view hgext/remotefilelog/shallowrepo.py @ 42354:dbd0fcca6dfc

discovery: slowly increase sampling size

Some pathological discovery runs can requires many roundtrip. When this happens
things can get very slow.

To make the algorithm more resilience again such pathological case. We slowly
increase the sample size with each roundtrip (+5%). This will have a negligible
impact on "normal" discovery with few roundtrips, but a large positive impact of
case with many roundtrips. Asking more question per roundtrip helps to reduce
the undecided set faster. Instead of reducing the undecided set a linear speed
(in the worst case), we reduce it as a guaranteed (small) exponential rate. The
data below show this slow ramp up in sample size:

round trip    |    1 |     5 |    10 |    20 |     50 |     100 |       130 |
sample size   |  200 |   254 |   321 |   517 |  2 199 |  25 123 |   108 549 |
covered nodes |  200 | 1 357 | 2 821 | 7 031 | 42 658 | 524 530 | 2 276 755 |

To be a bit more concrete, lets take a very pathological case as an example. We
are doing discovery from a copy of Mozilla-try to a more recent version of
mozilla-unified. Mozilla-unified heads are unknown to the mozilla-try repo and
there are over 1 million "missing" changesets. (the discovery is "local" to
avoid network interference)

Without this change, the discovery:
- last 1858 seconds (31 minutes),
- does 1700 round trip,
- asking about 340 000 nodes.

With this change, the discovery:
- last 218 seconds (3 minutes, 38 seconds a -88% improvement),
- does 94 round trip (-94%),
- asking about 344 211 nodes (+1%).

Of course, this is an extreme case (and 3 minutes is still slow). However this
give a good example of how this sample size increase act as a safety net
catching any bad situations.

We could image a steeper increase than 5%. For example 10% would give the
following number:

round trip    |    1 |     5 |    10 |     20 |     50  |        75 |        100 |
sample size   |  200 |   321 |   514 |  1 326 |  23 060 |   249 812 |  2 706 594 |
covered nodes |  200 | 1 541 | 3 690 | 12 671 | 251 871 | 2 746 254 | 29 770 966 |

In parallel, it is useful to understand these pathological cases and improve
them. However the current change provides a general purpose safety net to smooth
the impact of pathological cases.

To avoid issue with older http server, the increase in sample size only occurs
if the protocol has not limit on command argument size.

# shallowrepo.py - shallow repository that uses remote filelogs
#
# Copyright 2013 Facebook, Inc.
#
# 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 os

from mercurial.i18n import _
from mercurial.node import hex, nullid, nullrev
from mercurial import (
    encoding,
    error,
    localrepo,
    match,
    scmutil,
    sparse,
    util,
)
from mercurial.utils import procutil
from . import (
    connectionpool,
    constants,
    contentstore,
    datapack,
    fileserverclient,
    historypack,
    metadatastore,
    remotefilectx,
    remotefilelog,
    shallowutil,
)

if util.safehasattr(util, '_hgexecutable'):
    # Before 5be286db
    _hgexecutable = util.hgexecutable
else:
    from mercurial.utils import procutil
    _hgexecutable = procutil.hgexecutable

# These make*stores functions are global so that other extensions can replace
# them.
def makelocalstores(repo):
    """In-repo stores, like .hg/store/data; can not be discarded."""
    localpath = os.path.join(repo.svfs.vfs.base, 'data')
    if not os.path.exists(localpath):
        os.makedirs(localpath)

    # Instantiate local data stores
    localcontent = contentstore.remotefilelogcontentstore(
        repo, localpath, repo.name, shared=False)
    localmetadata = metadatastore.remotefilelogmetadatastore(
        repo, localpath, repo.name, shared=False)
    return localcontent, localmetadata

def makecachestores(repo):
    """Typically machine-wide, cache of remote data; can be discarded."""
    # Instantiate shared cache stores
    cachepath = shallowutil.getcachepath(repo.ui)
    cachecontent = contentstore.remotefilelogcontentstore(
        repo, cachepath, repo.name, shared=True)
    cachemetadata = metadatastore.remotefilelogmetadatastore(
        repo, cachepath, repo.name, shared=True)

    repo.sharedstore = cachecontent
    repo.shareddatastores.append(cachecontent)
    repo.sharedhistorystores.append(cachemetadata)

    return cachecontent, cachemetadata

def makeremotestores(repo, cachecontent, cachemetadata):
    """These stores fetch data from a remote server."""
    # Instantiate remote stores
    repo.fileservice = fileserverclient.fileserverclient(repo)
    remotecontent = contentstore.remotecontentstore(
        repo.ui, repo.fileservice, cachecontent)
    remotemetadata = metadatastore.remotemetadatastore(
        repo.ui, repo.fileservice, cachemetadata)
    return remotecontent, remotemetadata

def makepackstores(repo):
    """Packs are more efficient (to read from) cache stores."""
    # Instantiate pack stores
    packpath = shallowutil.getcachepackpath(repo,
                                            constants.FILEPACK_CATEGORY)
    packcontentstore = datapack.datapackstore(repo.ui, packpath)
    packmetadatastore = historypack.historypackstore(repo.ui, packpath)

    repo.shareddatastores.append(packcontentstore)
    repo.sharedhistorystores.append(packmetadatastore)
    shallowutil.reportpackmetrics(repo.ui, 'filestore', packcontentstore,
        packmetadatastore)
    return packcontentstore, packmetadatastore

def makeunionstores(repo):
    """Union stores iterate the other stores and return the first result."""
    repo.shareddatastores = []
    repo.sharedhistorystores = []

    packcontentstore, packmetadatastore = makepackstores(repo)
    cachecontent, cachemetadata = makecachestores(repo)
    localcontent, localmetadata = makelocalstores(repo)
    remotecontent, remotemetadata = makeremotestores(repo, cachecontent,
                                                     cachemetadata)

    # Instantiate union stores
    repo.contentstore = contentstore.unioncontentstore(
        packcontentstore, cachecontent,
        localcontent, remotecontent, writestore=localcontent)
    repo.metadatastore = metadatastore.unionmetadatastore(
        packmetadatastore, cachemetadata, localmetadata, remotemetadata,
        writestore=localmetadata)

    fileservicedatawrite = cachecontent
    fileservicehistorywrite = cachemetadata
    repo.fileservice.setstore(repo.contentstore, repo.metadatastore,
                              fileservicedatawrite, fileservicehistorywrite)
    shallowutil.reportpackmetrics(repo.ui, 'filestore',
        packcontentstore, packmetadatastore)

def wraprepo(repo):
    class shallowrepository(repo.__class__):
        @util.propertycache
        def name(self):
            return self.ui.config('remotefilelog', 'reponame')

        @util.propertycache
        def fallbackpath(self):
            path = repo.ui.config("remotefilelog", "fallbackpath",
                                  repo.ui.config('paths', 'default'))
            if not path:
                raise error.Abort("no remotefilelog server "
                    "configured - is your .hg/hgrc trusted?")

            return path

        def maybesparsematch(self, *revs, **kwargs):
            '''
            A wrapper that allows the remotefilelog to invoke sparsematch() if
            this is a sparse repository, or returns None if this is not a
            sparse repository.
            '''
            if revs:
                ret = sparse.matcher(repo, revs=revs)
            else:
                ret = sparse.matcher(repo)

            if ret.always():
                return None
            return ret

        def file(self, f):
            if f[0] == '/':
                f = f[1:]

            if self.shallowmatch(f):
                return remotefilelog.remotefilelog(self.svfs, f, self)
            else:
                return super(shallowrepository, self).file(f)

        def filectx(self, path, *args, **kwargs):
            if self.shallowmatch(path):
                return remotefilectx.remotefilectx(self, path, *args, **kwargs)
            else:
                return super(shallowrepository, self).filectx(path, *args,
                                                              **kwargs)

        @localrepo.unfilteredmethod
        def commitctx(self, ctx, error=False):
            """Add a new revision to current repository.
            Revision information is passed via the context argument.
            """

            # some contexts already have manifest nodes, they don't need any
            # prefetching (for example if we're just editing a commit message
            # we can reuse manifest
            if not ctx.manifestnode():
                # prefetch files that will likely be compared
                m1 = ctx.p1().manifest()
                files = []
                for f in ctx.modified() + ctx.added():
                    fparent1 = m1.get(f, nullid)
                    if fparent1 != nullid:
                        files.append((f, hex(fparent1)))
                self.fileservice.prefetch(files)
            return super(shallowrepository, self).commitctx(ctx,
                                                            error=error)

        def backgroundprefetch(self, revs, base=None, repack=False, pats=None,
                               opts=None):
            """Runs prefetch in background with optional repack
            """
            cmd = [_hgexecutable(), '-R', repo.origroot, 'prefetch']
            if repack:
                cmd.append('--repack')
            if revs:
                cmd += ['-r', revs]
            procutil.runbgcommand(cmd, encoding.environ)

        def prefetch(self, revs, base=None, pats=None, opts=None):
            """Prefetches all the necessary file revisions for the given revs
            Optionally runs repack in background
            """
            with repo._lock(repo.svfs, 'prefetchlock', True, None, None,
                            _('prefetching in %s') % repo.origroot):
                self._prefetch(revs, base, pats, opts)

        def _prefetch(self, revs, base=None, pats=None, opts=None):
            fallbackpath = self.fallbackpath
            if fallbackpath:
                # If we know a rev is on the server, we should fetch the server
                # version of those files, since our local file versions might
                # become obsolete if the local commits are stripped.
                localrevs = repo.revs('outgoing(%s)', fallbackpath)
                if base is not None and base != nullrev:
                    serverbase = list(repo.revs('first(reverse(::%s) - %ld)',
                                                base, localrevs))
                    if serverbase:
                        base = serverbase[0]
            else:
                localrevs = repo

            mfl = repo.manifestlog
            mfrevlog = mfl.getstorage('')
            if base is not None:
                mfdict = mfl[repo[base].manifestnode()].read()
                skip = set(mfdict.iteritems())
            else:
                skip = set()

            # Copy the skip set to start large and avoid constant resizing,
            # and since it's likely to be very similar to the prefetch set.
            files = skip.copy()
            serverfiles = skip.copy()
            visited = set()
            visited.add(nullrev)
            revcount = len(revs)
            progress = self.ui.makeprogress(_('prefetching'), total=revcount)
            progress.update(0)
            for rev in sorted(revs):
                ctx = repo[rev]
                if pats:
                    m = scmutil.match(ctx, pats, opts)
                sparsematch = repo.maybesparsematch(rev)

                mfnode = ctx.manifestnode()
                mfrev = mfrevlog.rev(mfnode)

                # Decompressing manifests is expensive.
                # When possible, only read the deltas.
                p1, p2 = mfrevlog.parentrevs(mfrev)
                if p1 in visited and p2 in visited:
                    mfdict = mfl[mfnode].readfast()
                else:
                    mfdict = mfl[mfnode].read()

                diff = mfdict.iteritems()
                if pats:
                    diff = (pf for pf in diff if m(pf[0]))
                if sparsematch:
                    diff = (pf for pf in diff if sparsematch(pf[0]))
                if rev not in localrevs:
                    serverfiles.update(diff)
                else:
                    files.update(diff)

                visited.add(mfrev)
                progress.increment()

            files.difference_update(skip)
            serverfiles.difference_update(skip)
            progress.complete()

            # Fetch files known to be on the server
            if serverfiles:
                results = [(path, hex(fnode)) for (path, fnode) in serverfiles]
                repo.fileservice.prefetch(results, force=True)

            # Fetch files that may or may not be on the server
            if files:
                results = [(path, hex(fnode)) for (path, fnode) in files]
                repo.fileservice.prefetch(results)

        def close(self):
            super(shallowrepository, self).close()
            self.connectionpool.close()

    repo.__class__ = shallowrepository

    repo.shallowmatch = match.always()

    makeunionstores(repo)

    repo.includepattern = repo.ui.configlist("remotefilelog", "includepattern",
                                             None)
    repo.excludepattern = repo.ui.configlist("remotefilelog", "excludepattern",
                                             None)
    if not util.safehasattr(repo, 'connectionpool'):
        repo.connectionpool = connectionpool.connectionpool(repo)

    if repo.includepattern or repo.excludepattern:
        repo.shallowmatch = match.match(repo.root, '', None,
            repo.includepattern, repo.excludepattern)