Mercurial > hg
view hgext/convert/convcmd.py @ 26380:56a640b0f656
revlog: don't flush data file after every added revision
The current behavior of revlogs is to flush the data file when writing
data to it. Tracing system calls revealed that changegroup processing
incurred numerous write(2) calls for values much smaller than the
default buffer size (Python defaults to 4096, but it can be adjusted
based on detected block size at run time by CPython).
The reason we flush revlogs is so readers have all data available.
For example, the current code in revlog.py will re-open the revlog
file (instead of seeking an existing file handle) to read the text
of a revision. This happens when starting a new delta chain when
adding several revisions from changegroups, for example. Yes, this
is likely sub-optimal (we should probably be sharing file descriptors
between readers and writers to avoid the flushing and associated
overhead of re-opening files).
While flushing revlogs is necessary, it appears all callers are
diligent about flushing files before a read is performed (see
buildtext() in _addrevision()), making the flush in
_writeentry() redundant and unncessary. So, we remove it. In practice,
this means we incur a write(2) a) when the buffer is full (typically
4096 bytes) b) when a new delta chain is created rather than after
every added revision. This applies to every revlog, but by volume
it mostly impacts filelogs.
Removing the redundant flush from _writeentry() significantly
reduces the number of write(2) calls during changegroup processing on
my Linux machine. When applying a changegroup of the hg repo based on
my local repo, the total number of write(2) calls during application
of the mercurial/localrepo.py revlogs dropped from 1,320 to 217 with
this patch applied. Total I/O related system calls dropped from 1,577
to 474.
When unbundling a mozilla-central gzipped bundle (264,403 changesets
with 1,492,215 changes to 222,507 files), total write(2) calls
dropped from 1,252,881 to 827,106 and total system calls dropped from
3,601,259 to 3,178,636 - a reduction of 425,775!
While the system call reduction is significant, it appears
to have no impact on wall time on my Linux and Windows machines. Still,
fewer syscalls is fewer syscalls. Surely this can't hurt. If nothing
else, it makes examining remaining system call usage simpler and opens
the door to experimenting with the performance impact of different
buffer sizes.
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Sat, 26 Sep 2015 21:43:13 -0700 |
parents | 86598f4fe1cf |
children | 56b2bcea2529 |
line wrap: on
line source
# convcmd - convert extension commands definition # # Copyright 2005-2007 Matt Mackall <mpm@selenic.com> # # This software may be used and distributed according to the terms of the # GNU General Public License version 2 or any later version. from common import NoRepo, MissingTool, SKIPREV, mapfile from cvs import convert_cvs from darcs import darcs_source from git import convert_git from hg import mercurial_source, mercurial_sink from subversion import svn_source, svn_sink from monotone import monotone_source from gnuarch import gnuarch_source from bzr import bzr_source from p4 import p4_source import filemap import os, shutil, shlex from mercurial import hg, util, encoding from mercurial.i18n import _ orig_encoding = 'ascii' def recode(s): if isinstance(s, unicode): return s.encode(orig_encoding, 'replace') else: return s.decode('utf-8').encode(orig_encoding, 'replace') def mapbranch(branch, branchmap): ''' >>> bmap = {'default': 'branch1'} >>> for i in ['', None]: ... mapbranch(i, bmap) 'branch1' 'branch1' >>> bmap = {'None': 'branch2'} >>> for i in ['', None]: ... mapbranch(i, bmap) 'branch2' 'branch2' >>> bmap = {'None': 'branch3', 'default': 'branch4'} >>> for i in ['None', '', None, 'default', 'branch5']: ... mapbranch(i, bmap) 'branch3' 'branch4' 'branch4' 'branch4' 'branch5' ''' # If branch is None or empty, this commit is coming from the source # repository's default branch and destined for the default branch in the # destination repository. For such commits, using a literal "default" # in branchmap below allows the user to map "default" to an alternate # default branch in the destination repository. branch = branchmap.get(branch or 'default', branch) # At some point we used "None" literal to denote the default branch, # attempt to use that for backward compatibility. if (not branch): branch = branchmap.get(str(None), branch) return branch source_converters = [ ('cvs', convert_cvs, 'branchsort'), ('git', convert_git, 'branchsort'), ('svn', svn_source, 'branchsort'), ('hg', mercurial_source, 'sourcesort'), ('darcs', darcs_source, 'branchsort'), ('mtn', monotone_source, 'branchsort'), ('gnuarch', gnuarch_source, 'branchsort'), ('bzr', bzr_source, 'branchsort'), ('p4', p4_source, 'branchsort'), ] sink_converters = [ ('hg', mercurial_sink), ('svn', svn_sink), ] def convertsource(ui, path, type, revs): exceptions = [] if type and type not in [s[0] for s in source_converters]: raise util.Abort(_('%s: invalid source repository type') % type) for name, source, sortmode in source_converters: try: if not type or name == type: return source(ui, path, revs), sortmode except (NoRepo, MissingTool) as inst: exceptions.append(inst) if not ui.quiet: for inst in exceptions: ui.write("%s\n" % inst) raise util.Abort(_('%s: missing or unsupported repository') % path) def convertsink(ui, path, type): if type and type not in [s[0] for s in sink_converters]: raise util.Abort(_('%s: invalid destination repository type') % type) for name, sink in sink_converters: try: if not type or name == type: return sink(ui, path) except NoRepo as inst: ui.note(_("convert: %s\n") % inst) except MissingTool as inst: raise util.Abort('%s\n' % inst) raise util.Abort(_('%s: unknown repository type') % path) class progresssource(object): def __init__(self, ui, source, filecount): self.ui = ui self.source = source self.filecount = filecount self.retrieved = 0 def getfile(self, file, rev): self.retrieved += 1 self.ui.progress(_('getting files'), self.retrieved, item=file, total=self.filecount) return self.source.getfile(file, rev) def targetfilebelongstosource(self, targetfilename): return self.source.targetfilebelongstosource(targetfilename) def lookuprev(self, rev): return self.source.lookuprev(rev) def close(self): self.ui.progress(_('getting files'), None) class converter(object): def __init__(self, ui, source, dest, revmapfile, opts): self.source = source self.dest = dest self.ui = ui self.opts = opts self.commitcache = {} self.authors = {} self.authorfile = None # Record converted revisions persistently: maps source revision # ID to target revision ID (both strings). (This is how # incremental conversions work.) self.map = mapfile(ui, revmapfile) # Read first the dst author map if any authorfile = self.dest.authorfile() if authorfile and os.path.exists(authorfile): self.readauthormap(authorfile) # Extend/Override with new author map if necessary if opts.get('authormap'): self.readauthormap(opts.get('authormap')) self.authorfile = self.dest.authorfile() self.splicemap = self.parsesplicemap(opts.get('splicemap')) self.branchmap = mapfile(ui, opts.get('branchmap')) def parsesplicemap(self, path): """ check and validate the splicemap format and return a child/parents dictionary. Format checking has two parts. 1. generic format which is same across all source types 2. specific format checking which may be different for different source type. This logic is implemented in checkrevformat function in source files like hg.py, subversion.py etc. """ if not path: return {} m = {} try: fp = open(path, 'r') for i, line in enumerate(fp): line = line.splitlines()[0].rstrip() if not line: # Ignore blank lines continue # split line lex = shlex.shlex(line, posix=True) lex.whitespace_split = True lex.whitespace += ',' line = list(lex) # check number of parents if not (2 <= len(line) <= 3): raise util.Abort(_('syntax error in %s(%d): child parent1' '[,parent2] expected') % (path, i + 1)) for part in line: self.source.checkrevformat(part) child, p1, p2 = line[0], line[1:2], line[2:] if p1 == p2: m[child] = p1 else: m[child] = p1 + p2 # if file does not exist or error reading, exit except IOError: raise util.Abort(_('splicemap file not found or error reading %s:') % path) return m def walktree(self, heads): '''Return a mapping that identifies the uncommitted parents of every uncommitted changeset.''' visit = heads known = set() parents = {} numcommits = self.source.numcommits() while visit: n = visit.pop(0) if n in known: continue if n in self.map: m = self.map[n] if m == SKIPREV or self.dest.hascommitfrommap(m): continue known.add(n) self.ui.progress(_('scanning'), len(known), unit=_('revisions'), total=numcommits) commit = self.cachecommit(n) parents[n] = [] for p in commit.parents: parents[n].append(p) visit.append(p) self.ui.progress(_('scanning'), None) return parents def mergesplicemap(self, parents, splicemap): """A splicemap redefines child/parent relationships. Check the map contains valid revision identifiers and merge the new links in the source graph. """ for c in sorted(splicemap): if c not in parents: if not self.dest.hascommitforsplicemap(self.map.get(c, c)): # Could be in source but not converted during this run self.ui.warn(_('splice map revision %s is not being ' 'converted, ignoring\n') % c) continue pc = [] for p in splicemap[c]: # We do not have to wait for nodes already in dest. if self.dest.hascommitforsplicemap(self.map.get(p, p)): continue # Parent is not in dest and not being converted, not good if p not in parents: raise util.Abort(_('unknown splice map parent: %s') % p) pc.append(p) parents[c] = pc def toposort(self, parents, sortmode): '''Return an ordering such that every uncommitted changeset is preceded by all its uncommitted ancestors.''' def mapchildren(parents): """Return a (children, roots) tuple where 'children' maps parent revision identifiers to children ones, and 'roots' is the list of revisions without parents. 'parents' must be a mapping of revision identifier to its parents ones. """ visit = sorted(parents) seen = set() children = {} roots = [] while visit: n = visit.pop(0) if n in seen: continue seen.add(n) # Ensure that nodes without parents are present in the # 'children' mapping. children.setdefault(n, []) hasparent = False for p in parents[n]: if p not in self.map: visit.append(p) hasparent = True children.setdefault(p, []).append(n) if not hasparent: roots.append(n) return children, roots # Sort functions are supposed to take a list of revisions which # can be converted immediately and pick one def makebranchsorter(): """If the previously converted revision has a child in the eligible revisions list, pick it. Return the list head otherwise. Branch sort attempts to minimize branch switching, which is harmful for Mercurial backend compression. """ prev = [None] def picknext(nodes): next = nodes[0] for n in nodes: if prev[0] in parents[n]: next = n break prev[0] = next return next return picknext def makesourcesorter(): """Source specific sort.""" keyfn = lambda n: self.commitcache[n].sortkey def picknext(nodes): return sorted(nodes, key=keyfn)[0] return picknext def makeclosesorter(): """Close order sort.""" keyfn = lambda n: ('close' not in self.commitcache[n].extra, self.commitcache[n].sortkey) def picknext(nodes): return sorted(nodes, key=keyfn)[0] return picknext def makedatesorter(): """Sort revisions by date.""" dates = {} def getdate(n): if n not in dates: dates[n] = util.parsedate(self.commitcache[n].date) return dates[n] def picknext(nodes): return min([(getdate(n), n) for n in nodes])[1] return picknext if sortmode == 'branchsort': picknext = makebranchsorter() elif sortmode == 'datesort': picknext = makedatesorter() elif sortmode == 'sourcesort': picknext = makesourcesorter() elif sortmode == 'closesort': picknext = makeclosesorter() else: raise util.Abort(_('unknown sort mode: %s') % sortmode) children, actives = mapchildren(parents) s = [] pendings = {} while actives: n = picknext(actives) actives.remove(n) s.append(n) # Update dependents list for c in children.get(n, []): if c not in pendings: pendings[c] = [p for p in parents[c] if p not in self.map] try: pendings[c].remove(n) except ValueError: raise util.Abort(_('cycle detected between %s and %s') % (recode(c), recode(n))) if not pendings[c]: # Parents are converted, node is eligible actives.insert(0, c) pendings[c] = None if len(s) != len(parents): raise util.Abort(_("not all revisions were sorted")) return s def writeauthormap(self): authorfile = self.authorfile if authorfile: self.ui.status(_('writing author map file %s\n') % authorfile) ofile = open(authorfile, 'w+') for author in self.authors: ofile.write("%s=%s\n" % (author, self.authors[author])) ofile.close() def readauthormap(self, authorfile): afile = open(authorfile, 'r') for line in afile: line = line.strip() if not line or line.startswith('#'): continue try: srcauthor, dstauthor = line.split('=', 1) except ValueError: msg = _('ignoring bad line in author map file %s: %s\n') self.ui.warn(msg % (authorfile, line.rstrip())) continue srcauthor = srcauthor.strip() dstauthor = dstauthor.strip() if self.authors.get(srcauthor) in (None, dstauthor): msg = _('mapping author %s to %s\n') self.ui.debug(msg % (srcauthor, dstauthor)) self.authors[srcauthor] = dstauthor continue m = _('overriding mapping for author %s, was %s, will be %s\n') self.ui.status(m % (srcauthor, self.authors[srcauthor], dstauthor)) afile.close() def cachecommit(self, rev): commit = self.source.getcommit(rev) commit.author = self.authors.get(commit.author, commit.author) commit.branch = mapbranch(commit.branch, self.branchmap) self.commitcache[rev] = commit return commit def copy(self, rev): commit = self.commitcache[rev] full = self.opts.get('full') changes = self.source.getchanges(rev, full) if isinstance(changes, basestring): if changes == SKIPREV: dest = SKIPREV else: dest = self.map[changes] self.map[rev] = dest return files, copies, cleanp2 = changes pbranches = [] if commit.parents: for prev in commit.parents: if prev not in self.commitcache: self.cachecommit(prev) pbranches.append((self.map[prev], self.commitcache[prev].branch)) self.dest.setbranch(commit.branch, pbranches) try: parents = self.splicemap[rev] self.ui.status(_('spliced in %s as parents of %s\n') % (parents, rev)) parents = [self.map.get(p, p) for p in parents] except KeyError: parents = [b[0] for b in pbranches] if len(pbranches) != 2: cleanp2 = set() if len(parents) < 3: source = progresssource(self.ui, self.source, len(files)) else: # For an octopus merge, we end up traversing the list of # changed files N-1 times. This tweak to the number of # files makes it so the progress bar doesn't overflow # itself. source = progresssource(self.ui, self.source, len(files) * (len(parents) - 1)) newnode = self.dest.putcommit(files, copies, parents, commit, source, self.map, full, cleanp2) source.close() self.source.converted(rev, newnode) self.map[rev] = newnode def convert(self, sortmode): try: self.source.before() self.dest.before() self.source.setrevmap(self.map) self.ui.status(_("scanning source...\n")) heads = self.source.getheads() parents = self.walktree(heads) self.mergesplicemap(parents, self.splicemap) self.ui.status(_("sorting...\n")) t = self.toposort(parents, sortmode) num = len(t) c = None self.ui.status(_("converting...\n")) for i, c in enumerate(t): num -= 1 desc = self.commitcache[c].desc if "\n" in desc: desc = desc.splitlines()[0] # convert log message to local encoding without using # tolocal() because the encoding.encoding convert() # uses is 'utf-8' self.ui.status("%d %s\n" % (num, recode(desc))) self.ui.note(_("source: %s\n") % recode(c)) self.ui.progress(_('converting'), i, unit=_('revisions'), total=len(t)) self.copy(c) self.ui.progress(_('converting'), None) if not self.ui.configbool('convert', 'skiptags'): tags = self.source.gettags() ctags = {} for k in tags: v = tags[k] if self.map.get(v, SKIPREV) != SKIPREV: ctags[k] = self.map[v] if c and ctags: nrev, tagsparent = self.dest.puttags(ctags) if nrev and tagsparent: # write another hash correspondence to override the # previous one so we don't end up with extra tag heads tagsparents = [e for e in self.map.iteritems() if e[1] == tagsparent] if tagsparents: self.map[tagsparents[0][0]] = nrev bookmarks = self.source.getbookmarks() cbookmarks = {} for k in bookmarks: v = bookmarks[k] if self.map.get(v, SKIPREV) != SKIPREV: cbookmarks[k] = self.map[v] if c and cbookmarks: self.dest.putbookmarks(cbookmarks) self.writeauthormap() finally: self.cleanup() def cleanup(self): try: self.dest.after() finally: self.source.after() self.map.close() def convert(ui, src, dest=None, revmapfile=None, **opts): global orig_encoding orig_encoding = encoding.encoding encoding.encoding = 'UTF-8' # support --authors as an alias for --authormap if not opts.get('authormap'): opts['authormap'] = opts.get('authors') if not dest: dest = hg.defaultdest(src) + "-hg" ui.status(_("assuming destination %s\n") % dest) destc = convertsink(ui, dest, opts.get('dest_type')) try: srcc, defaultsort = convertsource(ui, src, opts.get('source_type'), opts.get('rev')) except Exception: for path in destc.created: shutil.rmtree(path, True) raise sortmodes = ('branchsort', 'datesort', 'sourcesort', 'closesort') sortmode = [m for m in sortmodes if opts.get(m)] if len(sortmode) > 1: raise util.Abort(_('more than one sort mode specified')) if sortmode: sortmode = sortmode[0] else: sortmode = defaultsort if sortmode == 'sourcesort' and not srcc.hasnativeorder(): raise util.Abort(_('--sourcesort is not supported by this data source')) if sortmode == 'closesort' and not srcc.hasnativeclose(): raise util.Abort(_('--closesort is not supported by this data source')) fmap = opts.get('filemap') if fmap: srcc = filemap.filemap_source(ui, srcc, fmap) destc.setfilemapmode(True) if not revmapfile: revmapfile = destc.revmapfile() c = converter(ui, srcc, destc, revmapfile, opts) c.convert(sortmode)