##// END OF EJS Templates
repair: migrate revlogs during upgrade...
repair: migrate revlogs during upgrade Our next step for in-place upgrade is to migrate store data. Revlogs are the biggest source of data within the store and a store is useless without them, so we implement their migration first. Our strategy for migrating revlogs is to walk the store and call `revlog.clone()` on each revlog. There are some minor complications. Because revlogs have different storage options (e.g. changelog has generaldelta and delta chains disabled), we need to obtain the correct class of revlog so inserted data is encoded properly for its type. Various attempts at implementing progress indicators that didn't lead to frustration from false "it's almost done" indicators were made. I initially used a single progress bar based on number of revlogs. However, this quickly churned through all filelogs, got to 99% then effectively froze at 99.99% when it got to the manifest. So I converted the progress bar to total revision count. This was a little bit better. But the manifest was still significantly slower than filelogs and it took forever to process the last few percent. I then tried both revision/chunk bytes and raw bytes as the denominator. This had the opposite effect: because so much data is in manifests, it would churn through filelogs without showing much progress. When it got to manifests, it would fill in 90+% of the progress bar. I finally gave up having a unified progress bar and instead implemented 3 progress bars: 1 for filelog revisions, 1 for manifest revisions, and 1 for changelog revisions. I added extra messages indicating the total number of revisions of each so users know there are more progress bars coming. I also added extra messages before and after each stage to give extra details about what is happening. Strictly speaking, this isn't necessary. But the numbers are impressive. For example, when converting a non-generaldelta mozilla-central repository, the messages you see are: migrating 2475593 total revisions (1833043 in filelogs, 321156 in manifests, 321394 in changelog) migrating 1.67 GB in store; 2508 GB tracked data migrating 267868 filelogs containing 1833043 revisions (1.09 GB in store; 57.3 GB tracked data) finished migrating 1833043 filelog revisions across 267868 filelogs; change in size: -415776 bytes migrating 1 manifests containing 321156 revisions (518 MB in store; 2451 GB tracked data) That "2508 GB" figure really blew me away. I had no clue that the raw tracked data in mozilla-central was that large. Granted, 2451 GB is in the manifest and "only" 57.3 GB is in filelogs. But still. It's worth noting that gratuitous loading of source revlogs in order to display numbers and progress bars does serve a purpose: it ensures we can open all source revlogs. We don't want to spend several minutes copying revlogs only to encounter a permissions error or similar later. As part of this commit, we also add swapping of the store directory to the upgrade function. After revlogs are converted, we move the old store into the backup directory then move the temporary repo's store into the old store's location. On well-behaved systems, this should be 2 atomic operations and the window of inconsistency show be very narrow. There are still a few improvements to be made to store copying and upgrading. But this commit gets the bulk of the work out of the way.

File last commit:

r30750:378686af default
r30779:38aa1ca9 default
Show More
chgserver.py
588 lines | 20.6 KiB | text/x-python | PythonLexer
# chgserver.py - command server extension for cHg
#
# Copyright 2011 Yuya Nishihara <yuya@tcha.org>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
"""command server extension for cHg
'S' channel (read/write)
propagate ui.system() request to client
'attachio' command
attach client's stdio passed by sendmsg()
'chdir' command
change current directory
'setenv' command
replace os.environ completely
'setumask' command
set umask
'validate' command
reload the config and check if the server is up to date
Config
------
::
[chgserver]
idletimeout = 3600 # seconds, after which an idle server will exit
skiphash = False # whether to skip config or env change checks
"""
from __future__ import absolute_import
import errno
import hashlib
import inspect
import os
import re
import struct
import time
from .i18n import _
from . import (
commandserver,
encoding,
error,
extensions,
osutil,
pycompat,
util,
)
_log = commandserver.log
def _hashlist(items):
"""return sha1 hexdigest for a list"""
return hashlib.sha1(str(items)).hexdigest()
# sensitive config sections affecting confighash
_configsections = [
'alias', # affects global state commands.table
'extdiff', # uisetup will register new commands
'extensions',
]
# sensitive environment variables affecting confighash
_envre = re.compile(r'''\A(?:
CHGHG
|HG(?:[A-Z].*)?
|LANG(?:UAGE)?
|LC_.*
|LD_.*
|PATH
|PYTHON.*
|TERM(?:INFO)?
|TZ
)\Z''', re.X)
def _confighash(ui):
"""return a quick hash for detecting config/env changes
confighash is the hash of sensitive config items and environment variables.
for chgserver, it is designed that once confighash changes, the server is
not qualified to serve its client and should redirect the client to a new
server. different from mtimehash, confighash change will not mark the
server outdated and exit since the user can have different configs at the
same time.
"""
sectionitems = []
for section in _configsections:
sectionitems.append(ui.configitems(section))
sectionhash = _hashlist(sectionitems)
envitems = [(k, v) for k, v in encoding.environ.iteritems()
if _envre.match(k)]
envhash = _hashlist(sorted(envitems))
return sectionhash[:6] + envhash[:6]
def _getmtimepaths(ui):
"""get a list of paths that should be checked to detect change
The list will include:
- extensions (will not cover all files for complex extensions)
- mercurial/__version__.py
- python binary
"""
modules = [m for n, m in extensions.extensions(ui)]
try:
from . import __version__
modules.append(__version__)
except ImportError:
pass
files = [pycompat.sysexecutable]
for m in modules:
try:
files.append(inspect.getabsfile(m))
except TypeError:
pass
return sorted(set(files))
def _mtimehash(paths):
"""return a quick hash for detecting file changes
mtimehash calls stat on given paths and calculate a hash based on size and
mtime of each file. mtimehash does not read file content because reading is
expensive. therefore it's not 100% reliable for detecting content changes.
it's possible to return different hashes for same file contents.
it's also possible to return a same hash for different file contents for
some carefully crafted situation.
for chgserver, it is designed that once mtimehash changes, the server is
considered outdated immediately and should no longer provide service.
mtimehash is not included in confighash because we only know the paths of
extensions after importing them (there is imp.find_module but that faces
race conditions). We need to calculate confighash without importing.
"""
def trystat(path):
try:
st = os.stat(path)
return (st.st_mtime, st.st_size)
except OSError:
# could be ENOENT, EPERM etc. not fatal in any case
pass
return _hashlist(map(trystat, paths))[:12]
class hashstate(object):
"""a structure storing confighash, mtimehash, paths used for mtimehash"""
def __init__(self, confighash, mtimehash, mtimepaths):
self.confighash = confighash
self.mtimehash = mtimehash
self.mtimepaths = mtimepaths
@staticmethod
def fromui(ui, mtimepaths=None):
if mtimepaths is None:
mtimepaths = _getmtimepaths(ui)
confighash = _confighash(ui)
mtimehash = _mtimehash(mtimepaths)
_log('confighash = %s mtimehash = %s\n' % (confighash, mtimehash))
return hashstate(confighash, mtimehash, mtimepaths)
def _newchgui(srcui, csystem, attachio):
class chgui(srcui.__class__):
def __init__(self, src=None):
super(chgui, self).__init__(src)
if src:
self._csystem = getattr(src, '_csystem', csystem)
else:
self._csystem = csystem
def system(self, cmd, environ=None, cwd=None, onerr=None,
errprefix=None):
# fallback to the original system method if the output needs to be
# captured (to self._buffers), or the output stream is not stdout
# (e.g. stderr, cStringIO), because the chg client is not aware of
# these situations and will behave differently (write to stdout).
if (any(s[1] for s in self._bufferstates)
or not util.safehasattr(self.fout, 'fileno')
or self.fout.fileno() != util.stdout.fileno()):
return super(chgui, self).system(cmd, environ, cwd, onerr,
errprefix)
self.flush()
rc = self._csystem(cmd, util.shellenviron(environ), cwd)
if rc and onerr:
errmsg = '%s %s' % (os.path.basename(cmd.split(None, 1)[0]),
util.explainexit(rc)[0])
if errprefix:
errmsg = '%s: %s' % (errprefix, errmsg)
raise onerr(errmsg)
return rc
def _runpager(self, cmd):
self._csystem(cmd, util.shellenviron(), type='pager',
cmdtable={'attachio': attachio})
return chgui(srcui)
def _loadnewui(srcui, args):
from . import dispatch # avoid cycle
newui = srcui.__class__.load()
for a in ['fin', 'fout', 'ferr', 'environ']:
setattr(newui, a, getattr(srcui, a))
if util.safehasattr(srcui, '_csystem'):
newui._csystem = srcui._csystem
# command line args
args = args[:]
dispatch._parseconfig(newui, dispatch._earlygetopt(['--config'], args))
# stolen from tortoisehg.util.copydynamicconfig()
for section, name, value in srcui.walkconfig():
source = srcui.configsource(section, name)
if ':' in source or source == '--config':
# path:line or command line
continue
newui.setconfig(section, name, value, source)
# load wd and repo config, copied from dispatch.py
cwds = dispatch._earlygetopt(['--cwd'], args)
cwd = cwds and os.path.realpath(cwds[-1]) or None
rpath = dispatch._earlygetopt(["-R", "--repository", "--repo"], args)
path, newlui = dispatch._getlocal(newui, rpath, wd=cwd)
return (newui, newlui)
class channeledsystem(object):
"""Propagate ui.system() request in the following format:
payload length (unsigned int),
type, '\0',
cmd, '\0',
cwd, '\0',
envkey, '=', val, '\0',
...
envkey, '=', val
if type == 'system', waits for:
exitcode length (unsigned int),
exitcode (int)
if type == 'pager', repetitively waits for a command name ending with '\n'
and executes it defined by cmdtable, or exits the loop if the command name
is empty.
"""
def __init__(self, in_, out, channel):
self.in_ = in_
self.out = out
self.channel = channel
def __call__(self, cmd, environ, cwd=None, type='system', cmdtable=None):
args = [type, util.quotecommand(cmd), os.path.abspath(cwd or '.')]
args.extend('%s=%s' % (k, v) for k, v in environ.iteritems())
data = '\0'.join(args)
self.out.write(struct.pack('>cI', self.channel, len(data)))
self.out.write(data)
self.out.flush()
if type == 'system':
length = self.in_.read(4)
length, = struct.unpack('>I', length)
if length != 4:
raise error.Abort(_('invalid response'))
rc, = struct.unpack('>i', self.in_.read(4))
return rc
elif type == 'pager':
while True:
cmd = self.in_.readline()[:-1]
if not cmd:
break
if cmdtable and cmd in cmdtable:
_log('pager subcommand: %s' % cmd)
cmdtable[cmd]()
else:
raise error.Abort(_('unexpected command: %s') % cmd)
else:
raise error.ProgrammingError('invalid S channel type: %s' % type)
_iochannels = [
# server.ch, ui.fp, mode
('cin', 'fin', 'rb'),
('cout', 'fout', 'wb'),
('cerr', 'ferr', 'wb'),
]
class chgcmdserver(commandserver.server):
def __init__(self, ui, repo, fin, fout, sock, hashstate, baseaddress):
super(chgcmdserver, self).__init__(
_newchgui(ui, channeledsystem(fin, fout, 'S'), self.attachio),
repo, fin, fout)
self.clientsock = sock
self._oldios = [] # original (self.ch, ui.fp, fd) before "attachio"
self.hashstate = hashstate
self.baseaddress = baseaddress
if hashstate is not None:
self.capabilities = self.capabilities.copy()
self.capabilities['validate'] = chgcmdserver.validate
def cleanup(self):
super(chgcmdserver, self).cleanup()
# dispatch._runcatch() does not flush outputs if exception is not
# handled by dispatch._dispatch()
self.ui.flush()
self._restoreio()
def attachio(self):
"""Attach to client's stdio passed via unix domain socket; all
channels except cresult will no longer be used
"""
# tell client to sendmsg() with 1-byte payload, which makes it
# distinctive from "attachio\n" command consumed by client.read()
self.clientsock.sendall(struct.pack('>cI', 'I', 1))
clientfds = osutil.recvfds(self.clientsock.fileno())
_log('received fds: %r\n' % clientfds)
ui = self.ui
ui.flush()
first = self._saveio()
for fd, (cn, fn, mode) in zip(clientfds, _iochannels):
assert fd > 0
fp = getattr(ui, fn)
os.dup2(fd, fp.fileno())
os.close(fd)
if not first:
continue
# reset buffering mode when client is first attached. as we want
# to see output immediately on pager, the mode stays unchanged
# when client re-attached. ferr is unchanged because it should
# be unbuffered no matter if it is a tty or not.
if fn == 'ferr':
newfp = fp
else:
# make it line buffered explicitly because the default is
# decided on first write(), where fout could be a pager.
if fp.isatty():
bufsize = 1 # line buffered
else:
bufsize = -1 # system default
newfp = os.fdopen(fp.fileno(), mode, bufsize)
setattr(ui, fn, newfp)
setattr(self, cn, newfp)
self.cresult.write(struct.pack('>i', len(clientfds)))
def _saveio(self):
if self._oldios:
return False
ui = self.ui
for cn, fn, _mode in _iochannels:
ch = getattr(self, cn)
fp = getattr(ui, fn)
fd = os.dup(fp.fileno())
self._oldios.append((ch, fp, fd))
return True
def _restoreio(self):
ui = self.ui
for (ch, fp, fd), (cn, fn, _mode) in zip(self._oldios, _iochannels):
newfp = getattr(ui, fn)
# close newfp while it's associated with client; otherwise it
# would be closed when newfp is deleted
if newfp is not fp:
newfp.close()
# restore original fd: fp is open again
os.dup2(fd, fp.fileno())
os.close(fd)
setattr(self, cn, ch)
setattr(ui, fn, fp)
del self._oldios[:]
def validate(self):
"""Reload the config and check if the server is up to date
Read a list of '\0' separated arguments.
Write a non-empty list of '\0' separated instruction strings or '\0'
if the list is empty.
An instruction string could be either:
- "unlink $path", the client should unlink the path to stop the
outdated server.
- "redirect $path", the client should attempt to connect to $path
first. If it does not work, start a new server. It implies
"reconnect".
- "exit $n", the client should exit directly with code n.
This may happen if we cannot parse the config.
- "reconnect", the client should close the connection and
reconnect.
If neither "reconnect" nor "redirect" is included in the instruction
list, the client can continue with this server after completing all
the instructions.
"""
from . import dispatch # avoid cycle
args = self._readlist()
try:
self.ui, lui = _loadnewui(self.ui, args)
except error.ParseError as inst:
dispatch._formatparse(self.ui.warn, inst)
self.ui.flush()
self.cresult.write('exit 255')
return
newhash = hashstate.fromui(lui, self.hashstate.mtimepaths)
insts = []
if newhash.mtimehash != self.hashstate.mtimehash:
addr = _hashaddress(self.baseaddress, self.hashstate.confighash)
insts.append('unlink %s' % addr)
# mtimehash is empty if one or more extensions fail to load.
# to be compatible with hg, still serve the client this time.
if self.hashstate.mtimehash:
insts.append('reconnect')
if newhash.confighash != self.hashstate.confighash:
addr = _hashaddress(self.baseaddress, newhash.confighash)
insts.append('redirect %s' % addr)
_log('validate: %s\n' % insts)
self.cresult.write('\0'.join(insts) or '\0')
def chdir(self):
"""Change current directory
Note that the behavior of --cwd option is bit different from this.
It does not affect --config parameter.
"""
path = self._readstr()
if not path:
return
_log('chdir to %r\n' % path)
os.chdir(path)
def setumask(self):
"""Change umask"""
mask = struct.unpack('>I', self._read(4))[0]
_log('setumask %r\n' % mask)
os.umask(mask)
def runcommand(self):
return super(chgcmdserver, self).runcommand()
def setenv(self):
"""Clear and update os.environ
Note that not all variables can make an effect on the running process.
"""
l = self._readlist()
try:
newenv = dict(s.split('=', 1) for s in l)
except ValueError:
raise ValueError('unexpected value in setenv request')
_log('setenv: %r\n' % sorted(newenv.keys()))
encoding.environ.clear()
encoding.environ.update(newenv)
capabilities = commandserver.server.capabilities.copy()
capabilities.update({'attachio': attachio,
'chdir': chdir,
'runcommand': runcommand,
'setenv': setenv,
'setumask': setumask})
if util.safehasattr(osutil, 'setprocname'):
def setprocname(self):
"""Change process title"""
name = self._readstr()
_log('setprocname: %r\n' % name)
osutil.setprocname(name)
capabilities['setprocname'] = setprocname
def _tempaddress(address):
return '%s.%d.tmp' % (address, os.getpid())
def _hashaddress(address, hashstr):
# if the basename of address contains '.', use only the left part. this
# makes it possible for the client to pass 'server.tmp$PID' and follow by
# an atomic rename to avoid locking when spawning new servers.
dirname, basename = os.path.split(address)
basename = basename.split('.', 1)[0]
return '%s-%s' % (os.path.join(dirname, basename), hashstr)
class chgunixservicehandler(object):
"""Set of operations for chg services"""
pollinterval = 1 # [sec]
def __init__(self, ui):
self.ui = ui
self._idletimeout = ui.configint('chgserver', 'idletimeout', 3600)
self._lastactive = time.time()
def bindsocket(self, sock, address):
self._inithashstate(address)
self._checkextensions()
self._bind(sock)
self._createsymlink()
def _inithashstate(self, address):
self._baseaddress = address
if self.ui.configbool('chgserver', 'skiphash', False):
self._hashstate = None
self._realaddress = address
return
self._hashstate = hashstate.fromui(self.ui)
self._realaddress = _hashaddress(address, self._hashstate.confighash)
def _checkextensions(self):
if not self._hashstate:
return
if extensions.notloaded():
# one or more extensions failed to load. mtimehash becomes
# meaningless because we do not know the paths of those extensions.
# set mtimehash to an illegal hash value to invalidate the server.
self._hashstate.mtimehash = ''
def _bind(self, sock):
# use a unique temp address so we can stat the file and do ownership
# check later
tempaddress = _tempaddress(self._realaddress)
util.bindunixsocket(sock, tempaddress)
self._socketstat = os.stat(tempaddress)
# rename will replace the old socket file if exists atomically. the
# old server will detect ownership change and exit.
util.rename(tempaddress, self._realaddress)
def _createsymlink(self):
if self._baseaddress == self._realaddress:
return
tempaddress = _tempaddress(self._baseaddress)
os.symlink(os.path.basename(self._realaddress), tempaddress)
util.rename(tempaddress, self._baseaddress)
def _issocketowner(self):
try:
stat = os.stat(self._realaddress)
return (stat.st_ino == self._socketstat.st_ino and
stat.st_mtime == self._socketstat.st_mtime)
except OSError:
return False
def unlinksocket(self, address):
if not self._issocketowner():
return
# it is possible to have a race condition here that we may
# remove another server's socket file. but that's okay
# since that server will detect and exit automatically and
# the client will start a new server on demand.
try:
os.unlink(self._realaddress)
except OSError as exc:
if exc.errno != errno.ENOENT:
raise
def printbanner(self, address):
# no "listening at" message should be printed to simulate hg behavior
pass
def shouldexit(self):
if not self._issocketowner():
self.ui.debug('%s is not owned, exiting.\n' % self._realaddress)
return True
if time.time() - self._lastactive > self._idletimeout:
self.ui.debug('being idle too long. exiting.\n')
return True
return False
def newconnection(self):
self._lastactive = time.time()
def createcmdserver(self, repo, conn, fin, fout):
return chgcmdserver(self.ui, repo, fin, fout, conn,
self._hashstate, self._baseaddress)
def chgunixservice(ui, repo, opts):
# CHGINTERNALMARK is temporarily set by chg client to detect if chg will
# start another chg. drop it to avoid possible side effects.
if 'CHGINTERNALMARK' in encoding.environ:
del encoding.environ['CHGINTERNALMARK']
if repo:
# one chgserver can serve multiple repos. drop repo information
ui.setconfig('bundle', 'mainreporoot', '', 'repo')
h = chgunixservicehandler(ui)
return commandserver.unixforkingservice(ui, repo=None, opts=opts, handler=h)