Mercurial > public > mercurial-scm > hg-stable
view mercurial/lock.py @ 37060:2ec1fb9de638
wireproto: add request IDs to frames
One of my primary goals with the new wire protocol is to make
operations faster and enable both client and server-side
operations to scale to multiple CPU cores.
One of the ways we make server interactions faster is by reducing
the number of round trips to that server.
With the existing wire protocol, the "batch" command facilitates
executing multiple commands from a single request payload. The way
it works is the requests for multiple commands are serialized. The
server executes those commands sequentially then serializes all
their results. As an optimization for reducing round trips, this
is very effective. The technical implementation, however, is pretty
bad and suffers from a number of deficiencies. For example, it
creates a new place where authorization to run a command must be
checked. (The lack of this checking in older Mercurial releases
was CVE-2018-1000132.)
The principles behind the "batch" command are sound. However, the
execution is not. Therefore, I want to ditch "batch" in the
new wire protocol and have protocol level support for issuing
multiple requests in a single round trip.
This commit introduces support in the frame-based wire protocol to
facilitate this. We do this by adding a "request ID" to each frame.
If a server sees frames associated with different "request IDs," it
handles them as separate requests. All of this happening possibly
as part of the same message from client to server (the same request
body in the case of HTTP).
We /could/ model the exchange the way pipelined HTTP requests do,
where the server processes requests in order they are issued and
received. But this artifically constrains scalability. A better
model is to allow multi-requests to be executed concurrently and
for responses to be sent and handled concurrently. So the
specification explicitly allows this. There is some work to be done
around specifying dependencies between multi-requests. We take
the easy road for now and punt on this problem, declaring that
if order is important, clients must not issue the request until
responses to dependent requests have been received.
This commit focuses on the boilerplate of implementing the request
ID. The server reactor still can't manage multiple, in-flight
request IDs. This will be addressed in a subsequent commit.
Because the wire semantics have changed, we bump the version of the
media type.
Differential Revision: https://phab.mercurial-scm.org/D2869
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Wed, 14 Mar 2018 16:51:34 -0700 |
parents | d77c3b023393 |
children | a8a902d7176e |
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 os import signal import socket import time import warnings from .i18n import _ from . import ( encoding, error, pycompat, util, ) def _getlockprefix(): """Return a string which is used to differentiate pid namespaces It's useful to detect "dead" processes and remove stale locks with confidence. Typically it's just hostname. On modern linux, we include an extra Linux-specific pid namespace identifier. """ result = encoding.strtolocal(socket.gethostname()) if pycompat.sysplatform.startswith('linux'): try: result += '/%x' % os.stat('/proc/self/ns/pid').st_ino except OSError as ex: if ex.errno not in (errno.ENOENT, errno.EACCES, errno.ENOTDIR): raise return result @contextlib.contextmanager def _delayedinterrupt(): """Block signal interrupt while doing something critical This makes sure that the code block wrapped by this context manager won't be interrupted. For Windows developers: It appears not possible to guard time.sleep() from CTRL_C_EVENT, so please don't use time.sleep() to test if this is working. """ assertedsigs = [] blocked = False orighandlers = {} def raiseinterrupt(num): if (num == getattr(signal, 'SIGINT', None) or num == getattr(signal, 'CTRL_C_EVENT', None)): raise KeyboardInterrupt else: raise error.SignalInterrupt def catchterm(num, frame): if blocked: assertedsigs.append(num) else: raiseinterrupt(num) try: # save handlers first so they can be restored even if a setup is # interrupted between signal.signal() and orighandlers[] =. for name in ['CTRL_C_EVENT', 'SIGINT', 'SIGBREAK', 'SIGHUP', 'SIGTERM']: num = getattr(signal, name, None) if num and num not in orighandlers: orighandlers[num] = signal.getsignal(num) try: for num in orighandlers: signal.signal(num, catchterm) except ValueError: pass # in a thread? no luck blocked = True yield finally: # no simple way to reliably restore all signal handlers because # any loops, recursive function calls, except blocks, etc. can be # interrupted. so instead, make catchterm() raise interrupt. blocked = False try: for num, handler in orighandlers.items(): signal.signal(num, handler) except ValueError: pass # in a thread? # re-raise interrupt exception if any, which may be shadowed by a new # interrupt occurred while re-raising the first one if assertedsigs: raiseinterrupt(assertedsigs[0]) def trylock(ui, vfs, lockname, timeout, warntimeout, *args, **kwargs): """return an acquired lock or raise an a LockHeld exception This function is responsible to issue warnings and or debug messages about the held lock while trying to acquires it.""" def printwarning(printer, locker): """issue the usual "waiting on lock" message through any channel""" # show more details for new-style locks if ':' in locker: host, pid = locker.split(":", 1) msg = (_("waiting for lock on %s held by process %r on host %r\n") % (pycompat.bytestr(l.desc), pycompat.bytestr(pid), pycompat.bytestr(host))) else: msg = (_("waiting for lock on %s held by %r\n") % (l.desc, pycompat.bytestr(locker))) printer(msg) l = lock(vfs, lockname, 0, *args, dolock=False, **kwargs) debugidx = 0 if (warntimeout and timeout) else -1 warningidx = 0 if not timeout: warningidx = -1 elif warntimeout: warningidx = warntimeout delay = 0 while True: try: l._trylock() break except error.LockHeld as inst: if delay == debugidx: printwarning(ui.debug, inst.locker) if delay == warningidx: printwarning(ui.warn, inst.locker) if timeout <= delay: raise error.LockHeld(errno.ETIMEDOUT, inst.filename, l.desc, inst.locker) time.sleep(1) delay += 1 l.delay = delay if l.delay: if 0 <= warningidx <= l.delay: ui.warn(_("got lock after %d seconds\n") % l.delay) else: ui.debug("got lock after %d seconds\n" % l.delay) if l.acquirefn: l.acquirefn() return l 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, dolock=True): 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() if dolock: 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 = _getlockprefix() lockname = '%s:%d' % (lock._host, self.pid) retry = 5 while not self.held and retry: retry -= 1 try: with _delayedinterrupt(): self.vfs.makelock(lockname, self.f) self.held = 1 except (OSError, IOError) as why: if why.errno == errno.EEXIST: locker = self._readlock() if locker is None: continue # 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) if not self.held: # use empty locker to mean "busy for frequent lock/unlock # by many processes" raise error.LockHeld(errno.EAGAIN, self.vfs.join(self.f), 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()