hgext/largefiles/lfcommands.py

revlog: add an experimental option to mitigated delta issues (issue5480)

The general delta heuristic to select a delta do not scale with the number of
branch. The delta base is frequently too far away to be able to reuse a chain
according to the "distance" criteria. This leads to insertion of larger delta (or
even full text) that themselves push the bases for the next delta further away
leading to more large deltas and full texts. This full text and frequent
recomputation throw Mercurial performance in disarray.

For example of a slightly large repository

  280 000 files (2 150 000 versions)
  430 000 changesets (10 000 topological heads)

Number below compares repository with and without the distance criteria:

manifest size:
    with:    21.4 GB
    without:  0.3 GB

store size:
    with:    28.7 GB
    without   7.4 GB

bundle last 15 00 revisions:
    with:    800 seconds
             971 MB
    without:  50 seconds
              73 MB

unbundle time (of the last 15K revisions):
    with:    1150 seconds (~19 minutes)
    without:   35 seconds

Similar issues has been observed in other repositories.


Adding a new option or "feature" on stable is uncommon. However, given that this
issues is making Mercurial practically unusable, I'm exceptionally targeting
this patch for stable.

What is actually needed is a full rework of the delta building and reading
logic. However, that will be a longer process and churn not suitable for stable.

In the meantime, we introduces a quick and dirty mitigation of this in the
'experimental' config space. The new option introduces a way to set the maximum
amount of memory usable to store a diff in memory. This extend the ability for
Mercurial to create chains without removing all safe guard regarding memory
access. The option should be phased out when core has a more proper solution
available.

Setting the limit to '0' remove all limits, setting it to '-1' use the default
limit (textsize x 4).

# Copyright 2009-2010 Gregory P. Ward
# Copyright 2009-2010 Intelerad Medical Systems Incorporated
# Copyright 2010-2011 Fog Creek Software
# Copyright 2010-2011 Unity Technologies
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.

'''High-level command function for lfconvert, plus the cmdtable.'''
from __future__ import absolute_import

import errno
import hashlib
import os
import shutil

from mercurial.i18n import _

from mercurial import (
    cmdutil,
    commands,
    context,
    error,
    hg,
    lock,
    match as matchmod,
    node,
    scmutil,
    util,
)

from ..convert import (
    convcmd,
    filemap,
)

from . import (
    lfutil,
    storefactory
)

release = lock.release

# -- Commands ----------------------------------------------------------

cmdtable = {}
command = cmdutil.command(cmdtable)

@command('lfconvert',
    [('s', 'size', '',
      _('minimum size (MB) for files to be converted as largefiles'), 'SIZE'),
    ('', 'to-normal', False,
     _('convert from a largefiles repo to a normal repo')),
    ],
    _('hg lfconvert SOURCE DEST [FILE ...]'),
    norepo=True,
    inferrepo=True)
def lfconvert(ui, src, dest, *pats, **opts):
    '''convert a normal repository to a largefiles repository

    Convert repository SOURCE to a new repository DEST, identical to
    SOURCE except that certain files will be converted as largefiles:
    specifically, any file that matches any PATTERN *or* whose size is
    above the minimum size threshold is converted as a largefile. The
    size used to determine whether or not to track a file as a
    largefile is the size of the first version of the file. The
    minimum size can be specified either with --size or in
    configuration as ``largefiles.size``.

    After running this command you will need to make sure that
    largefiles is enabled anywhere you intend to push the new
    repository.

    Use --to-normal to convert largefiles back to normal files; after
    this, the DEST repository can be used without largefiles at all.'''

    if opts['to_normal']:
        tolfile = False
    else:
        tolfile = True
        size = lfutil.getminsize(ui, True, opts.get('size'), default=None)

    if not hg.islocal(src):
        raise error.Abort(_('%s is not a local Mercurial repo') % src)
    if not hg.islocal(dest):
        raise error.Abort(_('%s is not a local Mercurial repo') % dest)

    rsrc = hg.repository(ui, src)
    ui.status(_('initializing destination %s\n') % dest)
    rdst = hg.repository(ui, dest, create=True)

    success = False
    dstwlock = dstlock = None
    try:
        # Get a list of all changesets in the source.  The easy way to do this
        # is to simply walk the changelog, using changelog.nodesbetween().
        # Take a look at mercurial/revlog.py:639 for more details.
        # Use a generator instead of a list to decrease memory usage
        ctxs = (rsrc[ctx] for ctx in rsrc.changelog.nodesbetween(None,
            rsrc.heads())[0])
        revmap = {node.nullid: node.nullid}
        if tolfile:
            # Lock destination to prevent modification while it is converted to.
            # Don't need to lock src because we are just reading from its
            # history which can't change.
            dstwlock = rdst.wlock()
            dstlock = rdst.lock()

            lfiles = set()
            normalfiles = set()
            if not pats:
                pats = ui.configlist(lfutil.longname, 'patterns', default=[])
            if pats:
                matcher = matchmod.match(rsrc.root, '', list(pats))
            else:
                matcher = None

            lfiletohash = {}
            for ctx in ctxs:
                ui.progress(_('converting revisions'), ctx.rev(),
                    unit=_('revisions'), total=rsrc['tip'].rev())
                _lfconvert_addchangeset(rsrc, rdst, ctx, revmap,
                    lfiles, normalfiles, matcher, size, lfiletohash)
            ui.progress(_('converting revisions'), None)

            if rdst.wvfs.exists(lfutil.shortname):
                rdst.wvfs.rmtree(lfutil.shortname)

            for f in lfiletohash.keys():
                if rdst.wvfs.isfile(f):
                    rdst.wvfs.unlink(f)
                try:
                    rdst.wvfs.removedirs(rdst.wvfs.dirname(f))
                except OSError:
                    pass

            # If there were any files converted to largefiles, add largefiles
            # to the destination repository's requirements.
            if lfiles:
                rdst.requirements.add('largefiles')
                rdst._writerequirements()
        else:
            class lfsource(filemap.filemap_source):
                def __init__(self, ui, source):
                    super(lfsource, self).__init__(ui, source, None)
                    self.filemapper.rename[lfutil.shortname] = '.'

                def getfile(self, name, rev):
                    realname, realrev = rev
                    f = super(lfsource, self).getfile(name, rev)

                    if (not realname.startswith(lfutil.shortnameslash)
                            or f[0] is None):
                        return f

                    # Substitute in the largefile data for the hash
                    hash = f[0].strip()
                    path = lfutil.findfile(rsrc, hash)

                    if path is None:
                        raise error.Abort(_("missing largefile for '%s' in %s")
                                          % (realname, realrev))
                    return util.readfile(path), f[1]

            class converter(convcmd.converter):
                def __init__(self, ui, source, dest, revmapfile, opts):
                    src = lfsource(ui, source)

                    super(converter, self).__init__(ui, src, dest, revmapfile,
                                                    opts)

            found, missing = downloadlfiles(ui, rsrc)
            if missing != 0:
                raise error.Abort(_("all largefiles must be present locally"))

            orig = convcmd.converter
            convcmd.converter = converter

            try:
                convcmd.convert(ui, src, dest)
            finally:
                convcmd.converter = orig
        success = True
    finally:
        if tolfile:
            rdst.dirstate.clear()
            release(dstlock, dstwlock)
        if not success:
            # we failed, remove the new directory
            shutil.rmtree(rdst.root)

def _lfconvert_addchangeset(rsrc, rdst, ctx, revmap, lfiles, normalfiles,
        matcher, size, lfiletohash):
    # Convert src parents to dst parents
    parents = _convertparents(ctx, revmap)

    # Generate list of changed files
    files = _getchangedfiles(ctx, parents)

    dstfiles = []
    for f in files:
        if f not in lfiles and f not in normalfiles:
            islfile = _islfile(f, ctx, matcher, size)
            # If this file was renamed or copied then copy
            # the largefile-ness of its predecessor
            if f in ctx.manifest():
                fctx = ctx.filectx(f)
                renamed = fctx.renamed()
                renamedlfile = renamed and renamed[0] in lfiles
                islfile |= renamedlfile
                if 'l' in fctx.flags():
                    if renamedlfile:
                        raise error.Abort(
                            _('renamed/copied largefile %s becomes symlink')
                            % f)
                    islfile = False
            if islfile:
                lfiles.add(f)
            else:
                normalfiles.add(f)

        if f in lfiles:
            fstandin = lfutil.standin(f)
            dstfiles.append(fstandin)
            # largefile in manifest if it has not been removed/renamed
            if f in ctx.manifest():
                fctx = ctx.filectx(f)
                if 'l' in fctx.flags():
                    renamed = fctx.renamed()
                    if renamed and renamed[0] in lfiles:
                        raise error.Abort(_('largefile %s becomes symlink') % f)

                # largefile was modified, update standins
                m = hashlib.sha1('')
                m.update(ctx[f].data())
                hash = m.hexdigest()
                if f not in lfiletohash or lfiletohash[f] != hash:
                    rdst.wwrite(f, ctx[f].data(), ctx[f].flags())
                    executable = 'x' in ctx[f].flags()
                    lfutil.writestandin(rdst, fstandin, hash,
                        executable)
                    lfiletohash[f] = hash
        else:
            # normal file
            dstfiles.append(f)

    def getfilectx(repo, memctx, f):
        srcfname = lfutil.splitstandin(f)
        if srcfname is not None:
            # if the file isn't in the manifest then it was removed
            # or renamed, return None to indicate this
            try:
                fctx = ctx.filectx(srcfname)
            except error.LookupError:
                return None
            renamed = fctx.renamed()
            if renamed:
                # standin is always a largefile because largefile-ness
                # doesn't change after rename or copy
                renamed = lfutil.standin(renamed[0])

            return context.memfilectx(repo, f, lfiletohash[srcfname] + '\n',
                                      'l' in fctx.flags(), 'x' in fctx.flags(),
                                      renamed)
        else:
            return _getnormalcontext(repo, ctx, f, revmap)

    # Commit
    _commitcontext(rdst, parents, ctx, dstfiles, getfilectx, revmap)

def _commitcontext(rdst, parents, ctx, dstfiles, getfilectx, revmap):
    mctx = context.memctx(rdst, parents, ctx.description(), dstfiles,
                          getfilectx, ctx.user(), ctx.date(), ctx.extra())
    ret = rdst.commitctx(mctx)
    lfutil.copyalltostore(rdst, ret)
    rdst.setparents(ret)
    revmap[ctx.node()] = rdst.changelog.tip()

# Generate list of changed files
def _getchangedfiles(ctx, parents):
    files = set(ctx.files())
    if node.nullid not in parents:
        mc = ctx.manifest()
        mp1 = ctx.parents()[0].manifest()
        mp2 = ctx.parents()[1].manifest()
        files |= (set(mp1) | set(mp2)) - set(mc)
        for f in mc:
            if mc[f] != mp1.get(f, None) or mc[f] != mp2.get(f, None):
                files.add(f)
    return files

# Convert src parents to dst parents
def _convertparents(ctx, revmap):
    parents = []
    for p in ctx.parents():
        parents.append(revmap[p.node()])
    while len(parents) < 2:
        parents.append(node.nullid)
    return parents

# Get memfilectx for a normal file
def _getnormalcontext(repo, ctx, f, revmap):
    try:
        fctx = ctx.filectx(f)
    except error.LookupError:
        return None
    renamed = fctx.renamed()
    if renamed:
        renamed = renamed[0]

    data = fctx.data()
    if f == '.hgtags':
        data = _converttags (repo.ui, revmap, data)
    return context.memfilectx(repo, f, data, 'l' in fctx.flags(),
                              'x' in fctx.flags(), renamed)

# Remap tag data using a revision map
def _converttags(ui, revmap, data):
    newdata = []
    for line in data.splitlines():
        try:
            id, name = line.split(' ', 1)
        except ValueError:
            ui.warn(_('skipping incorrectly formatted tag %s\n')
                % line)
            continue
        try:
            newid = node.bin(id)
        except TypeError:
            ui.warn(_('skipping incorrectly formatted id %s\n')
                % id)
            continue
        try:
            newdata.append('%s %s\n' % (node.hex(revmap[newid]),
                name))
        except KeyError:
            ui.warn(_('no mapping for id %s\n') % id)
            continue
    return ''.join(newdata)

def _islfile(file, ctx, matcher, size):
    '''Return true if file should be considered a largefile, i.e.
    matcher matches it or it is larger than size.'''
    # never store special .hg* files as largefiles
    if file == '.hgtags' or file == '.hgignore' or file == '.hgsigs':
        return False
    if matcher and matcher(file):
        return True
    try:
        return ctx.filectx(file).size() >= size * 1024 * 1024
    except error.LookupError:
        return False

def uploadlfiles(ui, rsrc, rdst, files):
    '''upload largefiles to the central store'''

    if not files:
        return

    store = storefactory.openstore(rsrc, rdst, put=True)

    at = 0
    ui.debug("sending statlfile command for %d largefiles\n" % len(files))
    retval = store.exists(files)
    files = filter(lambda h: not retval[h], files)
    ui.debug("%d largefiles need to be uploaded\n" % len(files))

    for hash in files:
        ui.progress(_('uploading largefiles'), at, unit=_('files'),
                    total=len(files))
        source = lfutil.findfile(rsrc, hash)
        if not source:
            raise error.Abort(_('largefile %s missing from store'
                               ' (needs to be uploaded)') % hash)
        # XXX check for errors here
        store.put(source, hash)
        at += 1
    ui.progress(_('uploading largefiles'), None)

def verifylfiles(ui, repo, all=False, contents=False):
    '''Verify that every largefile revision in the current changeset
    exists in the central store.  With --contents, also verify that
    the contents of each local largefile file revision are correct (SHA-1 hash
    matches the revision ID).  With --all, check every changeset in
    this repository.'''
    if all:
        revs = repo.revs('all()')
    else:
        revs = ['.']

    store = storefactory.openstore(repo)
    return store.verify(revs, contents=contents)

def cachelfiles(ui, repo, node, filelist=None):
    '''cachelfiles ensures that all largefiles needed by the specified revision
    are present in the repository's largefile cache.

    returns a tuple (cached, missing).  cached is the list of files downloaded
    by this operation; missing is the list of files that were needed but could
    not be found.'''
    lfiles = lfutil.listlfiles(repo, node)
    if filelist:
        lfiles = set(lfiles) & set(filelist)
    toget = []

    ctx = repo[node]
    for lfile in lfiles:
        try:
            expectedhash = lfutil.readasstandin(ctx[lfutil.standin(lfile)])
        except IOError as err:
            if err.errno == errno.ENOENT:
                continue # node must be None and standin wasn't found in wctx
            raise
        if not lfutil.findfile(repo, expectedhash):
            toget.append((lfile, expectedhash))

    if toget:
        store = storefactory.openstore(repo)
        ret = store.get(toget)
        return ret

    return ([], [])

def downloadlfiles(ui, repo, rev=None):
    matchfn = scmutil.match(repo[None],
                            [repo.wjoin(lfutil.shortname)], {})
    def prepare(ctx, fns):
        pass
    totalsuccess = 0
    totalmissing = 0
    if rev != []: # walkchangerevs on empty list would return all revs
        for ctx in cmdutil.walkchangerevs(repo, matchfn, {'rev' : rev},
                                          prepare):
            success, missing = cachelfiles(ui, repo, ctx.node())
            totalsuccess += len(success)
            totalmissing += len(missing)
    ui.status(_("%d additional largefiles cached\n") % totalsuccess)
    if totalmissing > 0:
        ui.status(_("%d largefiles failed to download\n") % totalmissing)
    return totalsuccess, totalmissing

def updatelfiles(ui, repo, filelist=None, printmessage=None,
                 normallookup=False):
    '''Update largefiles according to standins in the working directory

    If ``printmessage`` is other than ``None``, it means "print (or
    ignore, for false) message forcibly".
    '''
    statuswriter = lfutil.getstatuswriter(ui, repo, printmessage)
    with repo.wlock():
        lfdirstate = lfutil.openlfdirstate(ui, repo)
        lfiles = set(lfutil.listlfiles(repo)) | set(lfdirstate)

        if filelist is not None:
            filelist = set(filelist)
            lfiles = [f for f in lfiles if f in filelist]

        update = {}
        updated, removed = 0, 0
        wvfs = repo.wvfs
        wctx = repo[None]
        for lfile in lfiles:
            rellfile = lfile
            rellfileorig = os.path.relpath(
                scmutil.origpath(ui, repo, wvfs.join(rellfile)),
                start=repo.root)
            relstandin = lfutil.standin(lfile)
            relstandinorig = os.path.relpath(
                scmutil.origpath(ui, repo, wvfs.join(relstandin)),
                start=repo.root)
            if wvfs.exists(relstandin):
                if (wvfs.exists(relstandinorig) and
                    wvfs.exists(rellfile)):
                    shutil.copyfile(wvfs.join(rellfile),
                                    wvfs.join(rellfileorig))
                    wvfs.unlinkpath(relstandinorig)
                expecthash = lfutil.readasstandin(wctx[relstandin])
                if expecthash != '':
                    if lfile not in wctx: # not switched to normal file
                        wvfs.unlinkpath(rellfile, ignoremissing=True)
                    # use normallookup() to allocate an entry in largefiles
                    # dirstate to prevent lfilesrepo.status() from reporting
                    # missing files as removed.
                    lfdirstate.normallookup(lfile)
                    update[lfile] = expecthash
            else:
                # Remove lfiles for which the standin is deleted, unless the
                # lfile is added to the repository again. This happens when a
                # largefile is converted back to a normal file: the standin
                # disappears, but a new (normal) file appears as the lfile.
                if (wvfs.exists(rellfile) and
                    repo.dirstate.normalize(lfile) not in wctx):
                    wvfs.unlinkpath(rellfile)
                    removed += 1

        # largefile processing might be slow and be interrupted - be prepared
        lfdirstate.write()

        if lfiles:
            statuswriter(_('getting changed largefiles\n'))
            cachelfiles(ui, repo, None, lfiles)

        for lfile in lfiles:
            update1 = 0

            expecthash = update.get(lfile)
            if expecthash:
                if not lfutil.copyfromcache(repo, expecthash, lfile):
                    # failed ... but already removed and set to normallookup
                    continue
                # Synchronize largefile dirstate to the last modified
                # time of the file
                lfdirstate.normal(lfile)
                update1 = 1

            # copy the exec mode of largefile standin from the repository's
            # dirstate to its state in the lfdirstate.
            rellfile = lfile
            relstandin = lfutil.standin(lfile)
            if wvfs.exists(relstandin):
                # exec is decided by the users permissions using mask 0o100
                standinexec = wvfs.stat(relstandin).st_mode & 0o100
                st = wvfs.stat(rellfile)
                mode = st.st_mode
                if standinexec != mode & 0o100:
                    # first remove all X bits, then shift all R bits to X
                    mode &= ~0o111
                    if standinexec:
                        mode |= (mode >> 2) & 0o111 & ~util.umask
                    wvfs.chmod(rellfile, mode)
                    update1 = 1

            updated += update1

            lfutil.synclfdirstate(repo, lfdirstate, lfile, normallookup)

        lfdirstate.write()
        if lfiles:
            statuswriter(_('%d largefiles updated, %d removed\n') % (updated,
                removed))

@command('lfpull',
    [('r', 'rev', [], _('pull largefiles for these revisions'))
    ] + commands.remoteopts,
    _('-r REV... [-e CMD] [--remotecmd CMD] [SOURCE]'))
def lfpull(ui, repo, source="default", **opts):
    """pull largefiles for the specified revisions from the specified source

    Pull largefiles that are referenced from local changesets but missing
    locally, pulling from a remote repository to the local cache.

    If SOURCE is omitted, the 'default' path will be used.
    See :hg:`help urls` for more information.

    .. container:: verbose

      Some examples:

      - pull largefiles for all branch heads::

          hg lfpull -r "head() and not closed()"

      - pull largefiles on the default branch::

          hg lfpull -r "branch(default)"
    """
    repo.lfpullsource = source

    revs = opts.get('rev', [])
    if not revs:
        raise error.Abort(_('no revisions specified'))
    revs = scmutil.revrange(repo, revs)

    numcached = 0
    for rev in revs:
        ui.note(_('pulling largefiles for revision %s\n') % rev)
        (cached, missing) = cachelfiles(ui, repo, rev)
        numcached += len(cached)
    ui.status(_("%d largefiles cached\n") % numcached)