Mercurial > hg
view mercurial/lock.py @ 29830:92ac2baaea86
revlog: use an LRU cache for delta chain bases
Profiling using statprof revealed a hotspot during changegroup
application calculating delta chain bases on generaldelta repos.
Essentially, revlog._addrevision() was performing a lot of redundant
work tracing the delta chain as part of determining when the chain
distance was acceptable. This was most pronounced when adding
revisions to manifests, which can have delta chains thousands of
revisions long.
There was a delta chain base cache on revlogs before, but it only
captured a single revision. This was acceptable before generaldelta,
when _addrevision would build deltas from the previous revision and
thus we'd pretty much guarantee a cache hit when resolving the delta
chain base on a subsequent _addrevision call. However, it isn't
suitable for generaldelta because parent revisions aren't necessarily
the last processed revision.
This patch converts the delta chain base cache to an LRU dict cache.
The cache can hold multiple entries, so generaldelta repos have a
higher chance of getting a cache hit.
The impact of this change when processing changegroup additions is
significant. On a generaldelta conversion of the "mozilla-unified"
repo (which contains heads of the main Firefox repositories in
chronological order - this means there are lots of transitions between
heads in revlog order), this change has the following impact when
performing an `hg unbundle` of an uncompressed bundle of the repo:
before: 5:42 CPU time
after: 4:34 CPU time
Most of this time is saved when applying the changelog and manifest
revlogs:
before: 2:30 CPU time
after: 1:17 CPU time
That nearly a 50% reduction in CPU time applying changesets and
manifests!
Applying a gzipped bundle of the same repo (effectively simulating a
`hg clone` over HTTP) showed a similar speedup:
before: 5:53 CPU time
after: 4:46 CPU time
Wall time improvements were basically the same as CPU time.
I didn't measure explicitly, but it feels like most of the time
is saved when processing manifests. This makes sense, as large
manifests tend to have very long delta chains and thus benefit the
most from this cache.
So, this change effectively makes changegroup application (which is
used by `hg unbundle`, `hg clone`, `hg pull`, `hg unshelve`, and
various other commands) significantly faster when delta chains are
long (which can happen on repos with large numbers of files and thus
large manifests).
In theory, this change can result in more memory utilization. However,
we're caching a dict of ints. At most we have 200 ints + Python object
overhead per revlog. And, the cache is really only populated when
performing read-heavy operations, such as adding changegroups or
scanning an individual revlog. For memory bloat to be an issue, we'd
need to scan/read several revisions from several revlogs all while
having active references to several revlogs. I don't think there are
many operations that do this, so I don't think memory bloat from the
cache will be an issue.
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Mon, 22 Aug 2016 21:48:50 -0700 |
parents | 518c3e392f75 |
children | dc9f086c7691 |
line wrap: on
line source
# lock.py - simple advisory locking scheme for mercurial # # Copyright 2005, 2006 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 __future__ import absolute_import import contextlib import errno import socket import time import warnings from . import ( error, util, ) class lock(object): '''An advisory lock held by one process to control access to a set of files. Non-cooperating processes or incorrectly written scripts can ignore Mercurial's locking scheme and stomp all over the repository, so don't do that. Typically used via localrepository.lock() to lock the repository store (.hg/store/) or localrepository.wlock() to lock everything else under .hg/.''' # lock is symlink on platforms that support it, file on others. # symlink is used because create of directory entry and contents # are atomic even over nfs. # old-style lock: symlink to pid # new-style lock: symlink to hostname:pid _host = None def __init__(self, vfs, file, timeout=-1, releasefn=None, acquirefn=None, desc=None, inheritchecker=None, parentlock=None): self.vfs = vfs self.f = file self.held = 0 self.timeout = timeout self.releasefn = releasefn self.acquirefn = acquirefn self.desc = desc self._inheritchecker = inheritchecker self.parentlock = parentlock self._parentheld = False self._inherited = False self.postrelease = [] self.pid = self._getpid() self.delay = self.lock() if self.acquirefn: self.acquirefn() def __enter__(self): return self def __exit__(self, exc_type, exc_value, exc_tb): self.release() def __del__(self): if self.held: warnings.warn("use lock.release instead of del lock", category=DeprecationWarning, stacklevel=2) # ensure the lock will be removed # even if recursive locking did occur self.held = 1 self.release() def _getpid(self): # wrapper around util.getpid() to make testing easier return util.getpid() def lock(self): timeout = self.timeout while True: try: self._trylock() return self.timeout - timeout except error.LockHeld as inst: if timeout != 0: time.sleep(1) if timeout > 0: timeout -= 1 continue raise error.LockHeld(errno.ETIMEDOUT, inst.filename, self.desc, inst.locker) def _trylock(self): if self.held: self.held += 1 return if lock._host is None: lock._host = socket.gethostname() lockname = '%s:%s' % (lock._host, self.pid) retry = 5 while not self.held and retry: retry -= 1 try: self.vfs.makelock(lockname, self.f) self.held = 1 except (OSError, IOError) as why: if why.errno == errno.EEXIST: locker = self._readlock() # special case where a parent process holds the lock -- this # is different from the pid being different because we do # want the unlock and postrelease functions to be called, # but the lockfile to not be removed. if locker == self.parentlock: self._parentheld = True self.held = 1 return locker = self._testlock(locker) if locker is not None: raise error.LockHeld(errno.EAGAIN, self.vfs.join(self.f), self.desc, locker) else: raise error.LockUnavailable(why.errno, why.strerror, why.filename, self.desc) def _readlock(self): """read lock and return its value Returns None if no lock exists, pid for old-style locks, and host:pid for new-style locks. """ try: return self.vfs.readlock(self.f) except (OSError, IOError) as why: if why.errno == errno.ENOENT: return None raise def _testlock(self, locker): if locker is None: return None try: host, pid = locker.split(":", 1) except ValueError: return locker if host != lock._host: return locker try: pid = int(pid) except ValueError: return locker if util.testpid(pid): return locker # if locker dead, break lock. must do this with another lock # held, or can race and break valid lock. try: l = lock(self.vfs, self.f + '.break', timeout=0) self.vfs.unlink(self.f) l.release() except error.LockError: return locker def testlock(self): """return id of locker if lock is valid, else None. If old-style lock, we cannot tell what machine locker is on. with new-style lock, if locker is on this machine, we can see if locker is alive. If locker is on this machine but not alive, we can safely break lock. The lock file is only deleted when None is returned. """ locker = self._readlock() return self._testlock(locker) @contextlib.contextmanager def inherit(self): """context for the lock to be inherited by a Mercurial subprocess. Yields a string that will be recognized by the lock in the subprocess. Communicating this string to the subprocess needs to be done separately -- typically by an environment variable. """ if not self.held: raise error.LockInheritanceContractViolation( 'inherit can only be called while lock is held') if self._inherited: raise error.LockInheritanceContractViolation( 'inherit cannot be called while lock is already inherited') if self._inheritchecker is not None: self._inheritchecker() if self.releasefn: self.releasefn() if self._parentheld: lockname = self.parentlock else: lockname = '%s:%s' % (lock._host, self.pid) self._inherited = True try: yield lockname finally: if self.acquirefn: self.acquirefn() self._inherited = False def release(self): """release the lock and execute callback function if any If the lock has been acquired multiple times, the actual release is delayed to the last release call.""" if self.held > 1: self.held -= 1 elif self.held == 1: self.held = 0 if self._getpid() != self.pid: # we forked, and are not the parent return try: if self.releasefn: self.releasefn() finally: if not self._parentheld: try: self.vfs.unlink(self.f) except OSError: pass # The postrelease functions typically assume the lock is not held # at all. if not self._parentheld: for callback in self.postrelease: callback() # Prevent double usage and help clear cycles. self.postrelease = None def release(*locks): for lock in locks: if lock is not None: lock.release()