath() while not validpath(path): path = pickpath() data = '%s contents\n' % path files[path] = data dir = os.path.dirname(path) while dir and dir not in dirs: dirs.add(dir) dir = os.path.dirname(dir) def filectxfn(repo, memctx, path): return context.memfilectx(repo, memctx, path, files[path]) progress.complete() message = 'synthesized wide repo with %d files' % (len(files),) mc = context.memctx(repo, [pctx.node(), nullid], message, files, filectxfn, ui.username(), '%d %d' % dateutil.makedate()) initnode = mc.commit() if ui.debugflag: hexfn = hex else: hexfn = short ui.status(_('added commit %s with %d files\n') % (hexfn(initnode), len(files))) # Synthesize incremental revisions to the repository, adding repo depth. count = int(opts['count']) heads = set(map(repo.changelog.rev, repo.heads())) progress = ui.makeprogress(_synthesizing, unit=_changesets, total=count) for i in xrange(count): progress.update(i) node = repo.changelog.node revs = len(repo) def pickhead(heads, distance): if heads: lheads = sorted(heads) rev = revs - min(pick(distance), revs) if rev < lheads[-1]: rev = lheads[bisect.bisect_left(lheads, rev)] else: rev = lheads[-1] return rev, node(rev) return nullrev, nullid r1 = revs - min(pick(p1distance), revs) p1 = node(r1) # the number of heads will grow without bound if we use a pure # model, so artificially constrain their proliferation toomanyheads = len(heads) > random.randint(1, 20) if p2distance[0] and (pick(parents) == 2 or toomanyheads): r2, p2 = pickhead(heads.difference([r1]), p2distance) else: r2, p2 = nullrev, nullid pl = [p1, p2] pctx = repo[r1] mf = pctx.manifest() mfk = mf.keys() changes = {} if mfk: for __ in xrange(pick(fileschanged)): for __ in xrange(10): fctx = pctx.filectx(random.choice(mfk)) path = fctx.path() if not (path in nevertouch or fctx.isbinary() or 'l' in fctx.flags()): break lines = fctx.data().splitlines() add, remove = pick(lineschanged) for __ in xrange(remove): if not lines: break del lines[random.randrange(0, len(lines))] for __ in xrange(add): lines.insert(random.randint(0, len(lines)), makeline()) path = fctx.path() changes[path] = '\n'.join(lines) + '\n' for __ in xrange(pick(filesremoved)): path = random.choice(mfk) for __ in xrange(10): path = random.choice(mfk) if path not in changes: break if filesadded: dirs = list(pctx.dirs()) dirs.insert(0, '') for __ in xrange(pick(filesadded)): pathstr = '' while pathstr in dirs: path = [random.choice(dirs)] if pick(dirsadded): path.append(random.choice(words)) path.append(random.choice(words)) pathstr = '/'.join(filter(None, path)) data = '\n'.join(makeline() for __ in xrange(pick(linesinfilesadded))) + '\n' changes[pathstr] = data def filectxfn(repo, memctx, path): if path not in changes: return None return context.memfilectx(repo, memctx, path, changes[path]) if not changes: continue if revs: date = repo['tip'].date()[0] + pick(interarrival) else: date = time.time() - (86400 * count) # dates in mercurial must be positive, fit in 32-bit signed integers. date = min(0x7fffffff, max(0, date)) user = random.choice(words) + '@' + random.choice(words) mc = context.memctx(repo, pl, makeline(minimum=2), sorted(changes), filectxfn, user, '%d %d' % (date, pick(tzoffset))) newnode = mc.commit() heads.add(repo.changelog.rev(newnode)) heads.discard(r1) heads.discard(r2) progress.complete() lock.release() wlock.release() def renamedirs(dirs, words): '''Randomly rename the directory names in the per-dir file count dict.''' wordgen = itertools.cycle(words) replacements = {'': ''} def rename(dirpath): '''Recursively rename the directory and all path prefixes. The mapping from path to renamed path is stored for all path prefixes as in dynamic programming, ensuring linear runtime and consistent renaming regardless of iteration order through the model. ''' if dirpath in replacements: return replacements[dirpath] head, _ = os.path.split(dirpath) if head: head = rename(head) else: head = '' renamed = os.path.join(head, next(wordgen)) replacements[dirpath] = renamed return renamed result = [] for dirpath, count in dirs.iteritems(): result.append([rename(dirpath.lstrip(os.sep)), count]) return result