changegroup: compute seen files as changesets are added (issue4750)
Before this patch, addchangegroup() would walk the changelog and compute
the set of seen files between applying changesets and applying
manifests. When cloning large repositories such as mozilla-central,
this consumed a non-trivial amount of time. On my MBP, this walk takes
~10s. On a dainty EC2 instance, this was measured to take ~125s! On the
latter machine, this delay was enough for the Mercurial server to
disconnect the client, thinking it had timed out, thus causing a clone
to abort.
This patch enables the changelog to compute the set of changed files as
new revisions are added. By doing so, we:
* avoid a potentially heavy computation between changelog and manifest
processing by spreading the computation across all changelog additions
* avoid extra reads from the changelog by operating on the data as it is
added
The downside of this is that the add revision callback does result in
extra I/O. Before, we would perform a flush (and subsequent read to
construct the full revision) when new delta chains were created. For
changelogs, this is typically every 2-4 revisions. Using the callback
guarantees there will be a flush after every added revision *and* an
open + read of the changelog to obtain the full revision in order to
read the added files. So, this increases the frequency of these
operations by the average chain length. In the future, the revlog
should be smart enough to know how to read revisions that haven't been
flushed yet, thus eliminating this extra I/O.
On my MBP, the total CPU times for an `hg unbundle` with a local
mozilla-central gzip bundle containing 251,934 changesets and 211,065
files did not have a statistically significant change with this patch,
holding steady around 360s. So, the increased revlog flushing did not
have an effect.
With this patch, there is no longer a visible pause between applying
changeset and manifest data. Before, it sure felt like Mercurial was
lethargic making this transition. Now, the transition is nearly
instantaneous, giving the impression that Mercurial is faster. Of course,
eliminating this pause means that the potential for network disconnect due
to channel inactivity during the changelog walk is eliminated as well.
And that is the impetus behind this change.
# monotone.py - monotone support for the convert extension
#
# Copyright 2008, 2009 Mikkel Fahnoe Jorgensen <mikkel@dvide.com> 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.
import os, re
from mercurial import util
from common import NoRepo, commit, converter_source, checktool
from common import commandline
from mercurial.i18n import _
class monotone_source(converter_source, commandline):
def __init__(self, ui, path=None, revs=None):
converter_source.__init__(self, ui, path, revs)
if revs and len(revs) > 1:
raise util.Abort(_('monotone source does not support specifying '
'multiple revs'))
commandline.__init__(self, ui, 'mtn')
self.ui = ui
self.path = path
self.automatestdio = False
self.revs = revs
norepo = NoRepo(_("%s does not look like a monotone repository")
% path)
if not os.path.exists(os.path.join(path, '_MTN')):
# Could be a monotone repository (SQLite db file)
try:
f = file(path, 'rb')
header = f.read(16)
f.close()
except IOError:
header = ''
if header != 'SQLite format 3\x00':
raise norepo
# regular expressions for parsing monotone output
space = r'\s*'
name = r'\s+"((?:\\"|[^"])*)"\s*'
value = name
revision = r'\s+\[(\w+)\]\s*'
lines = r'(?:.|\n)+'
self.dir_re = re.compile(space + "dir" + name)
self.file_re = re.compile(space + "file" + name +
"content" + revision)
self.add_file_re = re.compile(space + "add_file" + name +
"content" + revision)
self.patch_re = re.compile(space + "patch" + name +
"from" + revision + "to" + revision)
self.rename_re = re.compile(space + "rename" + name + "to" + name)
self.delete_re = re.compile(space + "delete" + name)
self.tag_re = re.compile(space + "tag" + name + "revision" +
revision)
self.cert_re = re.compile(lines + space + "name" + name +
"value" + value)
attr = space + "file" + lines + space + "attr" + space
self.attr_execute_re = re.compile(attr + '"mtn:execute"' +
space + '"true"')
# cached data
self.manifest_rev = None
self.manifest = None
self.files = None
self.dirs = None
checktool('mtn', abort=False)
def mtnrun(self, *args, **kwargs):
if self.automatestdio:
return self.mtnrunstdio(*args, **kwargs)
else:
return self.mtnrunsingle(*args, **kwargs)
def mtnrunsingle(self, *args, **kwargs):
kwargs['d'] = self.path
return self.run0('automate', *args, **kwargs)
def mtnrunstdio(self, *args, **kwargs):
# Prepare the command in automate stdio format
command = []
for k, v in kwargs.iteritems():
command.append("%s:%s" % (len(k), k))
if v:
command.append("%s:%s" % (len(v), v))
if command:
command.insert(0, 'o')
command.append('e')
command.append('l')
for arg in args:
command += "%s:%s" % (len(arg), arg)
command.append('e')
command = ''.join(command)
self.ui.debug("mtn: sending '%s'\n" % command)
self.mtnwritefp.write(command)
self.mtnwritefp.flush()
return self.mtnstdioreadcommandoutput(command)
def mtnstdioreadpacket(self):
read = None
commandnbr = ''
while read != ':':
read = self.mtnreadfp.read(1)
if not read:
raise util.Abort(_('bad mtn packet - no end of commandnbr'))
commandnbr += read
commandnbr = commandnbr[:-1]
stream = self.mtnreadfp.read(1)
if stream not in 'mewptl':
raise util.Abort(_('bad mtn packet - bad stream type %s') % stream)
read = self.mtnreadfp.read(1)
if read != ':':
raise util.Abort(_('bad mtn packet - no divider before size'))
read = None
lengthstr = ''
while read != ':':
read = self.mtnreadfp.read(1)
if not read:
raise util.Abort(_('bad mtn packet - no end of packet size'))
lengthstr += read
try:
length = long(lengthstr[:-1])
except TypeError:
raise util.Abort(_('bad mtn packet - bad packet size %s')
% lengthstr)
read = self.mtnreadfp.read(length)
if len(read) != length:
raise util.Abort(_("bad mtn packet - unable to read full packet "
"read %s of %s") % (len(read), length))
return (commandnbr, stream, length, read)
def mtnstdioreadcommandoutput(self, command):
retval = []
while True:
commandnbr, stream, length, output = self.mtnstdioreadpacket()
self.ui.debug('mtn: read packet %s:%s:%s\n' %
(commandnbr, stream, length))
if stream == 'l':
# End of command
if output != '0':
raise util.Abort(_("mtn command '%s' returned %s") %
(command, output))
break
elif stream in 'ew':
# Error, warning output
self.ui.warn(_('%s error:\n') % self.command)
self.ui.warn(output)
elif stream == 'p':
# Progress messages
self.ui.debug('mtn: ' + output)
elif stream == 'm':
# Main stream - command output
retval.append(output)
return ''.join(retval)
def mtnloadmanifest(self, rev):
if self.manifest_rev == rev:
return
self.manifest = self.mtnrun("get_manifest_of", rev).split("\n\n")
self.manifest_rev = rev
self.files = {}
self.dirs = {}
for e in self.manifest:
m = self.file_re.match(e)
if m:
attr = ""
name = m.group(1)
node = m.group(2)
if self.attr_execute_re.match(e):
attr += "x"
self.files[name] = (node, attr)
m = self.dir_re.match(e)
if m:
self.dirs[m.group(1)] = True
def mtnisfile(self, name, rev):
# a non-file could be a directory or a deleted or renamed file
self.mtnloadmanifest(rev)
return name in self.files
def mtnisdir(self, name, rev):
self.mtnloadmanifest(rev)
return name in self.dirs
def mtngetcerts(self, rev):
certs = {"author":"<missing>", "date":"<missing>",
"changelog":"<missing>", "branch":"<missing>"}
certlist = self.mtnrun("certs", rev)
# mtn < 0.45:
# key "test@selenic.com"
# mtn >= 0.45:
# key [ff58a7ffb771907c4ff68995eada1c4da068d328]
certlist = re.split('\n\n key ["\[]', certlist)
for e in certlist:
m = self.cert_re.match(e)
if m:
name, value = m.groups()
value = value.replace(r'\"', '"')
value = value.replace(r'\\', '\\')
certs[name] = value
# Monotone may have subsecond dates: 2005-02-05T09:39:12.364306
# and all times are stored in UTC
certs["date"] = certs["date"].split('.')[0] + " UTC"
return certs
# implement the converter_source interface:
def getheads(self):
if not self.revs:
return self.mtnrun("leaves").splitlines()
else:
return self.revs
def getchanges(self, rev, full):
if full:
raise util.Abort(_("convert from monotone do not support --full"))
revision = self.mtnrun("get_revision", rev).split("\n\n")
files = {}
ignoremove = {}
renameddirs = []
copies = {}
for e in revision:
m = self.add_file_re.match(e)
if m:
files[m.group(1)] = rev
ignoremove[m.group(1)] = rev
m = self.patch_re.match(e)
if m:
files[m.group(1)] = rev
# Delete/rename is handled later when the convert engine
# discovers an IOError exception from getfile,
# but only if we add the "from" file to the list of changes.
m = self.delete_re.match(e)
if m:
files[m.group(1)] = rev
m = self.rename_re.match(e)
if m:
toname = m.group(2)
fromname = m.group(1)
if self.mtnisfile(toname, rev):
ignoremove[toname] = 1
copies[toname] = fromname
files[toname] = rev
files[fromname] = rev
elif self.mtnisdir(toname, rev):
renameddirs.append((fromname, toname))
# Directory renames can be handled only once we have recorded
# all new files
for fromdir, todir in renameddirs:
renamed = {}
for tofile in self.files:
if tofile in ignoremove:
continue
if tofile.startswith(todir + '/'):
renamed[tofile] = fromdir + tofile[len(todir):]
# Avoid chained moves like:
# d1(/a) => d3/d1(/a)
# d2 => d3
ignoremove[tofile] = 1
for tofile, fromfile in renamed.items():
self.ui.debug (_("copying file in renamed directory "
"from '%s' to '%s'")
% (fromfile, tofile), '\n')
files[tofile] = rev
copies[tofile] = fromfile
for fromfile in renamed.values():
files[fromfile] = rev
return (files.items(), copies, set())
def getfile(self, name, rev):
if not self.mtnisfile(name, rev):
return None, None
try:
data = self.mtnrun("get_file_of", name, r=rev)
except Exception:
return None, None
self.mtnloadmanifest(rev)
node, attr = self.files.get(name, (None, ""))
return data, attr
def getcommit(self, rev):
extra = {}
certs = self.mtngetcerts(rev)
if certs.get('suspend') == certs["branch"]:
extra['close'] = 1
return commit(
author=certs["author"],
date=util.datestr(util.strdate(certs["date"], "%Y-%m-%dT%H:%M:%S")),
desc=certs["changelog"],
rev=rev,
parents=self.mtnrun("parents", rev).splitlines(),
branch=certs["branch"],
extra=extra)
def gettags(self):
tags = {}
for e in self.mtnrun("tags").split("\n\n"):
m = self.tag_re.match(e)
if m:
tags[m.group(1)] = m.group(2)
return tags
def getchangedfiles(self, rev, i):
# This function is only needed to support --filemap
# ... and we don't support that
raise NotImplementedError
def before(self):
# Check if we have a new enough version to use automate stdio
version = 0.0
try:
versionstr = self.mtnrunsingle("interface_version")
version = float(versionstr)
except Exception:
raise util.Abort(_("unable to determine mtn automate interface "
"version"))
if version >= 12.0:
self.automatestdio = True
self.ui.debug("mtn automate version %s - using automate stdio\n" %
version)
# launch the long-running automate stdio process
self.mtnwritefp, self.mtnreadfp = self._run2('automate', 'stdio',
'-d', self.path)
# read the headers
read = self.mtnreadfp.readline()
if read != 'format-version: 2\n':
raise util.Abort(_('mtn automate stdio header unexpected: %s')
% read)
while read != '\n':
read = self.mtnreadfp.readline()
if not read:
raise util.Abort(_("failed to reach end of mtn automate "
"stdio headers"))
else:
self.ui.debug("mtn automate version %s - not using automate stdio "
"(automate >= 12.0 - mtn >= 0.46 is needed)\n" % version)
def after(self):
if self.automatestdio:
self.mtnwritefp.close()
self.mtnwritefp = None
self.mtnreadfp.close()
self.mtnreadfp = None