Mercurial > hg
view hgext/convert/gnuarch.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 | baea47cafe75 |
children | 56b2bcea2529 |
line wrap: on
line source
# gnuarch.py - GNU Arch support for the convert extension # # Copyright 2008, 2009 Aleix Conchillo Flaque <aleix@member.fsf.org> # and others # # 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, commandline, commit, converter_source from mercurial.i18n import _ from mercurial import encoding, util import os, shutil, tempfile, stat from email.Parser import Parser class gnuarch_source(converter_source, commandline): class gnuarch_rev(object): def __init__(self, rev): self.rev = rev self.summary = '' self.date = None self.author = '' self.continuationof = None self.add_files = [] self.mod_files = [] self.del_files = [] self.ren_files = {} self.ren_dirs = {} def __init__(self, ui, path, revs=None): super(gnuarch_source, self).__init__(ui, path, revs=revs) if not os.path.exists(os.path.join(path, '{arch}')): raise NoRepo(_("%s does not look like a GNU Arch repository") % path) # Could use checktool, but we want to check for baz or tla. self.execmd = None if util.findexe('baz'): self.execmd = 'baz' else: if util.findexe('tla'): self.execmd = 'tla' else: raise util.Abort(_('cannot find a GNU Arch tool')) commandline.__init__(self, ui, self.execmd) self.path = os.path.realpath(path) self.tmppath = None self.treeversion = None self.lastrev = None self.changes = {} self.parents = {} self.tags = {} self.catlogparser = Parser() self.encoding = encoding.encoding self.archives = [] def before(self): # Get registered archives self.archives = [i.rstrip('\n') for i in self.runlines0('archives', '-n')] if self.execmd == 'tla': output = self.run0('tree-version', self.path) else: output = self.run0('tree-version', '-d', self.path) self.treeversion = output.strip() # Get name of temporary directory version = self.treeversion.split('/') self.tmppath = os.path.join(tempfile.gettempdir(), 'hg-%s' % version[1]) # Generate parents dictionary self.parents[None] = [] treeversion = self.treeversion child = None while treeversion: self.ui.status(_('analyzing tree version %s...\n') % treeversion) archive = treeversion.split('/')[0] if archive not in self.archives: self.ui.status(_('tree analysis stopped because it points to ' 'an unregistered archive %s...\n') % archive) break # Get the complete list of revisions for that tree version output, status = self.runlines('revisions', '-r', '-f', treeversion) self.checkexit(status, 'failed retrieving revisions for %s' % treeversion) # No new iteration unless a revision has a continuation-of header treeversion = None for l in output: rev = l.strip() self.changes[rev] = self.gnuarch_rev(rev) self.parents[rev] = [] # Read author, date and summary catlog, status = self.run('cat-log', '-d', self.path, rev) if status: catlog = self.run0('cat-archive-log', rev) self._parsecatlog(catlog, rev) # Populate the parents map self.parents[child].append(rev) # Keep track of the current revision as the child of the next # revision scanned child = rev # Check if we have to follow the usual incremental history # or if we have to 'jump' to a different treeversion given # by the continuation-of header. if self.changes[rev].continuationof: treeversion = '--'.join( self.changes[rev].continuationof.split('--')[:-1]) break # If we reached a base-0 revision w/o any continuation-of # header, it means the tree history ends here. if rev[-6:] == 'base-0': break def after(self): self.ui.debug('cleaning up %s\n' % self.tmppath) shutil.rmtree(self.tmppath, ignore_errors=True) def getheads(self): return self.parents[None] def getfile(self, name, rev): if rev != self.lastrev: raise util.Abort(_('internal calling inconsistency')) if not os.path.lexists(os.path.join(self.tmppath, name)): return None, None return self._getfile(name, rev) def getchanges(self, rev, full): if full: raise util.Abort(_("convert from arch do not support --full")) self._update(rev) changes = [] copies = {} for f in self.changes[rev].add_files: changes.append((f, rev)) for f in self.changes[rev].mod_files: changes.append((f, rev)) for f in self.changes[rev].del_files: changes.append((f, rev)) for src in self.changes[rev].ren_files: to = self.changes[rev].ren_files[src] changes.append((src, rev)) changes.append((to, rev)) copies[to] = src for src in self.changes[rev].ren_dirs: to = self.changes[rev].ren_dirs[src] chgs, cps = self._rendirchanges(src, to) changes += [(f, rev) for f in chgs] copies.update(cps) self.lastrev = rev return sorted(set(changes)), copies, set() def getcommit(self, rev): changes = self.changes[rev] return commit(author=changes.author, date=changes.date, desc=changes.summary, parents=self.parents[rev], rev=rev) def gettags(self): return self.tags def _execute(self, cmd, *args, **kwargs): cmdline = [self.execmd, cmd] cmdline += args cmdline = [util.shellquote(arg) for arg in cmdline] cmdline += ['>', os.devnull, '2>', os.devnull] cmdline = util.quotecommand(' '.join(cmdline)) self.ui.debug(cmdline, '\n') return os.system(cmdline) def _update(self, rev): self.ui.debug('applying revision %s...\n' % rev) changeset, status = self.runlines('replay', '-d', self.tmppath, rev) if status: # Something went wrong while merging (baz or tla # issue?), get latest revision and try from there shutil.rmtree(self.tmppath, ignore_errors=True) self._obtainrevision(rev) else: old_rev = self.parents[rev][0] self.ui.debug('computing changeset between %s and %s...\n' % (old_rev, rev)) self._parsechangeset(changeset, rev) def _getfile(self, name, rev): mode = os.lstat(os.path.join(self.tmppath, name)).st_mode if stat.S_ISLNK(mode): data = os.readlink(os.path.join(self.tmppath, name)) if mode: mode = 'l' else: mode = '' else: data = open(os.path.join(self.tmppath, name), 'rb').read() mode = (mode & 0o111) and 'x' or '' return data, mode def _exclude(self, name): exclude = ['{arch}', '.arch-ids', '.arch-inventory'] for exc in exclude: if name.find(exc) != -1: return True return False def _readcontents(self, path): files = [] contents = os.listdir(path) while len(contents) > 0: c = contents.pop() p = os.path.join(path, c) # os.walk could be used, but here we avoid internal GNU # Arch files and directories, thus saving a lot time. if not self._exclude(p): if os.path.isdir(p): contents += [os.path.join(c, f) for f in os.listdir(p)] else: files.append(c) return files def _rendirchanges(self, src, dest): changes = [] copies = {} files = self._readcontents(os.path.join(self.tmppath, dest)) for f in files: s = os.path.join(src, f) d = os.path.join(dest, f) changes.append(s) changes.append(d) copies[d] = s return changes, copies def _obtainrevision(self, rev): self.ui.debug('obtaining revision %s...\n' % rev) output = self._execute('get', rev, self.tmppath) self.checkexit(output) self.ui.debug('analyzing revision %s...\n' % rev) files = self._readcontents(self.tmppath) self.changes[rev].add_files += files def _stripbasepath(self, path): if path.startswith('./'): return path[2:] return path def _parsecatlog(self, data, rev): try: catlog = self.catlogparser.parsestr(data) # Commit date self.changes[rev].date = util.datestr( util.strdate(catlog['Standard-date'], '%Y-%m-%d %H:%M:%S')) # Commit author self.changes[rev].author = self.recode(catlog['Creator']) # Commit description self.changes[rev].summary = '\n\n'.join((catlog['Summary'], catlog.get_payload())) self.changes[rev].summary = self.recode(self.changes[rev].summary) # Commit revision origin when dealing with a branch or tag if 'Continuation-of' in catlog: self.changes[rev].continuationof = self.recode( catlog['Continuation-of']) except Exception: raise util.Abort(_('could not parse cat-log of %s') % rev) def _parsechangeset(self, data, rev): for l in data: l = l.strip() # Added file (ignore added directory) if l.startswith('A') and not l.startswith('A/'): file = self._stripbasepath(l[1:].strip()) if not self._exclude(file): self.changes[rev].add_files.append(file) # Deleted file (ignore deleted directory) elif l.startswith('D') and not l.startswith('D/'): file = self._stripbasepath(l[1:].strip()) if not self._exclude(file): self.changes[rev].del_files.append(file) # Modified binary file elif l.startswith('Mb'): file = self._stripbasepath(l[2:].strip()) if not self._exclude(file): self.changes[rev].mod_files.append(file) # Modified link elif l.startswith('M->'): file = self._stripbasepath(l[3:].strip()) if not self._exclude(file): self.changes[rev].mod_files.append(file) # Modified file elif l.startswith('M'): file = self._stripbasepath(l[1:].strip()) if not self._exclude(file): self.changes[rev].mod_files.append(file) # Renamed file (or link) elif l.startswith('=>'): files = l[2:].strip().split(' ') if len(files) == 1: files = l[2:].strip().split('\t') src = self._stripbasepath(files[0]) dst = self._stripbasepath(files[1]) if not self._exclude(src) and not self._exclude(dst): self.changes[rev].ren_files[src] = dst # Conversion from file to link or from link to file (modified) elif l.startswith('ch'): file = self._stripbasepath(l[2:].strip()) if not self._exclude(file): self.changes[rev].mod_files.append(file) # Renamed directory elif l.startswith('/>'): dirs = l[2:].strip().split(' ') if len(dirs) == 1: dirs = l[2:].strip().split('\t') src = self._stripbasepath(dirs[0]) dst = self._stripbasepath(dirs[1]) if not self._exclude(src) and not self._exclude(dst): self.changes[rev].ren_dirs[src] = dst