view hgext/share.py @ 40021:c537144fdbef

wireprotov2: support response caching

One of the things I've learned from managing VCS servers over the
years is that they are hard to scale. It is well known that some
companies have very beefy (read: very expensive) servers to power
their VCS needs. It is also known that specialized servers for
various VCS exist in order to facilitate scaling servers. (Mercurial
is in this boat.)

One of the aspects that make a VCS server hard to scale is the
high CPU load incurred by constant client clone/pull operations.

To alleviate the scaling pain associated with data retrieval
operations, I want to integrate caching into the Mercurial wire
protocol server as robustly as possible such that servers can
aggressively cache responses and defer as much server load as
possible.

This commit represents the initial implementation of a general
caching layer in wire protocol version 2.

We define a new interface and behavior for a wire protocol cacher
in repository.py. (This is probably where a reviewer should look
first to understand what is going on.)

The bulk of the added code is in wireprotov2server.py, where we
define how a command can opt in to being cached and integrate
caching into command dispatching.

From a very high-level:

* A command can declare itself as cacheable by providing a callable
  that can be used to derive a cache key.
* At dispatch time, if a command is cacheable, we attempt to
  construct a cacher and use it for serving the request and/or
  caching the request.
* The dispatch layer handles the bulk of the business logic for
  caching, making cachers mostly "dumb content stores."
* The mechanism for invalidating cached entries (one of the harder
  parts about caching in general) is by varying the cache key when
  state changes. As such, cachers don't need to be concerned with
  cache invalidation.

Initially, we've hooked up support for caching "manifestdata" and
"filedata" commands. These are the simplest to cache, as they should
be immutable over time. Caching of commands related to changeset
data is a bit harder (because cache validation is impacted by
changes to bookmarks, phases, etc). This will be implemented later.
(Strictly speaking, censoring a file should invalidate caches. I've
added an inline TODO to track this edge case.)

To prove it works, this commit implements a test-only extension
providing in-memory caching backed by an lrucachedict. A new test
showing this extension behaving properly is added. FWIW, the
cacher is ~50 lines of code, demonstrating the relative ease with
which a cache can be added to a server.

While the test cacher is not suitable for production workloads, just
for kicks I performed a clone of just the changeset and manifest data
for the mozilla-unified repository. With a fully warmed cache (of just
the manifest data since changeset data is not cached), server-side
CPU usage dropped from ~73s to ~28s. That's pretty significant and
demonstrates the potential that response caching has on server
scalability!

Differential Revision: https://phab.mercurial-scm.org/D4773

# Copyright 2006, 2007 Matt Mackall <[e-mail]>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.

'''share a common history between several working directories

Automatic Pooled Storage for Clones
-----------------------------------

When this extension is active, :hg:`clone` can be configured to
automatically share/pool storage across multiple clones. This
mode effectively converts :hg:`clone` to :hg:`clone` + :hg:`share`.
The benefit of using this mode is the automatic management of
store paths and intelligent pooling of related repositories.

The following ``share.`` config options influence this feature:

``share.pool``
    Filesystem path where shared repository data will be stored. When
    defined, :hg:`clone` will automatically use shared repository
    storage instead of creating a store inside each clone.

``share.poolnaming``
    How directory names in ``share.pool`` are constructed.

    "identity" means the name is derived from the first changeset in the
    repository. In this mode, different remotes share storage if their
    root/initial changeset is identical. In this mode, the local shared
    repository is an aggregate of all encountered remote repositories.

    "remote" means the name is derived from the source repository's
    path or URL. In this mode, storage is only shared if the path or URL
    requested in the :hg:`clone` command matches exactly to a repository
    that was cloned before.

    The default naming mode is "identity".
'''

from __future__ import absolute_import

import errno
from mercurial.i18n import _
from mercurial import (
    bookmarks,
    commands,
    error,
    extensions,
    hg,
    registrar,
    txnutil,
    util,
)

cmdtable = {}
command = registrar.command(cmdtable)
# Note for extension authors: ONLY specify testedwith = 'ships-with-hg-core' for
# extensions which SHIP WITH MERCURIAL. Non-mainline extensions should
# be specifying the version(s) of Mercurial they are tested with, or
# leave the attribute unspecified.
testedwith = 'ships-with-hg-core'

@command('share',
    [('U', 'noupdate', None, _('do not create a working directory')),
     ('B', 'bookmarks', None, _('also share bookmarks')),
     ('', 'relative', None, _('point to source using a relative path '
                              '(EXPERIMENTAL)')),
    ],
    _('[-U] [-B] SOURCE [DEST]'),
    norepo=True)
def share(ui, source, dest=None, noupdate=False, bookmarks=False,
          relative=False):
    """create a new shared repository

    Initialize a new repository and working directory that shares its
    history (and optionally bookmarks) with another repository.

    .. note::

       using rollback or extensions that destroy/modify history (mq,
       rebase, etc.) can cause considerable confusion with shared
       clones. In particular, if two shared clones are both updated to
       the same changeset, and one of them destroys that changeset
       with rollback, the other clone will suddenly stop working: all
       operations will fail with "abort: working directory has unknown
       parent". The only known workaround is to use debugsetparents on
       the broken clone to reset it to a changeset that still exists.
    """

    hg.share(ui, source, dest=dest, update=not noupdate,
             bookmarks=bookmarks, relative=relative)
    return 0

@command('unshare', [], '')
def unshare(ui, repo):
    """convert a shared repository to a normal one

    Copy the store data to the repo and remove the sharedpath data.
    """

    if not repo.shared():
        raise error.Abort(_("this is not a shared repo"))

    hg.unshare(ui, repo)

# Wrap clone command to pass auto share options.
def clone(orig, ui, source, *args, **opts):
    pool = ui.config('share', 'pool')
    if pool:
        pool = util.expandpath(pool)

    opts[r'shareopts'] = {
        'pool': pool,
        'mode': ui.config('share', 'poolnaming'),
    }

    return orig(ui, source, *args, **opts)

def extsetup(ui):
    extensions.wrapfunction(bookmarks, '_getbkfile', getbkfile)
    extensions.wrapfunction(bookmarks.bmstore, '_recordchange', recordchange)
    extensions.wrapfunction(bookmarks.bmstore, '_writerepo', writerepo)
    extensions.wrapcommand(commands.table, 'clone', clone)

def _hassharedbookmarks(repo):
    """Returns whether this repo has shared bookmarks"""
    try:
        shared = repo.vfs.read('shared').splitlines()
    except IOError as inst:
        if inst.errno != errno.ENOENT:
            raise
        return False
    return hg.sharedbookmarks in shared

def getbkfile(orig, repo):
    if _hassharedbookmarks(repo):
        srcrepo = hg.sharedreposource(repo)
        if srcrepo is not None:
            # just orig(srcrepo) doesn't work as expected, because
            # HG_PENDING refers repo.root.
            try:
                fp, pending = txnutil.trypending(repo.root, repo.vfs,
                                                 'bookmarks')
                if pending:
                    # only in this case, bookmark information in repo
                    # is up-to-date.
                    return fp
                fp.close()
            except IOError as inst:
                if inst.errno != errno.ENOENT:
                    raise

            # otherwise, we should read bookmarks from srcrepo,
            # because .hg/bookmarks in srcrepo might be already
            # changed via another sharing repo
            repo = srcrepo

            # TODO: Pending changes in repo are still invisible in
            # srcrepo, because bookmarks.pending is written only into repo.
            # See also https://www.mercurial-scm.org/wiki/SharedRepository
    return orig(repo)

def recordchange(orig, self, tr):
    # Continue with write to local bookmarks file as usual
    orig(self, tr)

    if _hassharedbookmarks(self._repo):
        srcrepo = hg.sharedreposource(self._repo)
        if srcrepo is not None:
            category = 'share-bookmarks'
            tr.addpostclose(category, lambda tr: self._writerepo(srcrepo))

def writerepo(orig, self, repo):
    # First write local bookmarks file in case we ever unshare
    orig(self, repo)

    if _hassharedbookmarks(self._repo):
        srcrepo = hg.sharedreposource(self._repo)
        if srcrepo is not None:
            orig(self, srcrepo)