Mercurial > public > mercurial-scm > hg
view mercurial/repocache.py @ 44118:f81c17ec303c
hgdemandimport: apply lazy module loading to sys.meta_path finders
Python's `sys.meta_path` finders are the primary objects whose job it
is to find a module at import time. When `import` is called, Python
iterates objects in this list and calls `o.find_spec(...)` to find
a `ModuleSpec` (or None if the module couldn't be found by that
finder). If no meta path finder can find a module, import fails.
One of the default meta path finders is `PathFinder`. Its job is to
import modules from the filesystem and is probably the most important
importer. This finder looks at `sys.path` and `sys.path_hooks` to do
its job.
The `ModuleSpec` returned by `MetaPathImporter.find_spec()` has a
`loader` attribute, which defines the concrete module loader to use.
`sys.path_hooks` is a hook point for teaching `PathFinder` to
instantiate custom loader types.
Previously, we injected a custom `sys.path_hook` that told `PathFinder`
to wrap the default loaders with a loader that creates a module object
that is lazy.
This approach worked. But its main limitation was that it only applied
to the `PathFinder` meta path importer. There are other meta path
importers that are registered. And in the case of PyOxidizer loading
modules from memory, `PathFinder` doesn't come into play since
PyOxidizer's own meta path importer was handling all imports.
This commit changes our approach to lazy module loading by proxying
all meta path importers. Specifically, we overload the `find_spec()`
method to swap in a wrapped loader on the `ModuleSpec` before it
is returned. The end result of this is all meta path importers should
be lazy.
As much as I would have loved to utilize .__class__ manipulation to
achieve this, some meta path importers are implemented in C/Rust
in such a way that they cannot be monkeypatched. This is why we
use __getattribute__ to define a proxy.
Also, this change could theoretically open us up to regressions in
meta path importers whose loader is creating module objects which
can't be monkeypatched. But I'm not aware of any of these in the
wild. So I think we'll be safe.
According to hyperfine, this change yields a decent startup time win of
5-6ms:
```
Benchmark #1: ~/.pyenv/versions/3.6.10/bin/python ./hg version
Time (mean ? ?): 86.8 ms ? 0.5 ms [User: 78.0 ms, System: 8.7 ms]
Range (min ? max): 86.0 ms ? 89.1 ms 50 runs
Time (mean ? ?): 81.1 ms ? 2.7 ms [User: 74.5 ms, System: 6.5 ms]
Range (min ? max): 77.8 ms ? 90.5 ms 50 runs
Benchmark #2: ~/.pyenv/versions/3.7.6/bin/python ./hg version
Time (mean ? ?): 78.9 ms ? 0.6 ms [User: 70.2 ms, System: 8.7 ms]
Range (min ? max): 78.1 ms ? 81.2 ms 50 runs
Time (mean ? ?): 73.4 ms ? 0.6 ms [User: 65.3 ms, System: 8.0 ms]
Range (min ? max): 72.4 ms ? 75.7 ms 50 runs
Benchmark #3: ~/.pyenv/versions/3.8.1/bin/python ./hg version
Time (mean ? ?): 78.1 ms ? 0.6 ms [User: 70.2 ms, System: 7.9 ms]
Range (min ? max): 77.4 ms ? 80.9 ms 50 runs
Time (mean ? ?): 72.1 ms ? 0.4 ms [User: 64.4 ms, System: 7.6 ms]
Range (min ? max): 71.4 ms ? 74.1 ms 50 runs
```
Differential Revision: https://phab.mercurial-scm.org/D7954
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Mon, 20 Jan 2020 23:51:25 -0800 |
parents | 8ff1ecfadcd1 |
children | 6000f5b25c9b |
line wrap: on
line source
# repocache.py - in-memory repository cache for long-running services # # Copyright 2018 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. from __future__ import absolute_import import collections import gc import threading from . import ( error, hg, obsolete, scmutil, util, ) class repoloader(object): """Load repositories in background thread This is designed for a forking server. A cached repo cannot be obtained until the server fork()s a worker and the loader thread stops. """ def __init__(self, ui, maxlen): self._ui = ui.copy() self._cache = util.lrucachedict(max=maxlen) # use deque and Event instead of Queue since deque can discard # old items to keep at most maxlen items. self._inqueue = collections.deque(maxlen=maxlen) self._accepting = False self._newentry = threading.Event() self._thread = None def start(self): assert not self._thread if self._inqueue.maxlen == 0: # no need to spawn loader thread as the cache is disabled return self._accepting = True self._thread = threading.Thread(target=self._mainloop) self._thread.start() def stop(self): if not self._thread: return self._accepting = False self._newentry.set() self._thread.join() self._thread = None self._cache.clear() self._inqueue.clear() def load(self, path): """Request to load the specified repository in background""" self._inqueue.append(path) self._newentry.set() def get(self, path): """Return a cached repo if available This function must be called after fork(), where the loader thread is stopped. Otherwise, the returned repo might be updated by the loader thread. """ if self._thread and self._thread.is_alive(): raise error.ProgrammingError( b'cannot obtain cached repo while loader is active' ) return self._cache.peek(path, None) def _mainloop(self): while self._accepting: # Avoid heavy GC after fork(), which would cancel the benefit of # COW. We assume that GIL is acquired while GC is underway in the # loader thread. If that isn't true, we might have to move # gc.collect() to the main thread so that fork() would never stop # the thread where GC is in progress. gc.collect() self._newentry.wait() while self._accepting: self._newentry.clear() try: path = self._inqueue.popleft() except IndexError: break scmutil.callcatch(self._ui, lambda: self._load(path)) def _load(self, path): start = util.timer() # TODO: repo should be recreated if storage configuration changed try: # pop before loading so inconsistent state wouldn't be exposed repo = self._cache.pop(path) except KeyError: repo = hg.repository(self._ui, path).unfiltered() _warmupcache(repo) repo.ui.log( b'repocache', b'loaded repo into cache: %s (in %.3fs)\n', path, util.timer() - start, ) self._cache.insert(path, repo) # TODO: think about proper API of preloading cache def _warmupcache(repo): repo.invalidateall() repo.changelog repo.obsstore._all repo.obsstore.successors repo.obsstore.predecessors repo.obsstore.children for name in obsolete.cachefuncs: obsolete.getrevs(repo, name) repo._phasecache.loadphaserevs(repo) # TODO: think about proper API of attaching preloaded attributes def copycache(srcrepo, destrepo): """Copy cached attributes from srcrepo to destrepo""" destfilecache = destrepo._filecache srcfilecache = srcrepo._filecache if b'changelog' in srcfilecache: destfilecache[b'changelog'] = ce = srcfilecache[b'changelog'] ce.obj.opener = ce.obj._realopener = destrepo.svfs if b'obsstore' in srcfilecache: destfilecache[b'obsstore'] = ce = srcfilecache[b'obsstore'] ce.obj.svfs = destrepo.svfs if b'_phasecache' in srcfilecache: destfilecache[b'_phasecache'] = ce = srcfilecache[b'_phasecache'] ce.obj.opener = destrepo.svfs