remotefilelog: add a developer option to wait for background processes

In order to block the main command on the subprocess exiting, we ensure the
repo's ui object will call the subprocess.wait() method to ensure the top-level
hg process doesn't exit until all background processes have also done so.

Currently, in the tests, most operation spawning background process as followed
by commands waiting for these operations to complete. However this waiting is
racy. First because it seems like we can start waiting before the background
operation actually start, in which case it is prematurely detected as "done".
Second, because some commands may spawn multiple background operation for the
same operation (eg: rebase can apparently trigger multiple prefetch). The
current approach could be updated to maybe handle the first issue, but the
second one will never be properly handled.

In most case, we do not care that the bg process keep running after the command
end. (Since we explicitly wait for them to end before doing anything else).  So
we add an option to wait on the background process before exiting the command.
We'll put it in use in the next changeset.

Differential Revision: https://phab.mercurial-scm.org/D7585

hgext/remotefilelog/__init__.py

@@ -1,1302 +1,1303 b''
 # __init__.py - remotefilelog extension
 #
 # 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.
 """remotefilelog causes Mercurial to lazilly fetch file contents (EXPERIMENTAL)
 
 This extension is HIGHLY EXPERIMENTAL. There are NO BACKWARDS COMPATIBILITY
 GUARANTEES. This means that repositories created with this extension may
 only be usable with the exact version of this extension/Mercurial that was
 used. The extension attempts to enforce this in order to prevent repository
 corruption.
 
 remotefilelog works by fetching file contents lazily and storing them
 in a cache on the client rather than in revlogs. This allows enormous
 histories to be transferred only partially, making them easier to
 operate on.
 
 Configs:
 
     ``packs.maxchainlen`` specifies the maximum delta chain length in pack files
 
     ``packs.maxpacksize`` specifies the maximum pack file size
 
     ``packs.maxpackfilecount`` specifies the maximum number of packs in the
       shared cache (trees only for now)
 
     ``remotefilelog.backgroundprefetch`` runs prefetch in background when True
 
     ``remotefilelog.bgprefetchrevs`` specifies revisions to fetch on commit and
       update, and on other commands that use them. Different from pullprefetch.
 
     ``remotefilelog.gcrepack`` does garbage collection during repack when True
 
     ``remotefilelog.nodettl`` specifies maximum TTL of a node in seconds before
       it is garbage collected
 
     ``remotefilelog.repackonhggc`` runs repack on hg gc when True
 
     ``remotefilelog.prefetchdays`` specifies the maximum age of a commit in
       days after which it is no longer prefetched.
 
     ``remotefilelog.prefetchdelay`` specifies delay between background
       prefetches in seconds after operations that change the working copy parent
 
     ``remotefilelog.data.gencountlimit`` constraints the minimum number of data
       pack files required to be considered part of a generation. In particular,
       minimum number of packs files > gencountlimit.
 
     ``remotefilelog.data.generations`` list for specifying the lower bound of
       each generation of the data pack files. For example, list ['100MB','1MB']
       or ['1MB', '100MB'] will lead to three generations: [0, 1MB), [
       1MB, 100MB) and [100MB, infinity).
 
     ``remotefilelog.data.maxrepackpacks`` the maximum number of pack files to
       include in an incremental data repack.
 
     ``remotefilelog.data.repackmaxpacksize`` the maximum size of a pack file for
       it to be considered for an incremental data repack.
 
     ``remotefilelog.data.repacksizelimit`` the maximum total size of pack files
       to include in an incremental data repack.
 
     ``remotefilelog.history.gencountlimit`` constraints the minimum number of
       history pack files required to be considered part of a generation. In
       particular, minimum number of packs files > gencountlimit.
 
     ``remotefilelog.history.generations`` list for specifying the lower bound of
       each generation of the history pack files. For example, list [
       '100MB', '1MB'] or ['1MB', '100MB'] will lead to three generations: [
       0, 1MB), [1MB, 100MB) and [100MB, infinity).
 
     ``remotefilelog.history.maxrepackpacks`` the maximum number of pack files to
       include in an incremental history repack.
 
     ``remotefilelog.history.repackmaxpacksize`` the maximum size of a pack file
       for it to be considered for an incremental history repack.
 
     ``remotefilelog.history.repacksizelimit`` the maximum total size of pack
       files to include in an incremental history repack.
 
     ``remotefilelog.backgroundrepack`` automatically consolidate packs in the
       background
 
     ``remotefilelog.cachepath`` path to cache
 
     ``remotefilelog.cachegroup`` if set, make cache directory sgid to this
       group
 
     ``remotefilelog.cacheprocess`` binary to invoke for fetching file data
 
     ``remotefilelog.debug`` turn on remotefilelog-specific debug output
 
     ``remotefilelog.excludepattern`` pattern of files to exclude from pulls
 
     ``remotefilelog.includepattern`` pattern of files to include in pulls
 
     ``remotefilelog.fetchwarning``: message to print when too many
       single-file fetches occur
 
     ``remotefilelog.getfilesstep`` number of files to request in a single RPC
 
     ``remotefilelog.getfilestype`` if set to 'threaded' use threads to fetch
       files, otherwise use optimistic fetching
 
     ``remotefilelog.pullprefetch`` revset for selecting files that should be
       eagerly downloaded rather than lazily
 
     ``remotefilelog.reponame`` name of the repo. If set, used to partition
       data from other repos in a shared store.
 
     ``remotefilelog.server`` if true, enable server-side functionality
 
     ``remotefilelog.servercachepath`` path for caching blobs on the server
 
     ``remotefilelog.serverexpiration`` number of days to keep cached server
       blobs
 
     ``remotefilelog.validatecache`` if set, check cache entries for corruption
       before returning blobs
 
     ``remotefilelog.validatecachelog`` if set, check cache entries for
       corruption before returning metadata
 
 """
 from __future__ import absolute_import
 
 import os
 import time
 import traceback
 
 from mercurial.node import hex
 from mercurial.i18n import _
 from mercurial.pycompat import open
 from mercurial import (
     changegroup,
     changelog,
     cmdutil,
     commands,
     configitems,
     context,
     copies,
     debugcommands as hgdebugcommands,
     dispatch,
     error,
     exchange,
     extensions,
     hg,
     localrepo,
     match,
     merge,
     node as nodemod,
     patch,
     pycompat,
     registrar,
     repair,
     repoview,
     revset,
     scmutil,
     smartset,
     streamclone,
     util,
 )
 from . import (
     constants,
     debugcommands,
     fileserverclient,
     remotefilectx,
     remotefilelog,
     remotefilelogserver,
     repack as repackmod,
     shallowbundle,
     shallowrepo,
     shallowstore,
     shallowutil,
     shallowverifier,
 )
 
 # ensures debug commands are registered
 hgdebugcommands.command
 
 cmdtable = {}
 command = registrar.command(cmdtable)
 
 configtable = {}
 configitem = registrar.configitem(configtable)
 
 configitem(b'remotefilelog', b'debug', default=False)
 
 configitem(b'remotefilelog', b'reponame', default=b'')
 configitem(b'remotefilelog', b'cachepath', default=None)
 configitem(b'remotefilelog', b'cachegroup', default=None)
 configitem(b'remotefilelog', b'cacheprocess', default=None)
 configitem(b'remotefilelog', b'cacheprocess.includepath', default=None)
 configitem(b"remotefilelog", b"cachelimit", default=b"1000 GB")
 
 configitem(
     b'remotefilelog',
     b'fallbackpath',
     default=configitems.dynamicdefault,
     alias=[(b'remotefilelog', b'fallbackrepo')],
 )
 
 configitem(b'remotefilelog', b'validatecachelog', default=None)
 configitem(b'remotefilelog', b'validatecache', default=b'on')
 configitem(b'remotefilelog', b'server', default=None)
 configitem(b'remotefilelog', b'servercachepath', default=None)
 configitem(b"remotefilelog", b"serverexpiration", default=30)
 configitem(b'remotefilelog', b'backgroundrepack', default=False)
 configitem(b'remotefilelog', b'bgprefetchrevs', default=None)
 configitem(b'remotefilelog', b'pullprefetch', default=None)
 configitem(b'remotefilelog', b'backgroundprefetch', default=False)
 configitem(b'remotefilelog', b'prefetchdelay', default=120)
 configitem(b'remotefilelog', b'prefetchdays', default=14)
 
 configitem(b'remotefilelog', b'getfilesstep', default=10000)
 configitem(b'remotefilelog', b'getfilestype', default=b'optimistic')
 configitem(b'remotefilelog', b'batchsize', configitems.dynamicdefault)
 configitem(b'remotefilelog', b'fetchwarning', default=b'')
 
 configitem(b'remotefilelog', b'includepattern', default=None)
 configitem(b'remotefilelog', b'excludepattern', default=None)
 
 configitem(b'remotefilelog', b'gcrepack', default=False)
 configitem(b'remotefilelog', b'repackonhggc', default=False)
 configitem(b'repack', b'chainorphansbysize', default=True, experimental=True)
 
 configitem(b'packs', b'maxpacksize', default=0)
 configitem(b'packs', b'maxchainlen', default=1000)
 
 configitem(b'devel', b'remotefilelog.ensurestart', default=False)
+configitem(b'devel', b'remotefilelog.bg-wait', default=False)
 
 #  default TTL limit is 30 days
 _defaultlimit = 60 * 60 * 24 * 30
 configitem(b'remotefilelog', b'nodettl', default=_defaultlimit)
 
 configitem(b'remotefilelog', b'data.gencountlimit', default=2),
 configitem(
     b'remotefilelog', b'data.generations', default=[b'1GB', b'100MB', b'1MB']
 )
 configitem(b'remotefilelog', b'data.maxrepackpacks', default=50)
 configitem(b'remotefilelog', b'data.repackmaxpacksize', default=b'4GB')
 configitem(b'remotefilelog', b'data.repacksizelimit', default=b'100MB')
 
 configitem(b'remotefilelog', b'history.gencountlimit', default=2),
 configitem(b'remotefilelog', b'history.generations', default=[b'100MB'])
 configitem(b'remotefilelog', b'history.maxrepackpacks', default=50)
 configitem(b'remotefilelog', b'history.repackmaxpacksize', default=b'400MB')
 configitem(b'remotefilelog', b'history.repacksizelimit', default=b'100MB')
 
 # 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 = b'ships-with-hg-core'
 
 repoclass = localrepo.localrepository
 repoclass._basesupported.add(constants.SHALLOWREPO_REQUIREMENT)
 
 isenabled = shallowutil.isenabled
 
 
 def uisetup(ui):
     """Wraps user facing Mercurial commands to swap them out with shallow
     versions.
     """
     hg.wirepeersetupfuncs.append(fileserverclient.peersetup)
 
     entry = extensions.wrapcommand(commands.table, b'clone', cloneshallow)
     entry[1].append(
         (
             b'',
             b'shallow',
             None,
             _(b"create a shallow clone which uses remote file history"),
         )
     )
 
     extensions.wrapcommand(
         commands.table, b'debugindex', debugcommands.debugindex
     )
     extensions.wrapcommand(
         commands.table, b'debugindexdot', debugcommands.debugindexdot
     )
     extensions.wrapcommand(commands.table, b'log', log)
     extensions.wrapcommand(commands.table, b'pull', pull)
 
     # Prevent 'hg manifest --all'
     def _manifest(orig, ui, repo, *args, **opts):
         if isenabled(repo) and opts.get(r'all'):
             raise error.Abort(_(b"--all is not supported in a shallow repo"))
 
         return orig(ui, repo, *args, **opts)
 
     extensions.wrapcommand(commands.table, b"manifest", _manifest)
 
     # Wrap remotefilelog with lfs code
     def _lfsloaded(loaded=False):
         lfsmod = None
         try:
             lfsmod = extensions.find(b'lfs')
         except KeyError:
             pass
         if lfsmod:
             lfsmod.wrapfilelog(remotefilelog.remotefilelog)
             fileserverclient._lfsmod = lfsmod
 
     extensions.afterloaded(b'lfs', _lfsloaded)
 
     # debugdata needs remotefilelog.len to work
     extensions.wrapcommand(commands.table, b'debugdata', debugdatashallow)
 
     changegroup.cgpacker = shallowbundle.shallowcg1packer
 
     extensions.wrapfunction(
         changegroup, b'_addchangegroupfiles', shallowbundle.addchangegroupfiles
     )
     extensions.wrapfunction(
         changegroup, b'makechangegroup', shallowbundle.makechangegroup
     )
     extensions.wrapfunction(localrepo, b'makestore', storewrapper)
     extensions.wrapfunction(exchange, b'pull', exchangepull)
     extensions.wrapfunction(merge, b'applyupdates', applyupdates)
     extensions.wrapfunction(merge, b'_checkunknownfiles', checkunknownfiles)
     extensions.wrapfunction(context.workingctx, b'_checklookup', checklookup)
     extensions.wrapfunction(scmutil, b'_findrenames', findrenames)
     extensions.wrapfunction(
         copies, b'_computeforwardmissing', computeforwardmissing
     )
     extensions.wrapfunction(dispatch, b'runcommand', runcommand)
     extensions.wrapfunction(repair, b'_collectbrokencsets', _collectbrokencsets)
     extensions.wrapfunction(context.changectx, b'filectx', filectx)
     extensions.wrapfunction(context.workingctx, b'filectx', workingfilectx)
     extensions.wrapfunction(patch, b'trydiff', trydiff)
     extensions.wrapfunction(hg, b'verify', _verify)
     scmutil.fileprefetchhooks.add(b'remotefilelog', _fileprefetchhook)
 
     # disappointing hacks below
     extensions.wrapfunction(scmutil, b'getrenamedfn', getrenamedfn)
     extensions.wrapfunction(revset, b'filelog', filelogrevset)
     revset.symbols[b'filelog'] = revset.filelog
     extensions.wrapfunction(cmdutil, b'walkfilerevs', walkfilerevs)
 
 
 def cloneshallow(orig, ui, repo, *args, **opts):
     if opts.get(r'shallow'):
         repos = []
 
         def pull_shallow(orig, self, *args, **kwargs):
             if not isenabled(self):
                 repos.append(self.unfiltered())
                 # set up the client hooks so the post-clone update works
                 setupclient(self.ui, self.unfiltered())
 
                 # setupclient fixed the class on the repo itself
                 # but we also need to fix it on the repoview
                 if isinstance(self, repoview.repoview):
                     self.__class__.__bases__ = (
                         self.__class__.__bases__[0],
                         self.unfiltered().__class__,
                     )
                 self.requirements.add(constants.SHALLOWREPO_REQUIREMENT)
                 self._writerequirements()
 
                 # Since setupclient hadn't been called, exchange.pull was not
                 # wrapped. So we need to manually invoke our version of it.
                 return exchangepull(orig, self, *args, **kwargs)
             else:
                 return orig(self, *args, **kwargs)
 
         extensions.wrapfunction(exchange, b'pull', pull_shallow)
 
         # Wrap the stream logic to add requirements and to pass include/exclude
         # patterns around.
         def setup_streamout(repo, remote):
             # Replace remote.stream_out with a version that sends file
             # patterns.
             def stream_out_shallow(orig):
                 caps = remote.capabilities()
                 if constants.NETWORK_CAP_LEGACY_SSH_GETFILES in caps:
                     opts = {}
                     if repo.includepattern:
                         opts[r'includepattern'] = b'\0'.join(
                             repo.includepattern
                         )
                     if repo.excludepattern:
                         opts[r'excludepattern'] = b'\0'.join(
                             repo.excludepattern
                         )
                     return remote._callstream(b'stream_out_shallow', **opts)
                 else:
                     return orig()
 
             extensions.wrapfunction(remote, b'stream_out', stream_out_shallow)
 
         def stream_wrap(orig, op):
             setup_streamout(op.repo, op.remote)
             return orig(op)
 
         extensions.wrapfunction(
             streamclone, b'maybeperformlegacystreamclone', stream_wrap
         )
 
         def canperformstreamclone(orig, pullop, bundle2=False):
             # remotefilelog is currently incompatible with the
             # bundle2 flavor of streamclones, so force us to use
             # v1 instead.
             if b'v2' in pullop.remotebundle2caps.get(b'stream', []):
                 pullop.remotebundle2caps[b'stream'] = [
                     c for c in pullop.remotebundle2caps[b'stream'] if c != b'v2'
                 ]
             if bundle2:
                 return False, None
             supported, requirements = orig(pullop, bundle2=bundle2)
             if requirements is not None:
                 requirements.add(constants.SHALLOWREPO_REQUIREMENT)
             return supported, requirements
 
         extensions.wrapfunction(
             streamclone, b'canperformstreamclone', canperformstreamclone
         )
 
     try:
         orig(ui, repo, *args, **opts)
     finally:
         if opts.get(r'shallow'):
             for r in repos:
                 if util.safehasattr(r, b'fileservice'):
                     r.fileservice.close()
 
 
 def debugdatashallow(orig, *args, **kwds):
     oldlen = remotefilelog.remotefilelog.__len__
     try:
         remotefilelog.remotefilelog.__len__ = lambda x: 1
         return orig(*args, **kwds)
     finally:
         remotefilelog.remotefilelog.__len__ = oldlen
 
 
 def reposetup(ui, repo):
     if not repo.local():
         return
 
     # put here intentionally bc doesnt work in uisetup
     ui.setconfig(b'hooks', b'update.prefetch', wcpprefetch)
     ui.setconfig(b'hooks', b'commit.prefetch', wcpprefetch)
 
     isserverenabled = ui.configbool(b'remotefilelog', b'server')
     isshallowclient = isenabled(repo)
 
     if isserverenabled and isshallowclient:
         raise RuntimeError(b"Cannot be both a server and shallow client.")
 
     if isshallowclient:
         setupclient(ui, repo)
 
     if isserverenabled:
         remotefilelogserver.setupserver(ui, repo)
 
 
 def setupclient(ui, repo):
     if not isinstance(repo, localrepo.localrepository):
         return
 
     # Even clients get the server setup since they need to have the
     # wireprotocol endpoints registered.
     remotefilelogserver.onetimesetup(ui)
     onetimeclientsetup(ui)
 
     shallowrepo.wraprepo(repo)
     repo.store = shallowstore.wrapstore(repo.store)
 
 
 def storewrapper(orig, requirements, path, vfstype):
     s = orig(requirements, path, vfstype)
     if constants.SHALLOWREPO_REQUIREMENT in requirements:
         s = shallowstore.wrapstore(s)
 
     return s
 
 
 # prefetch files before update
 def applyupdates(
     orig, repo, actions, wctx, mctx, overwrite, wantfiledata, labels=None
 ):
     if isenabled(repo):
         manifest = mctx.manifest()
         files = []
         for f, args, msg in actions[b'g']:
             files.append((f, hex(manifest[f])))
         # batch fetch the needed files from the server
         repo.fileservice.prefetch(files)
     return orig(
         repo, actions, wctx, mctx, overwrite, wantfiledata, labels=labels
     )
 
 
 # Prefetch merge checkunknownfiles
 def checkunknownfiles(orig, repo, wctx, mctx, force, actions, *args, **kwargs):
     if isenabled(repo):
         files = []
         sparsematch = repo.maybesparsematch(mctx.rev())
         for f, (m, actionargs, msg) in pycompat.iteritems(actions):
             if sparsematch and not sparsematch(f):
                 continue
             if m in (b'c', b'dc', b'cm'):
                 files.append((f, hex(mctx.filenode(f))))
             elif m == b'dg':
                 f2 = actionargs[0]
                 files.append((f2, hex(mctx.filenode(f2))))
         # batch fetch the needed files from the server
         repo.fileservice.prefetch(files)
     return orig(repo, wctx, mctx, force, actions, *args, **kwargs)
 
 
 # Prefetch files before status attempts to look at their size and contents
 def checklookup(orig, self, files):
     repo = self._repo
     if isenabled(repo):
         prefetchfiles = []
         for parent in self._parents:
             for f in files:
                 if f in parent:
                     prefetchfiles.append((f, hex(parent.filenode(f))))
         # batch fetch the needed files from the server
         repo.fileservice.prefetch(prefetchfiles)
     return orig(self, files)
 
 
 # Prefetch the logic that compares added and removed files for renames
 def findrenames(orig, repo, matcher, added, removed, *args, **kwargs):
     if isenabled(repo):
         files = []
         pmf = repo[b'.'].manifest()
         for f in removed:
             if f in pmf:
                 files.append((f, hex(pmf[f])))
         # batch fetch the needed files from the server
         repo.fileservice.prefetch(files)
     return orig(repo, matcher, added, removed, *args, **kwargs)
 
 
 # prefetch files before pathcopies check
 def computeforwardmissing(orig, a, b, match=None):
     missing = orig(a, b, match=match)
     repo = a._repo
     if isenabled(repo):
         mb = b.manifest()
 
         files = []
         sparsematch = repo.maybesparsematch(b.rev())
         if sparsematch:
             sparsemissing = set()
             for f in missing:
                 if sparsematch(f):
                     files.append((f, hex(mb[f])))
                     sparsemissing.add(f)
             missing = sparsemissing
 
         # batch fetch the needed files from the server
         repo.fileservice.prefetch(files)
     return missing
 
 
 # close cache miss server connection after the command has finished
 def runcommand(orig, lui, repo, *args, **kwargs):
     fileservice = None
     # repo can be None when running in chg:
     # - at startup, reposetup was called because serve is not norepo
     # - a norepo command like "help" is called
     if repo and isenabled(repo):
         fileservice = repo.fileservice
     try:
         return orig(lui, repo, *args, **kwargs)
     finally:
         if fileservice:
             fileservice.close()
 
 
 # prevent strip from stripping remotefilelogs
 def _collectbrokencsets(orig, repo, files, striprev):
     if isenabled(repo):
         files = list([f for f in files if not repo.shallowmatch(f)])
     return orig(repo, files, striprev)
 
 
 # changectx wrappers
 def filectx(orig, self, path, fileid=None, filelog=None):
     if fileid is None:
         fileid = self.filenode(path)
     if isenabled(self._repo) and self._repo.shallowmatch(path):
         return remotefilectx.remotefilectx(
             self._repo, path, fileid=fileid, changectx=self, filelog=filelog
         )
     return orig(self, path, fileid=fileid, filelog=filelog)
 
 
 def workingfilectx(orig, self, path, filelog=None):
     if isenabled(self._repo) and self._repo.shallowmatch(path):
         return remotefilectx.remoteworkingfilectx(
             self._repo, path, workingctx=self, filelog=filelog
         )
     return orig(self, path, filelog=filelog)
 
 
 # prefetch required revisions before a diff
 def trydiff(
     orig,
     repo,
     revs,
     ctx1,
     ctx2,
     modified,
     added,
     removed,
     copy,
     getfilectx,
     *args,
     **kwargs
 ):
     if isenabled(repo):
         prefetch = []
         mf1 = ctx1.manifest()
         for fname in modified + added + removed:
             if fname in mf1:
                 fnode = getfilectx(fname, ctx1).filenode()
                 # fnode can be None if it's a edited working ctx file
                 if fnode:
                     prefetch.append((fname, hex(fnode)))
             if fname not in removed:
                 fnode = getfilectx(fname, ctx2).filenode()
                 if fnode:
                     prefetch.append((fname, hex(fnode)))
 
         repo.fileservice.prefetch(prefetch)
 
     return orig(
         repo,
         revs,
         ctx1,
         ctx2,
         modified,
         added,
         removed,
         copy,
         getfilectx,
         *args,
         **kwargs
     )
 
 
 # Prevent verify from processing files
 # a stub for mercurial.hg.verify()
 def _verify(orig, repo, level=None):
     lock = repo.lock()
     try:
         return shallowverifier.shallowverifier(repo).verify()
     finally:
         lock.release()
 
 
 clientonetime = False
 
 
 def onetimeclientsetup(ui):
     global clientonetime
     if clientonetime:
         return
     clientonetime = True
 
     # Don't commit filelogs until we know the commit hash, since the hash
     # is present in the filelog blob.
     # This violates Mercurial's filelog->manifest->changelog write order,
     # but is generally fine for client repos.
     pendingfilecommits = []
 
     def addrawrevision(
         orig,
         self,
         rawtext,
         transaction,
         link,
         p1,
         p2,
         node,
         flags,
         cachedelta=None,
         _metatuple=None,
     ):
         if isinstance(link, int):
             pendingfilecommits.append(
                 (
                     self,
                     rawtext,
                     transaction,
                     link,
                     p1,
                     p2,
                     node,
                     flags,
                     cachedelta,
                     _metatuple,
                 )
             )
             return node
         else:
             return orig(
                 self,
                 rawtext,
                 transaction,
                 link,
                 p1,
                 p2,
                 node,
                 flags,
                 cachedelta,
                 _metatuple=_metatuple,
             )
 
     extensions.wrapfunction(
         remotefilelog.remotefilelog, b'addrawrevision', addrawrevision
     )
 
     def changelogadd(orig, self, *args):
         oldlen = len(self)
         node = orig(self, *args)
         newlen = len(self)
         if oldlen != newlen:
             for oldargs in pendingfilecommits:
                 log, rt, tr, link, p1, p2, n, fl, c, m = oldargs
                 linknode = self.node(link)
                 if linknode == node:
                     log.addrawrevision(rt, tr, linknode, p1, p2, n, fl, c, m)
                 else:
                     raise error.ProgrammingError(
                         b'pending multiple integer revisions are not supported'
                     )
         else:
             # "link" is actually wrong here (it is set to len(changelog))
             # if changelog remains unchanged, skip writing file revisions
             # but still do a sanity check about pending multiple revisions
             if len(set(x[3] for x in pendingfilecommits)) > 1:
                 raise error.ProgrammingError(
                     b'pending multiple integer revisions are not supported'
                 )
         del pendingfilecommits[:]
         return node
 
     extensions.wrapfunction(changelog.changelog, b'add', changelogadd)
 
 
 def getrenamedfn(orig, repo, endrev=None):
     if not isenabled(repo) or copies.usechangesetcentricalgo(repo):
         return orig(repo, endrev)
 
     rcache = {}
 
     def getrenamed(fn, rev):
         '''looks up all renames for a file (up to endrev) the first
         time the file is given. It indexes on the changerev and only
         parses the manifest if linkrev != changerev.
         Returns rename info for fn at changerev rev.'''
         if rev in rcache.setdefault(fn, {}):
             return rcache[fn][rev]
 
         try:
             fctx = repo[rev].filectx(fn)
             for ancestor in fctx.ancestors():
                 if ancestor.path() == fn:
                     renamed = ancestor.renamed()
                     rcache[fn][ancestor.rev()] = renamed and renamed[0]
 
             renamed = fctx.renamed()
             return renamed and renamed[0]
         except error.LookupError:
             return None
 
     return getrenamed
 
 
 def walkfilerevs(orig, repo, match, follow, revs, fncache):
     if not isenabled(repo):
         return orig(repo, match, follow, revs, fncache)
 
     # remotefilelog's can't be walked in rev order, so throw.
     # The caller will see the exception and walk the commit tree instead.
     if not follow:
         raise cmdutil.FileWalkError(b"Cannot walk via filelog")
 
     wanted = set()
     minrev, maxrev = min(revs), max(revs)
 
     pctx = repo[b'.']
     for filename in match.files():
         if filename not in pctx:
             raise error.Abort(
                 _(b'cannot follow file not in parent revision: "%s"') % filename
             )
         fctx = pctx[filename]
 
         linkrev = fctx.linkrev()
         if linkrev >= minrev and linkrev <= maxrev:
             fncache.setdefault(linkrev, []).append(filename)
             wanted.add(linkrev)
 
         for ancestor in fctx.ancestors():
             linkrev = ancestor.linkrev()
             if linkrev >= minrev and linkrev <= maxrev:
                 fncache.setdefault(linkrev, []).append(ancestor.path())
                 wanted.add(linkrev)
 
     return wanted
 
 
 def filelogrevset(orig, repo, subset, x):
     """``filelog(pattern)``
     Changesets connected to the specified filelog.
 
     For performance reasons, ``filelog()`` does not show every changeset
     that affects the requested file(s). See :hg:`help log` for details. For
     a slower, more accurate result, use ``file()``.
     """
 
     if not isenabled(repo):
         return orig(repo, subset, x)
 
     # i18n: "filelog" is a keyword
     pat = revset.getstring(x, _(b"filelog requires a pattern"))
     m = match.match(
         repo.root, repo.getcwd(), [pat], default=b'relpath', ctx=repo[None]
     )
     s = set()
 
     if not match.patkind(pat):
         # slow
         for r in subset:
             ctx = repo[r]
             cfiles = ctx.files()
             for f in m.files():
                 if f in cfiles:
                     s.add(ctx.rev())
                     break
     else:
         # partial
         files = (f for f in repo[None] if m(f))
         for f in files:
             fctx = repo[None].filectx(f)
             s.add(fctx.linkrev())
             for actx in fctx.ancestors():
                 s.add(actx.linkrev())
 
     return smartset.baseset([r for r in subset if r in s])
 
 
 @command(b'gc', [], _(b'hg gc [REPO...]'), norepo=True)
 def gc(ui, *args, **opts):
     '''garbage collect the client and server filelog caches
     '''
     cachepaths = set()
 
     # get the system client cache
     systemcache = shallowutil.getcachepath(ui, allowempty=True)
     if systemcache:
         cachepaths.add(systemcache)
 
     # get repo client and server cache
     repopaths = []
     pwd = ui.environ.get(b'PWD')
     if pwd:
         repopaths.append(pwd)
 
     repopaths.extend(args)
     repos = []
     for repopath in repopaths:
         try:
             repo = hg.peer(ui, {}, repopath)
             repos.append(repo)
 
             repocache = shallowutil.getcachepath(repo.ui, allowempty=True)
             if repocache:
                 cachepaths.add(repocache)
         except error.RepoError:
             pass
 
     # gc client cache
     for cachepath in cachepaths:
         gcclient(ui, cachepath)
 
     # gc server cache
     for repo in repos:
         remotefilelogserver.gcserver(ui, repo._repo)
 
 
 def gcclient(ui, cachepath):
     # get list of repos that use this cache
     repospath = os.path.join(cachepath, b'repos')
     if not os.path.exists(repospath):
         ui.warn(_(b"no known cache at %s\n") % cachepath)
         return
 
     reposfile = open(repospath, b'rb')
     repos = {r[:-1] for r in reposfile.readlines()}
     reposfile.close()
 
     # build list of useful files
     validrepos = []
     keepkeys = set()
 
     sharedcache = None
     filesrepacked = False
 
     count = 0
     progress = ui.makeprogress(
         _(b"analyzing repositories"), unit=b"repos", total=len(repos)
     )
     for path in repos:
         progress.update(count)
         count += 1
         try:
             path = ui.expandpath(os.path.normpath(path))
         except TypeError as e:
             ui.warn(_(b"warning: malformed path: %r:%s\n") % (path, e))
             traceback.print_exc()
             continue
         try:
             peer = hg.peer(ui, {}, path)
             repo = peer._repo
         except error.RepoError:
             continue
 
         validrepos.append(path)
 
         # Protect against any repo or config changes that have happened since
         # this repo was added to the repos file. We'd rather this loop succeed
         # and too much be deleted, than the loop fail and nothing gets deleted.
         if not isenabled(repo):
             continue
 
         if not util.safehasattr(repo, b'name'):
             ui.warn(
                 _(b"repo %s is a misconfigured remotefilelog repo\n") % path
             )
             continue
 
         # If garbage collection on repack and repack on hg gc are enabled
         # then loose files are repacked and garbage collected.
         # Otherwise regular garbage collection is performed.
         repackonhggc = repo.ui.configbool(b'remotefilelog', b'repackonhggc')
         gcrepack = repo.ui.configbool(b'remotefilelog', b'gcrepack')
         if repackonhggc and gcrepack:
             try:
                 repackmod.incrementalrepack(repo)
                 filesrepacked = True
                 continue
             except (IOError, repackmod.RepackAlreadyRunning):
                 # If repack cannot be performed due to not enough disk space
                 # continue doing garbage collection of loose files w/o repack
                 pass
 
         reponame = repo.name
         if not sharedcache:
             sharedcache = repo.sharedstore
 
         # Compute a keepset which is not garbage collected
         def keyfn(fname, fnode):
             return fileserverclient.getcachekey(reponame, fname, hex(fnode))
 
         keepkeys = repackmod.keepset(repo, keyfn=keyfn, lastkeepkeys=keepkeys)
 
     progress.complete()
 
     # write list of valid repos back
     oldumask = os.umask(0o002)
     try:
         reposfile = open(repospath, b'wb')
         reposfile.writelines([(b"%s\n" % r) for r in validrepos])
         reposfile.close()
     finally:
         os.umask(oldumask)
 
     # prune cache
     if sharedcache is not None:
         sharedcache.gc(keepkeys)
     elif not filesrepacked:
         ui.warn(_(b"warning: no valid repos in repofile\n"))
 
 
 def log(orig, ui, repo, *pats, **opts):
     if not isenabled(repo):
         return orig(ui, repo, *pats, **opts)
 
     follow = opts.get(r'follow')
     revs = opts.get(r'rev')
     if pats:
         # Force slowpath for non-follow patterns and follows that start from
         # non-working-copy-parent revs.
         if not follow or revs:
             # This forces the slowpath
             opts[r'removed'] = True
 
         # If this is a non-follow log without any revs specified, recommend that
         # the user add -f to speed it up.
         if not follow and not revs:
             match = scmutil.match(repo[b'.'], pats, pycompat.byteskwargs(opts))
             isfile = not match.anypats()
             if isfile:
                 for file in match.files():
                     if not os.path.isfile(repo.wjoin(file)):
                         isfile = False
                         break
 
             if isfile:
                 ui.warn(
                     _(
                         b"warning: file log can be slow on large repos - "
                         + b"use -f to speed it up\n"
                     )
                 )
 
     return orig(ui, repo, *pats, **opts)
 
 
 def revdatelimit(ui, revset):
     """Update revset so that only changesets no older than 'prefetchdays' days
     are included. The default value is set to 14 days. If 'prefetchdays' is set
     to zero or negative value then date restriction is not applied.
     """
     days = ui.configint(b'remotefilelog', b'prefetchdays')
     if days > 0:
         revset = b'(%s) & date(-%s)' % (revset, days)
     return revset
 
 
 def readytofetch(repo):
     """Check that enough time has passed since the last background prefetch.
     This only relates to prefetches after operations that change the working
     copy parent. Default delay between background prefetches is 2 minutes.
     """
     timeout = repo.ui.configint(b'remotefilelog', b'prefetchdelay')
     fname = repo.vfs.join(b'lastprefetch')
 
     ready = False
     with open(fname, b'a'):
         # the with construct above is used to avoid race conditions
         modtime = os.path.getmtime(fname)
         if (time.time() - modtime) > timeout:
             os.utime(fname, None)
             ready = True
 
     return ready
 
 
 def wcpprefetch(ui, repo, **kwargs):
     """Prefetches in background revisions specified by bgprefetchrevs revset.
     Does background repack if backgroundrepack flag is set in config.
     """
     shallow = isenabled(repo)
     bgprefetchrevs = ui.config(b'remotefilelog', b'bgprefetchrevs')
     isready = readytofetch(repo)
 
     if not (shallow and bgprefetchrevs and isready):
         return
 
     bgrepack = repo.ui.configbool(b'remotefilelog', b'backgroundrepack')
     # update a revset with a date limit
     bgprefetchrevs = revdatelimit(ui, bgprefetchrevs)
 
     def anon():
         if util.safehasattr(repo, b'ranprefetch') and repo.ranprefetch:
             return
         repo.ranprefetch = True
         repo.backgroundprefetch(bgprefetchrevs, repack=bgrepack)
 
     repo._afterlock(anon)
 
 
 def pull(orig, ui, repo, *pats, **opts):
     result = orig(ui, repo, *pats, **opts)
 
     if isenabled(repo):
         # prefetch if it's configured
         prefetchrevset = ui.config(b'remotefilelog', b'pullprefetch')
         bgrepack = repo.ui.configbool(b'remotefilelog', b'backgroundrepack')
         bgprefetch = repo.ui.configbool(b'remotefilelog', b'backgroundprefetch')
         ensurestart = repo.ui.configbool(b'devel', b'remotefilelog.ensurestart')
 
         if prefetchrevset:
             ui.status(_(b"prefetching file contents\n"))
             revs = scmutil.revrange(repo, [prefetchrevset])
             base = repo[b'.'].rev()
             if bgprefetch:
                 repo.backgroundprefetch(
                     prefetchrevset, repack=bgrepack, ensurestart=ensurestart
                 )
             else:
                 repo.prefetch(revs, base=base)
                 if bgrepack:
                     repackmod.backgroundrepack(
                         repo, incremental=True, ensurestart=ensurestart
                     )
         elif bgrepack:
             repackmod.backgroundrepack(
                 repo, incremental=True, ensurestart=ensurestart
             )
 
     return result
 
 
 def exchangepull(orig, repo, remote, *args, **kwargs):
     # Hook into the callstream/getbundle to insert bundle capabilities
     # during a pull.
     def localgetbundle(
         orig, source, heads=None, common=None, bundlecaps=None, **kwargs
     ):
         if not bundlecaps:
             bundlecaps = set()
         bundlecaps.add(constants.BUNDLE2_CAPABLITY)
         return orig(
             source, heads=heads, common=common, bundlecaps=bundlecaps, **kwargs
         )
 
     if util.safehasattr(remote, b'_callstream'):
         remote._localrepo = repo
     elif util.safehasattr(remote, b'getbundle'):
         extensions.wrapfunction(remote, b'getbundle', localgetbundle)
 
     return orig(repo, remote, *args, **kwargs)
 
 
 def _fileprefetchhook(repo, revs, match):
     if isenabled(repo):
         allfiles = []
         for rev in revs:
             if rev == nodemod.wdirrev or rev is None:
                 continue
             ctx = repo[rev]
             mf = ctx.manifest()
             sparsematch = repo.maybesparsematch(ctx.rev())
             for path in ctx.walk(match):
                 if (not sparsematch or sparsematch(path)) and path in mf:
                     allfiles.append((path, hex(mf[path])))
         repo.fileservice.prefetch(allfiles)
 
 
 @command(
     b'debugremotefilelog',
     [(b'd', b'decompress', None, _(b'decompress the filelog first')),],
     _(b'hg debugremotefilelog <path>'),
     norepo=True,
 )
 def debugremotefilelog(ui, path, **opts):
     return debugcommands.debugremotefilelog(ui, path, **opts)
 
 
 @command(
     b'verifyremotefilelog',
     [(b'd', b'decompress', None, _(b'decompress the filelogs first')),],
     _(b'hg verifyremotefilelogs <directory>'),
     norepo=True,
 )
 def verifyremotefilelog(ui, path, **opts):
     return debugcommands.verifyremotefilelog(ui, path, **opts)
 
 
 @command(
     b'debugdatapack',
     [
         (b'', b'long', None, _(b'print the long hashes')),
         (b'', b'node', b'', _(b'dump the contents of node'), b'NODE'),
     ],
     _(b'hg debugdatapack <paths>'),
     norepo=True,
 )
 def debugdatapack(ui, *paths, **opts):
     return debugcommands.debugdatapack(ui, *paths, **opts)
 
 
 @command(b'debughistorypack', [], _(b'hg debughistorypack <path>'), norepo=True)
 def debughistorypack(ui, path, **opts):
     return debugcommands.debughistorypack(ui, path)
 
 
 @command(b'debugkeepset', [], _(b'hg debugkeepset'))
 def debugkeepset(ui, repo, **opts):
     # The command is used to measure keepset computation time
     def keyfn(fname, fnode):
         return fileserverclient.getcachekey(repo.name, fname, hex(fnode))
 
     repackmod.keepset(repo, keyfn)
     return
 
 
 @command(b'debugwaitonrepack', [], _(b'hg debugwaitonrepack'))
 def debugwaitonrepack(ui, repo, **opts):
     return debugcommands.debugwaitonrepack(repo)
 
 
 @command(b'debugwaitonprefetch', [], _(b'hg debugwaitonprefetch'))
 def debugwaitonprefetch(ui, repo, **opts):
     return debugcommands.debugwaitonprefetch(repo)
 
 
 def resolveprefetchopts(ui, opts):
     if not opts.get(b'rev'):
         revset = [b'.', b'draft()']
 
         prefetchrevset = ui.config(b'remotefilelog', b'pullprefetch', None)
         if prefetchrevset:
             revset.append(b'(%s)' % prefetchrevset)
         bgprefetchrevs = ui.config(b'remotefilelog', b'bgprefetchrevs', None)
         if bgprefetchrevs:
             revset.append(b'(%s)' % bgprefetchrevs)
         revset = b'+'.join(revset)
 
         # update a revset with a date limit
         revset = revdatelimit(ui, revset)
 
         opts[b'rev'] = [revset]
 
     if not opts.get(b'base'):
         opts[b'base'] = None
 
     return opts
 
 
 @command(
     b'prefetch',
     [
         (b'r', b'rev', [], _(b'prefetch the specified revisions'), _(b'REV')),
         (b'', b'repack', False, _(b'run repack after prefetch')),
         (b'b', b'base', b'', _(b"rev that is assumed to already be local")),
     ]
     + commands.walkopts,
     _(b'hg prefetch [OPTIONS] [FILE...]'),
     helpcategory=command.CATEGORY_MAINTENANCE,
 )
 def prefetch(ui, repo, *pats, **opts):
     """prefetch file revisions from the server
 
     Prefetchs file revisions for the specified revs and stores them in the
     local remotefilelog cache.  If no rev is specified, the default rev is
     used which is the union of dot, draft, pullprefetch and bgprefetchrev.
     File names or patterns can be used to limit which files are downloaded.
 
     Return 0 on success.
     """
     opts = pycompat.byteskwargs(opts)
     if not isenabled(repo):
         raise error.Abort(_(b"repo is not shallow"))
 
     opts = resolveprefetchopts(ui, opts)
     revs = scmutil.revrange(repo, opts.get(b'rev'))
     repo.prefetch(revs, opts.get(b'base'), pats, opts)
 
     ensurestart = repo.ui.configbool(b'devel', b'remotefilelog.ensurestart')
 
     # Run repack in background
     if opts.get(b'repack'):
         repackmod.backgroundrepack(
             repo, incremental=True, ensurestart=ensurestart
         )
 
 
 @command(
     b'repack',
     [
         (b'', b'background', None, _(b'run in a background process'), None),
         (b'', b'incremental', None, _(b'do an incremental repack'), None),
         (
             b'',
             b'packsonly',
             None,
             _(b'only repack packs (skip loose objects)'),
             None,
         ),
     ],
     _(b'hg repack [OPTIONS]'),
 )
 def repack_(ui, repo, *pats, **opts):
     if opts.get(r'background'):
         ensurestart = repo.ui.configbool(b'devel', b'remotefilelog.ensurestart')
         repackmod.backgroundrepack(
             repo,
             incremental=opts.get(r'incremental'),
             packsonly=opts.get(r'packsonly', False),
             ensurestart=ensurestart,
         )
         return
 
     options = {b'packsonly': opts.get(r'packsonly')}
 
     try:
         if opts.get(r'incremental'):
             repackmod.incrementalrepack(repo, options=options)
         else:
             repackmod.fullrepack(repo, options=options)
     except repackmod.RepackAlreadyRunning as ex:
         # Don't propogate the exception if the repack is already in
         # progress, since we want the command to exit 0.
         repo.ui.warn(b'%s\n' % ex)

hgext/remotefilelog/repack.py

@@ -1,912 +1,918 b''
 from __future__ import absolute_import
 
 import os
 import time
 
 from mercurial.i18n import _
 from mercurial.node import (
     nullid,
     short,
 )
 from mercurial import (
     encoding,
     error,
     lock as lockmod,
     mdiff,
     policy,
     pycompat,
     scmutil,
     util,
     vfs,
 )
 from mercurial.utils import procutil
 from . import (
     constants,
     contentstore,
     datapack,
     historypack,
     metadatastore,
     shallowutil,
 )
 
 osutil = policy.importmod(r'osutil')
 
 
 class RepackAlreadyRunning(error.Abort):
     pass
 
 
 def backgroundrepack(
     repo, incremental=True, packsonly=False, ensurestart=False
 ):
     cmd = [procutil.hgexecutable(), b'-R', repo.origroot, b'repack']
     msg = _(b"(running background repack)\n")
     if incremental:
         cmd.append(b'--incremental')
         msg = _(b"(running background incremental repack)\n")
     if packsonly:
         cmd.append(b'--packsonly')
     repo.ui.warn(msg)
     # We know this command will find a binary, so don't block on it starting.
-    procutil.runbgcommand(cmd, encoding.environ, ensurestart=ensurestart)
+    kwargs = {}
+    if repo.ui.configbool(b'devel', b'remotefilelog.bg-wait'):
+        kwargs['record_wait'] = repo.ui.atexit
+
+    procutil.runbgcommand(
+        cmd, encoding.environ, ensurestart=ensurestart, **kwargs
+    )
 
 
 def fullrepack(repo, options=None):
     """If ``packsonly`` is True, stores creating only loose objects are skipped.
     """
     if util.safehasattr(repo, 'shareddatastores'):
         datasource = contentstore.unioncontentstore(*repo.shareddatastores)
         historysource = metadatastore.unionmetadatastore(
             *repo.sharedhistorystores, allowincomplete=True
         )
 
         packpath = shallowutil.getcachepackpath(
             repo, constants.FILEPACK_CATEGORY
         )
         _runrepack(
             repo,
             datasource,
             historysource,
             packpath,
             constants.FILEPACK_CATEGORY,
             options=options,
         )
 
     if util.safehasattr(repo.manifestlog, 'datastore'):
         localdata, shareddata = _getmanifeststores(repo)
         lpackpath, ldstores, lhstores = localdata
         spackpath, sdstores, shstores = shareddata
 
         # Repack the shared manifest store
         datasource = contentstore.unioncontentstore(*sdstores)
         historysource = metadatastore.unionmetadatastore(
             *shstores, allowincomplete=True
         )
         _runrepack(
             repo,
             datasource,
             historysource,
             spackpath,
             constants.TREEPACK_CATEGORY,
             options=options,
         )
 
         # Repack the local manifest store
         datasource = contentstore.unioncontentstore(
             *ldstores, allowincomplete=True
         )
         historysource = metadatastore.unionmetadatastore(
             *lhstores, allowincomplete=True
         )
         _runrepack(
             repo,
             datasource,
             historysource,
             lpackpath,
             constants.TREEPACK_CATEGORY,
             options=options,
         )
 
 
 def incrementalrepack(repo, options=None):
     """This repacks the repo by looking at the distribution of pack files in the
     repo and performing the most minimal repack to keep the repo in good shape.
     """
     if util.safehasattr(repo, 'shareddatastores'):
         packpath = shallowutil.getcachepackpath(
             repo, constants.FILEPACK_CATEGORY
         )
         _incrementalrepack(
             repo,
             repo.shareddatastores,
             repo.sharedhistorystores,
             packpath,
             constants.FILEPACK_CATEGORY,
             options=options,
         )
 
     if util.safehasattr(repo.manifestlog, 'datastore'):
         localdata, shareddata = _getmanifeststores(repo)
         lpackpath, ldstores, lhstores = localdata
         spackpath, sdstores, shstores = shareddata
 
         # Repack the shared manifest store
         _incrementalrepack(
             repo,
             sdstores,
             shstores,
             spackpath,
             constants.TREEPACK_CATEGORY,
             options=options,
         )
 
         # Repack the local manifest store
         _incrementalrepack(
             repo,
             ldstores,
             lhstores,
             lpackpath,
             constants.TREEPACK_CATEGORY,
             allowincompletedata=True,
             options=options,
         )
 
 
 def _getmanifeststores(repo):
     shareddatastores = repo.manifestlog.shareddatastores
     localdatastores = repo.manifestlog.localdatastores
     sharedhistorystores = repo.manifestlog.sharedhistorystores
     localhistorystores = repo.manifestlog.localhistorystores
 
     sharedpackpath = shallowutil.getcachepackpath(
         repo, constants.TREEPACK_CATEGORY
     )
     localpackpath = shallowutil.getlocalpackpath(
         repo.svfs.vfs.base, constants.TREEPACK_CATEGORY
     )
 
     return (
         (localpackpath, localdatastores, localhistorystores),
         (sharedpackpath, shareddatastores, sharedhistorystores),
     )
 
 
 def _topacks(packpath, files, constructor):
     paths = list(os.path.join(packpath, p) for p in files)
     packs = list(constructor(p) for p in paths)
     return packs
 
 
 def _deletebigpacks(repo, folder, files):
     """Deletes packfiles that are bigger than ``packs.maxpacksize``.
 
     Returns ``files` with the removed files omitted."""
     maxsize = repo.ui.configbytes(b"packs", b"maxpacksize")
     if maxsize <= 0:
         return files
 
     # This only considers datapacks today, but we could broaden it to include
     # historypacks.
     VALIDEXTS = [b".datapack", b".dataidx"]
 
     # Either an oversize index or datapack will trigger cleanup of the whole
     # pack:
     oversized = {
         os.path.splitext(path)[0]
         for path, ftype, stat in files
         if (stat.st_size > maxsize and (os.path.splitext(path)[1] in VALIDEXTS))
     }
 
     for rootfname in oversized:
         rootpath = os.path.join(folder, rootfname)
         for ext in VALIDEXTS:
             path = rootpath + ext
             repo.ui.debug(
                 b'removing oversize packfile %s (%s)\n'
                 % (path, util.bytecount(os.stat(path).st_size))
             )
             os.unlink(path)
     return [row for row in files if os.path.basename(row[0]) not in oversized]
 
 
 def _incrementalrepack(
     repo,
     datastore,
     historystore,
     packpath,
     category,
     allowincompletedata=False,
     options=None,
 ):
     shallowutil.mkstickygroupdir(repo.ui, packpath)
 
     files = osutil.listdir(packpath, stat=True)
     files = _deletebigpacks(repo, packpath, files)
     datapacks = _topacks(
         packpath, _computeincrementaldatapack(repo.ui, files), datapack.datapack
     )
     datapacks.extend(
         s for s in datastore if not isinstance(s, datapack.datapackstore)
     )
 
     historypacks = _topacks(
         packpath,
         _computeincrementalhistorypack(repo.ui, files),
         historypack.historypack,
     )
     historypacks.extend(
         s
         for s in historystore
         if not isinstance(s, historypack.historypackstore)
     )
 
     # ``allhistory{files,packs}`` contains all known history packs, even ones we
     # don't plan to repack. They are used during the datapack repack to ensure
     # good ordering of nodes.
     allhistoryfiles = _allpackfileswithsuffix(
         files, historypack.PACKSUFFIX, historypack.INDEXSUFFIX
     )
     allhistorypacks = _topacks(
         packpath,
         (f for f, mode, stat in allhistoryfiles),
         historypack.historypack,
     )
     allhistorypacks.extend(
         s
         for s in historystore
         if not isinstance(s, historypack.historypackstore)
     )
     _runrepack(
         repo,
         contentstore.unioncontentstore(
             *datapacks, allowincomplete=allowincompletedata
         ),
         metadatastore.unionmetadatastore(*historypacks, allowincomplete=True),
         packpath,
         category,
         fullhistory=metadatastore.unionmetadatastore(
             *allhistorypacks, allowincomplete=True
         ),
         options=options,
     )
 
 
 def _computeincrementaldatapack(ui, files):
     opts = {
         b'gencountlimit': ui.configint(b'remotefilelog', b'data.gencountlimit'),
         b'generations': ui.configlist(b'remotefilelog', b'data.generations'),
         b'maxrepackpacks': ui.configint(
             b'remotefilelog', b'data.maxrepackpacks'
         ),
         b'repackmaxpacksize': ui.configbytes(
             b'remotefilelog', b'data.repackmaxpacksize'
         ),
         b'repacksizelimit': ui.configbytes(
             b'remotefilelog', b'data.repacksizelimit'
         ),
     }
 
     packfiles = _allpackfileswithsuffix(
         files, datapack.PACKSUFFIX, datapack.INDEXSUFFIX
     )
     return _computeincrementalpack(packfiles, opts)
 
 
 def _computeincrementalhistorypack(ui, files):
     opts = {
         b'gencountlimit': ui.configint(
             b'remotefilelog', b'history.gencountlimit'
         ),
         b'generations': ui.configlist(
             b'remotefilelog', b'history.generations', [b'100MB']
         ),
         b'maxrepackpacks': ui.configint(
             b'remotefilelog', b'history.maxrepackpacks'
         ),
         b'repackmaxpacksize': ui.configbytes(
             b'remotefilelog', b'history.repackmaxpacksize', b'400MB'
         ),
         b'repacksizelimit': ui.configbytes(
             b'remotefilelog', b'history.repacksizelimit'
         ),
     }
 
     packfiles = _allpackfileswithsuffix(
         files, historypack.PACKSUFFIX, historypack.INDEXSUFFIX
     )
     return _computeincrementalpack(packfiles, opts)
 
 
 def _allpackfileswithsuffix(files, packsuffix, indexsuffix):
     result = []
     fileset = set(fn for fn, mode, stat in files)
     for filename, mode, stat in files:
         if not filename.endswith(packsuffix):
             continue
 
         prefix = filename[: -len(packsuffix)]
 
         # Don't process a pack if it doesn't have an index.
         if (prefix + indexsuffix) not in fileset:
             continue
         result.append((prefix, mode, stat))
 
     return result
 
 
 def _computeincrementalpack(files, opts):
     """Given a set of pack files along with the configuration options, this
     function computes the list of files that should be packed as part of an
     incremental repack.
 
     It tries to strike a balance between keeping incremental repacks cheap (i.e.
     packing small things when possible, and rolling the packs up to the big ones
     over time).
     """
 
     limits = list(
         sorted((util.sizetoint(s) for s in opts[b'generations']), reverse=True)
     )
     limits.append(0)
 
     # Group the packs by generation (i.e. by size)
     generations = []
     for i in pycompat.xrange(len(limits)):
         generations.append([])
 
     sizes = {}
     for prefix, mode, stat in files:
         size = stat.st_size
         if size > opts[b'repackmaxpacksize']:
             continue
 
         sizes[prefix] = size
         for i, limit in enumerate(limits):
             if size > limit:
                 generations[i].append(prefix)
                 break
 
     # Steps for picking what packs to repack:
     # 1. Pick the largest generation with > gencountlimit pack files.
     # 2. Take the smallest three packs.
     # 3. While total-size-of-packs < repacksizelimit: add another pack
 
     # Find the largest generation with more than gencountlimit packs
     genpacks = []
     for i, limit in enumerate(limits):
         if len(generations[i]) > opts[b'gencountlimit']:
             # Sort to be smallest last, for easy popping later
             genpacks.extend(
                 sorted(generations[i], reverse=True, key=lambda x: sizes[x])
             )
             break
 
     # Take as many packs from the generation as we can
     chosenpacks = genpacks[-3:]
     genpacks = genpacks[:-3]
     repacksize = sum(sizes[n] for n in chosenpacks)
     while (
         repacksize < opts[b'repacksizelimit']
         and genpacks
         and len(chosenpacks) < opts[b'maxrepackpacks']
     ):
         chosenpacks.append(genpacks.pop())
         repacksize += sizes[chosenpacks[-1]]
 
     return chosenpacks
 
 
 def _runrepack(
     repo, data, history, packpath, category, fullhistory=None, options=None
 ):
     shallowutil.mkstickygroupdir(repo.ui, packpath)
 
     def isold(repo, filename, node):
         """Check if the file node is older than a limit.
         Unless a limit is specified in the config the default limit is taken.
         """
         filectx = repo.filectx(filename, fileid=node)
         filetime = repo[filectx.linkrev()].date()
 
         ttl = repo.ui.configint(b'remotefilelog', b'nodettl')
 
         limit = time.time() - ttl
         return filetime[0] < limit
 
     garbagecollect = repo.ui.configbool(b'remotefilelog', b'gcrepack')
     if not fullhistory:
         fullhistory = history
     packer = repacker(
         repo,
         data,
         history,
         fullhistory,
         category,
         gc=garbagecollect,
         isold=isold,
         options=options,
     )
 
     with datapack.mutabledatapack(repo.ui, packpath) as dpack:
         with historypack.mutablehistorypack(repo.ui, packpath) as hpack:
             try:
                 packer.run(dpack, hpack)
             except error.LockHeld:
                 raise RepackAlreadyRunning(
                     _(
                         b"skipping repack - another repack "
                         b"is already running"
                     )
                 )
 
 
 def keepset(repo, keyfn, lastkeepkeys=None):
     """Computes a keepset which is not garbage collected.
     'keyfn' is a function that maps filename, node to a unique key.
     'lastkeepkeys' is an optional argument and if provided the keepset
     function updates lastkeepkeys with more keys and returns the result.
     """
     if not lastkeepkeys:
         keepkeys = set()
     else:
         keepkeys = lastkeepkeys
 
     # We want to keep:
     # 1. Working copy parent
     # 2. Draft commits
     # 3. Parents of draft commits
     # 4. Pullprefetch and bgprefetchrevs revsets if specified
     revs = [b'.', b'draft()', b'parents(draft())']
     prefetchrevs = repo.ui.config(b'remotefilelog', b'pullprefetch', None)
     if prefetchrevs:
         revs.append(b'(%s)' % prefetchrevs)
     prefetchrevs = repo.ui.config(b'remotefilelog', b'bgprefetchrevs', None)
     if prefetchrevs:
         revs.append(b'(%s)' % prefetchrevs)
     revs = b'+'.join(revs)
 
     revs = [b'sort((%s), "topo")' % revs]
     keep = scmutil.revrange(repo, revs)
 
     processed = set()
     lastmanifest = None
 
     # process the commits in toposorted order starting from the oldest
     for r in reversed(keep._list):
         if repo[r].p1().rev() in processed:
             # if the direct parent has already been processed
             # then we only need to process the delta
             m = repo[r].manifestctx().readdelta()
         else:
             # otherwise take the manifest and diff it
             # with the previous manifest if one exists
             if lastmanifest:
                 m = repo[r].manifest().diff(lastmanifest)
             else:
                 m = repo[r].manifest()
         lastmanifest = repo[r].manifest()
         processed.add(r)
 
         # populate keepkeys with keys from the current manifest
         if type(m) is dict:
             # m is a result of diff of two manifests and is a dictionary that
             # maps filename to ((newnode, newflag), (oldnode, oldflag)) tuple
             for filename, diff in pycompat.iteritems(m):
                 if diff[0][0] is not None:
                     keepkeys.add(keyfn(filename, diff[0][0]))
         else:
             # m is a manifest object
             for filename, filenode in pycompat.iteritems(m):
                 keepkeys.add(keyfn(filename, filenode))
 
     return keepkeys
 
 
 class repacker(object):
     """Class for orchestrating the repack of data and history information into a
     new format.
     """
 
     def __init__(
         self,
         repo,
         data,
         history,
         fullhistory,
         category,
         gc=False,
         isold=None,
         options=None,
     ):
         self.repo = repo
         self.data = data
         self.history = history
         self.fullhistory = fullhistory
         self.unit = constants.getunits(category)
         self.garbagecollect = gc
         self.options = options
         if self.garbagecollect:
             if not isold:
                 raise ValueError(b"Function 'isold' is not properly specified")
             # use (filename, node) tuple as a keepset key
             self.keepkeys = keepset(repo, lambda f, n: (f, n))
             self.isold = isold
 
     def run(self, targetdata, targethistory):
         ledger = repackledger()
 
         with lockmod.lock(
             repacklockvfs(self.repo), b"repacklock", desc=None, timeout=0
         ):
             self.repo.hook(b'prerepack')
 
             # Populate ledger from source
             self.data.markledger(ledger, options=self.options)
             self.history.markledger(ledger, options=self.options)
 
             # Run repack
             self.repackdata(ledger, targetdata)
             self.repackhistory(ledger, targethistory)
 
             # Call cleanup on each source
             for source in ledger.sources:
                 source.cleanup(ledger)
 
     def _chainorphans(self, ui, filename, nodes, orphans, deltabases):
         """Reorderes ``orphans`` into a single chain inside ``nodes`` and
         ``deltabases``.
 
         We often have orphan entries (nodes without a base that aren't
         referenced by other nodes -- i.e., part of a chain) due to gaps in
         history. Rather than store them as individual fulltexts, we prefer to
         insert them as one chain sorted by size.
         """
         if not orphans:
             return nodes
 
         def getsize(node, default=0):
             meta = self.data.getmeta(filename, node)
             if constants.METAKEYSIZE in meta:
                 return meta[constants.METAKEYSIZE]
             else:
                 return default
 
         # Sort orphans by size; biggest first is preferred, since it's more
         # likely to be the newest version assuming files grow over time.
         # (Sort by node first to ensure the sort is stable.)
         orphans = sorted(orphans)
         orphans = list(sorted(orphans, key=getsize, reverse=True))
         if ui.debugflag:
             ui.debug(
                 b"%s: orphan chain: %s\n"
                 % (filename, b", ".join([short(s) for s in orphans]))
             )
 
         # Create one contiguous chain and reassign deltabases.
         for i, node in enumerate(orphans):
             if i == 0:
                 deltabases[node] = (nullid, 0)
             else:
                 parent = orphans[i - 1]
                 deltabases[node] = (parent, deltabases[parent][1] + 1)
         nodes = [n for n in nodes if n not in orphans]
         nodes += orphans
         return nodes
 
     def repackdata(self, ledger, target):
         ui = self.repo.ui
         maxchainlen = ui.configint(b'packs', b'maxchainlen', 1000)
 
         byfile = {}
         for entry in pycompat.itervalues(ledger.entries):
             if entry.datasource:
                 byfile.setdefault(entry.filename, {})[entry.node] = entry
 
         count = 0
         repackprogress = ui.makeprogress(
             _(b"repacking data"), unit=self.unit, total=len(byfile)
         )
         for filename, entries in sorted(pycompat.iteritems(byfile)):
             repackprogress.update(count)
 
             ancestors = {}
             nodes = list(node for node in entries)
             nohistory = []
             buildprogress = ui.makeprogress(
                 _(b"building history"), unit=b'nodes', total=len(nodes)
             )
             for i, node in enumerate(nodes):
                 if node in ancestors:
                     continue
                 buildprogress.update(i)
                 try:
                     ancestors.update(
                         self.fullhistory.getancestors(
                             filename, node, known=ancestors
                         )
                     )
                 except KeyError:
                     # Since we're packing data entries, we may not have the
                     # corresponding history entries for them. It's not a big
                     # deal, but the entries won't be delta'd perfectly.
                     nohistory.append(node)
             buildprogress.complete()
 
             # Order the nodes children first, so we can produce reverse deltas
             orderednodes = list(reversed(self._toposort(ancestors)))
             if len(nohistory) > 0:
                 ui.debug(
                     b'repackdata: %d nodes without history\n' % len(nohistory)
                 )
             orderednodes.extend(sorted(nohistory))
 
             # Filter orderednodes to just the nodes we want to serialize (it
             # currently also has the edge nodes' ancestors).
             orderednodes = list(
                 filter(lambda node: node in nodes, orderednodes)
             )
 
             # Garbage collect old nodes:
             if self.garbagecollect:
                 neworderednodes = []
                 for node in orderednodes:
                     # If the node is old and is not in the keepset, we skip it,
                     # and mark as garbage collected
                     if (filename, node) not in self.keepkeys and self.isold(
                         self.repo, filename, node
                     ):
                         entries[node].gced = True
                         continue
                     neworderednodes.append(node)
                 orderednodes = neworderednodes
 
             # Compute delta bases for nodes:
             deltabases = {}
             nobase = set()
             referenced = set()
             nodes = set(nodes)
             processprogress = ui.makeprogress(
                 _(b"processing nodes"), unit=b'nodes', total=len(orderednodes)
             )
             for i, node in enumerate(orderednodes):
                 processprogress.update(i)
                 # Find delta base
                 # TODO: allow delta'ing against most recent descendant instead
                 # of immediate child
                 deltatuple = deltabases.get(node, None)
                 if deltatuple is None:
                     deltabase, chainlen = nullid, 0
                     deltabases[node] = (nullid, 0)
                     nobase.add(node)
                 else:
                     deltabase, chainlen = deltatuple
                     referenced.add(deltabase)
 
                 # Use available ancestor information to inform our delta choices
                 ancestorinfo = ancestors.get(node)
                 if ancestorinfo:
                     p1, p2, linknode, copyfrom = ancestorinfo
 
                     # The presence of copyfrom means we're at a point where the
                     # file was copied from elsewhere. So don't attempt to do any
                     # deltas with the other file.
                     if copyfrom:
                         p1 = nullid
 
                     if chainlen < maxchainlen:
                         # Record this child as the delta base for its parents.
                         # This may be non optimal, since the parents may have
                         # many children, and this will only choose the last one.
                         # TODO: record all children and try all deltas to find
                         # best
                         if p1 != nullid:
                             deltabases[p1] = (node, chainlen + 1)
                         if p2 != nullid:
                             deltabases[p2] = (node, chainlen + 1)
 
             # experimental config: repack.chainorphansbysize
             if ui.configbool(b'repack', b'chainorphansbysize'):
                 orphans = nobase - referenced
                 orderednodes = self._chainorphans(
                     ui, filename, orderednodes, orphans, deltabases
                 )
 
             # Compute deltas and write to the pack
             for i, node in enumerate(orderednodes):
                 deltabase, chainlen = deltabases[node]
                 # Compute delta
                 # TODO: Optimize the deltachain fetching. Since we're
                 # iterating over the different version of the file, we may
                 # be fetching the same deltachain over and over again.
                 if deltabase != nullid:
                     deltaentry = self.data.getdelta(filename, node)
                     delta, deltabasename, origdeltabase, meta = deltaentry
                     size = meta.get(constants.METAKEYSIZE)
                     if (
                         deltabasename != filename
                         or origdeltabase != deltabase
                         or size is None
                     ):
                         deltabasetext = self.data.get(filename, deltabase)
                         original = self.data.get(filename, node)
                         size = len(original)
                         delta = mdiff.textdiff(deltabasetext, original)
                 else:
                     delta = self.data.get(filename, node)
                     size = len(delta)
                     meta = self.data.getmeta(filename, node)
 
                 # TODO: don't use the delta if it's larger than the fulltext
                 if constants.METAKEYSIZE not in meta:
                     meta[constants.METAKEYSIZE] = size
                 target.add(filename, node, deltabase, delta, meta)
 
                 entries[node].datarepacked = True
 
             processprogress.complete()
             count += 1
 
         repackprogress.complete()
         target.close(ledger=ledger)
 
     def repackhistory(self, ledger, target):
         ui = self.repo.ui
 
         byfile = {}
         for entry in pycompat.itervalues(ledger.entries):
             if entry.historysource:
                 byfile.setdefault(entry.filename, {})[entry.node] = entry
 
         progress = ui.makeprogress(
             _(b"repacking history"), unit=self.unit, total=len(byfile)
         )
         for filename, entries in sorted(pycompat.iteritems(byfile)):
             ancestors = {}
             nodes = list(node for node in entries)
 
             for node in nodes:
                 if node in ancestors:
                     continue
                 ancestors.update(
                     self.history.getancestors(filename, node, known=ancestors)
                 )
 
             # Order the nodes children first
             orderednodes = reversed(self._toposort(ancestors))
 
             # Write to the pack
             dontprocess = set()
             for node in orderednodes:
                 p1, p2, linknode, copyfrom = ancestors[node]
 
                 # If the node is marked dontprocess, but it's also in the
                 # explicit entries set, that means the node exists both in this
                 # file and in another file that was copied to this file.
                 # Usually this happens if the file was copied to another file,
                 # then the copy was deleted, then reintroduced without copy
                 # metadata. The original add and the new add have the same hash
                 # since the content is identical and the parents are null.
                 if node in dontprocess and node not in entries:
                     # If copyfrom == filename, it means the copy history
                     # went to come other file, then came back to this one, so we
                     # should continue processing it.
                     if p1 != nullid and copyfrom != filename:
                         dontprocess.add(p1)
                     if p2 != nullid:
                         dontprocess.add(p2)
                     continue
 
                 if copyfrom:
                     dontprocess.add(p1)
 
                 target.add(filename, node, p1, p2, linknode, copyfrom)
 
                 if node in entries:
                     entries[node].historyrepacked = True
 
             progress.increment()
 
         progress.complete()
         target.close(ledger=ledger)
 
     def _toposort(self, ancestors):
         def parentfunc(node):
             p1, p2, linknode, copyfrom = ancestors[node]
             parents = []
             if p1 != nullid:
                 parents.append(p1)
             if p2 != nullid:
                 parents.append(p2)
             return parents
 
         sortednodes = shallowutil.sortnodes(ancestors.keys(), parentfunc)
         return sortednodes
 
 
 class repackledger(object):
     """Storage for all the bookkeeping that happens during a repack. It contains
     the list of revisions being repacked, what happened to each revision, and
     which source store contained which revision originally (for later cleanup).
     """
 
     def __init__(self):
         self.entries = {}
         self.sources = {}
         self.created = set()
 
     def markdataentry(self, source, filename, node):
         """Mark the given filename+node revision as having a data rev in the
         given source.
         """
         entry = self._getorcreateentry(filename, node)
         entry.datasource = True
         entries = self.sources.get(source)
         if not entries:
             entries = set()
             self.sources[source] = entries
         entries.add(entry)
 
     def markhistoryentry(self, source, filename, node):
         """Mark the given filename+node revision as having a history rev in the
         given source.
         """
         entry = self._getorcreateentry(filename, node)
         entry.historysource = True
         entries = self.sources.get(source)
         if not entries:
             entries = set()
             self.sources[source] = entries
         entries.add(entry)
 
     def _getorcreateentry(self, filename, node):
         key = (filename, node)
         value = self.entries.get(key)
         if not value:
             value = repackentry(filename, node)
             self.entries[key] = value
 
         return value
 
     def addcreated(self, value):
         self.created.add(value)
 
 
 class repackentry(object):
     """Simple class representing a single revision entry in the repackledger.
     """
 
     __slots__ = (
         r'filename',
         r'node',
         r'datasource',
         r'historysource',
         r'datarepacked',
         r'historyrepacked',
         r'gced',
     )
 
     def __init__(self, filename, node):
         self.filename = filename
         self.node = node
         # If the revision has a data entry in the source
         self.datasource = False
         # If the revision has a history entry in the source
         self.historysource = False
         # If the revision's data entry was repacked into the repack target
         self.datarepacked = False
         # If the revision's history entry was repacked into the repack target
         self.historyrepacked = False
         # If garbage collected
         self.gced = False
 
 
 def repacklockvfs(repo):
     if util.safehasattr(repo, 'name'):
         # Lock in the shared cache so repacks across multiple copies of the same
         # repo are coordinated.
         sharedcachepath = shallowutil.getcachepackpath(
             repo, constants.FILEPACK_CATEGORY
         )
         return vfs.vfs(sharedcachepath)
     else:
         return repo.svfs

hgext/remotefilelog/shallowrepo.py

@@ -1,354 +1,358 b''
 # 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,
     pycompat,
     scmutil,
     sparse,
     util,
 )
 from mercurial.utils import procutil
 from . import (
     connectionpool,
     constants,
     contentstore,
     datapack,
     fileserverclient,
     historypack,
     metadatastore,
     remotefilectx,
     remotefilelog,
     shallowutil,
 )
 
 # 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, b'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, b'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, b'filestore', packcontentstore, packmetadatastore
     )
 
 
 def wraprepo(repo):
     class shallowrepository(repo.__class__):
         @util.propertycache
         def name(self):
             return self.ui.config(b'remotefilelog', b'reponame')
 
         @util.propertycache
         def fallbackpath(self):
             path = repo.ui.config(
                 b"remotefilelog",
                 b"fallbackpath",
                 repo.ui.config(b'paths', b'default'),
             )
             if not path:
                 raise error.Abort(
                     b"no remotefilelog server "
                     b"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] == b'/':
                 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, origctx=None):
             """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, origctx=origctx
             )
 
         def backgroundprefetch(
             self,
             revs,
             base=None,
             repack=False,
             pats=None,
             opts=None,
             ensurestart=False,
         ):
             """Runs prefetch in background with optional repack
             """
             cmd = [procutil.hgexecutable(), b'-R', repo.origroot, b'prefetch']
             if repack:
                 cmd.append(b'--repack')
             if revs:
                 cmd += [b'-r', revs]
             # We know this command will find a binary, so don't block
             # on it starting.
+            kwargs = {}
+            if repo.ui.configbool(b'devel', b'remotefilelog.bg-wait'):
+                kwargs['record_wait'] = repo.ui.atexit
+
             procutil.runbgcommand(
-                cmd, encoding.environ, ensurestart=ensurestart
+                cmd, encoding.environ, ensurestart=ensurestart, **kwargs
             )
 
         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,
                 b'prefetchlock',
                 True,
                 None,
                 None,
                 _(b'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(b'outgoing(%s)', fallbackpath)
                 if base is not None and base != nullrev:
                     serverbase = list(
                         repo.revs(
                             b'first(reverse(::%s) - %ld)', base, localrevs
                         )
                     )
                     if serverbase:
                         base = serverbase[0]
             else:
                 localrevs = repo
 
             mfl = repo.manifestlog
             mfrevlog = mfl.getstorage(b'')
             if base is not None:
                 mfdict = mfl[repo[base].manifestnode()].read()
                 skip = set(pycompat.iteritems(mfdict))
             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(_(b'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 = pycompat.iteritems(mfdict)
                 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(
         b"remotefilelog", b"includepattern", None
     )
     repo.excludepattern = repo.ui.configlist(
         b"remotefilelog", b"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, b'', None, repo.includepattern, repo.excludepattern
         )