view mercurial/branchmap.py @ 26623:5a95fe44121d

clonebundles: support for seeding clones from pre-generated bundles

Cloning can be an expensive operation for servers because the server
generates a bundle from existing repository data at request time. For
a large repository like mozilla-central, this consumes 4+ minutes
of CPU time on the server. It also results in significant network
utilization. Multiplied by hundreds or even thousands of clients and
the ensuing load can result in difficulties scaling the Mercurial server.

Despite generation of bundles being deterministic until the next
changeset is added, the generation of bundles to service a clone request
is not cached. Each clone thus performs redundant work. This is
wasteful.

This patch introduces the "clonebundles" extension and related
client-side functionality to help alleviate this deficiency. The
client-side feature is behind an experimental flag and is not enabled by
default.

It works as follows:

1) Server operator generates a bundle and makes it available on a
   server (likely HTTP).
2) Server operator defines the URL of a bundle file in a
   .hg/clonebundles.manifest file.
3) Client `hg clone`ing sees the server is advertising bundle URLs.
4) Client fetches and applies the advertised bundle.
5) Client performs equivalent of `hg pull` to fetch changes made since
   the bundle was created.

Essentially, the server performs the expensive work of generating a
bundle once and all subsequent clones fetch a static file from
somewhere. Scaling static file serving is a much more manageable
problem than scaling a Python application like Mercurial. Assuming your
repository grows less than 1% per day, the end result is 99+% of CPU
and network load from clones is eliminated, allowing Mercurial servers
to scale more easily. Serving static files also means data can be
transferred to clients as fast as they can consume it, rather than as
fast as servers can generate it. This makes clones faster.

Mozilla has implemented similar functionality of this patch on
hg.mozilla.org using a custom extension. We are hosting bundle files in
Amazon S3 and CloudFront (a CDN) and have successfully offloaded
>1 TB/day in data transfer from hg.mozilla.org, freeing up significant
bandwidth and CPU resources. The positive impact has been stellar and
I believe it has proved its value to be included in Mercurial core. I
feel it is important for the client-side support to be enabled in core
by default because it means that clients will get faster, more reliable
clones and will enable server operators to reduce load without
requiring any client-side configuration changes (assuming clients are
up to date, of course).

The scope of this feature is narrowly and specifically tailored to
cloning, despite "serve pulls from pre-generated bundles" being a valid
and useful feature. I would eventually like for Mercurial servers to
support transferring *all* repository data via statically hosted files.
You could imagine a server that siphons all pushed data to bundle files
and instructs clients to apply a stream of bundles to reconstruct all
repository data. This feature, while useful and powerful, is
significantly more work to implement because it requires the server
component have awareness of discovery and a mapping of which changesets
are in which files. Full, clone bundles, by contrast, are much simpler.

The wire protocol command is named "clonebundles" instead of something
more generic like "staticbundles" to leave the door open for a new, more
powerful and more generic server-side component with minimal backwards
compatibility implications. The name "bundleclone" is used by Mozilla's
extension and would cause problems since there are subtle differences
in Mozilla's extension.

Mozilla's experience with this idea has taught us that some form of
"content negotiation" is required. Not all clients will support all
bundle formats or even URLs (advanced TLS requirements, etc). To ensure
the highest uptake possible, a server needs to advertise multiple
versions of bundles and clients need to be able to choose the most
appropriate from that list one. The "attributes" in each
server-advertised entry facilitate this filtering and sorting. Their
use will become apparent in subsequent patches.

Initial inspiration and credit for the idea of cloning from static files
belongs to Augie Fackler and his "lookaside clone" extension proof of
concept.

# branchmap.py - logic to computes, maintain and stores branchmap for local repo
#
# Copyright 2005-2007 Matt Mackall <[e-mail]>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.

from __future__ import absolute_import

import array
import struct
import time

from .node import (
    bin,
    hex,
    nullid,
    nullrev,
)
from . import (
    encoding,
    error,
    scmutil,
)

array = array.array
calcsize = struct.calcsize
pack = struct.pack
unpack = struct.unpack

def _filename(repo):
    """name of a branchcache file for a given repo or repoview"""
    filename = "cache/branch2"
    if repo.filtername:
        filename = '%s-%s' % (filename, repo.filtername)
    return filename

def read(repo):
    try:
        f = repo.vfs(_filename(repo))
        lines = f.read().split('\n')
        f.close()
    except (IOError, OSError):
        return None

    try:
        cachekey = lines.pop(0).split(" ", 2)
        last, lrev = cachekey[:2]
        last, lrev = bin(last), int(lrev)
        filteredhash = None
        if len(cachekey) > 2:
            filteredhash = bin(cachekey[2])
        partial = branchcache(tipnode=last, tiprev=lrev,
                              filteredhash=filteredhash)
        if not partial.validfor(repo):
            # invalidate the cache
            raise ValueError('tip differs')
        for l in lines:
            if not l:
                continue
            node, state, label = l.split(" ", 2)
            if state not in 'oc':
                raise ValueError('invalid branch state')
            label = encoding.tolocal(label.strip())
            if not node in repo:
                raise ValueError('node %s does not exist' % node)
            node = bin(node)
            partial.setdefault(label, []).append(node)
            if state == 'c':
                partial._closednodes.add(node)
    except KeyboardInterrupt:
        raise
    except Exception as inst:
        if repo.ui.debugflag:
            msg = 'invalid branchheads cache'
            if repo.filtername is not None:
                msg += ' (%s)' % repo.filtername
            msg += ': %s\n'
            repo.ui.debug(msg % inst)
        partial = None
    return partial

### Nearest subset relation
# Nearest subset of filter X is a filter Y so that:
# * Y is included in X,
# * X - Y is as small as possible.
# This create and ordering used for branchmap purpose.
# the ordering may be partial
subsettable = {None: 'visible',
               'visible': 'served',
               'served': 'immutable',
               'immutable': 'base'}

def updatecache(repo):
    cl = repo.changelog
    filtername = repo.filtername
    partial = repo._branchcaches.get(filtername)

    revs = []
    if partial is None or not partial.validfor(repo):
        partial = read(repo)
        if partial is None:
            subsetname = subsettable.get(filtername)
            if subsetname is None:
                partial = branchcache()
            else:
                subset = repo.filtered(subsetname)
                partial = subset.branchmap().copy()
                extrarevs = subset.changelog.filteredrevs - cl.filteredrevs
                revs.extend(r for  r in extrarevs if r <= partial.tiprev)
    revs.extend(cl.revs(start=partial.tiprev + 1))
    if revs:
        partial.update(repo, revs)
        partial.write(repo)

    assert partial.validfor(repo), filtername
    repo._branchcaches[repo.filtername] = partial

def replacecache(repo, bm):
    """Replace the branchmap cache for a repo with a branch mapping.

    This is likely only called during clone with a branch map from a remote.
    """
    rbheads = []
    closed = []
    for bheads in bm.itervalues():
        rbheads.extend(bheads)
        for h in bheads:
            r = repo.changelog.rev(h)
            b, c = repo.changelog.branchinfo(r)
            if c:
                closed.append(h)

    if rbheads:
        rtiprev = max((int(repo.changelog.rev(node))
                for node in rbheads))
        cache = branchcache(bm,
                            repo[rtiprev].node(),
                            rtiprev,
                            closednodes=closed)

        # Try to stick it as low as possible
        # filter above served are unlikely to be fetch from a clone
        for candidate in ('base', 'immutable', 'served'):
            rview = repo.filtered(candidate)
            if cache.validfor(rview):
                repo._branchcaches[candidate] = cache
                cache.write(rview)
                break

class branchcache(dict):
    """A dict like object that hold branches heads cache.

    This cache is used to avoid costly computations to determine all the
    branch heads of a repo.

    The cache is serialized on disk in the following format:

    <tip hex node> <tip rev number> [optional filtered repo hex hash]
    <branch head hex node> <open/closed state> <branch name>
    <branch head hex node> <open/closed state> <branch name>
    ...

    The first line is used to check if the cache is still valid. If the
    branch cache is for a filtered repo view, an optional third hash is
    included that hashes the hashes of all filtered revisions.

    The open/closed state is represented by a single letter 'o' or 'c'.
    This field can be used to avoid changelog reads when determining if a
    branch head closes a branch or not.
    """

    def __init__(self, entries=(), tipnode=nullid, tiprev=nullrev,
                 filteredhash=None, closednodes=None):
        super(branchcache, self).__init__(entries)
        self.tipnode = tipnode
        self.tiprev = tiprev
        self.filteredhash = filteredhash
        # closednodes is a set of nodes that close their branch. If the branch
        # cache has been updated, it may contain nodes that are no longer
        # heads.
        if closednodes is None:
            self._closednodes = set()
        else:
            self._closednodes = closednodes

    def validfor(self, repo):
        """Is the cache content valid regarding a repo

        - False when cached tipnode is unknown or if we detect a strip.
        - True when cache is up to date or a subset of current repo."""
        try:
            return ((self.tipnode == repo.changelog.node(self.tiprev))
                    and (self.filteredhash == \
                         scmutil.filteredhash(repo, self.tiprev)))
        except IndexError:
            return False

    def _branchtip(self, heads):
        '''Return tuple with last open head in heads and false,
        otherwise return last closed head and true.'''
        tip = heads[-1]
        closed = True
        for h in reversed(heads):
            if h not in self._closednodes:
                tip = h
                closed = False
                break
        return tip, closed

    def branchtip(self, branch):
        '''Return the tipmost open head on branch head, otherwise return the
        tipmost closed head on branch.
        Raise KeyError for unknown branch.'''
        return self._branchtip(self[branch])[0]

    def branchheads(self, branch, closed=False):
        heads = self[branch]
        if not closed:
            heads = [h for h in heads if h not in self._closednodes]
        return heads

    def iterbranches(self):
        for bn, heads in self.iteritems():
            yield (bn, heads) + self._branchtip(heads)

    def copy(self):
        """return an deep copy of the branchcache object"""
        return branchcache(self, self.tipnode, self.tiprev, self.filteredhash,
                           self._closednodes)

    def write(self, repo):
        try:
            f = repo.vfs(_filename(repo), "w", atomictemp=True)
            cachekey = [hex(self.tipnode), str(self.tiprev)]
            if self.filteredhash is not None:
                cachekey.append(hex(self.filteredhash))
            f.write(" ".join(cachekey) + '\n')
            nodecount = 0
            for label, nodes in sorted(self.iteritems()):
                for node in nodes:
                    nodecount += 1
                    if node in self._closednodes:
                        state = 'c'
                    else:
                        state = 'o'
                    f.write("%s %s %s\n" % (hex(node), state,
                                            encoding.fromlocal(label)))
            f.close()
            repo.ui.log('branchcache',
                        'wrote %s branch cache with %d labels and %d nodes\n',
                        repo.filtername, len(self), nodecount)
        except (IOError, OSError, error.Abort) as inst:
            repo.ui.debug("couldn't write branch cache: %s\n" % inst)
            # Abort may be raise by read only opener
            pass

    def update(self, repo, revgen):
        """Given a branchhead cache, self, that may have extra nodes or be
        missing heads, and a generator of nodes that are strictly a superset of
        heads missing, this function updates self to be correct.
        """
        starttime = time.time()
        cl = repo.changelog
        # collect new branch entries
        newbranches = {}
        getbranchinfo = repo.revbranchcache().branchinfo
        for r in revgen:
            branch, closesbranch = getbranchinfo(r)
            newbranches.setdefault(branch, []).append(r)
            if closesbranch:
                self._closednodes.add(cl.node(r))

        # fetch current topological heads to speed up filtering
        topoheads = set(cl.headrevs())

        # if older branchheads are reachable from new ones, they aren't
        # really branchheads. Note checking parents is insufficient:
        # 1 (branch a) -> 2 (branch b) -> 3 (branch a)
        for branch, newheadrevs in newbranches.iteritems():
            bheads = self.setdefault(branch, [])
            bheadset = set(cl.rev(node) for node in bheads)

            # This have been tested True on all internal usage of this function.
            # run it again in case of doubt
            # assert not (set(bheadrevs) & set(newheadrevs))
            newheadrevs.sort()
            bheadset.update(newheadrevs)

            # This prunes out two kinds of heads - heads that are superseded by
            # a head in newheadrevs, and newheadrevs that are not heads because
            # an existing head is their descendant.
            uncertain = bheadset - topoheads
            if uncertain:
                floorrev = min(uncertain)
                ancestors = set(cl.ancestors(newheadrevs, floorrev))
                bheadset -= ancestors
            bheadrevs = sorted(bheadset)
            self[branch] = [cl.node(rev) for rev in bheadrevs]
            tiprev = bheadrevs[-1]
            if tiprev > self.tiprev:
                self.tipnode = cl.node(tiprev)
                self.tiprev = tiprev

        if not self.validfor(repo):
            # cache key are not valid anymore
            self.tipnode = nullid
            self.tiprev = nullrev
            for heads in self.values():
                tiprev = max(cl.rev(node) for node in heads)
                if tiprev > self.tiprev:
                    self.tipnode = cl.node(tiprev)
                    self.tiprev = tiprev
        self.filteredhash = scmutil.filteredhash(repo, self.tiprev)

        duration = time.time() - starttime
        repo.ui.log('branchcache', 'updated %s branch cache in %.4f seconds\n',
                    repo.filtername, duration)

# Revision branch info cache

_rbcversion = '-v1'
_rbcnames = 'cache/rbc-names' + _rbcversion
_rbcrevs = 'cache/rbc-revs' + _rbcversion
# [4 byte hash prefix][4 byte branch name number with sign bit indicating open]
_rbcrecfmt = '>4sI'
_rbcrecsize = calcsize(_rbcrecfmt)
_rbcnodelen = 4
_rbcbranchidxmask = 0x7fffffff
_rbccloseflag = 0x80000000

class revbranchcache(object):
    """Persistent cache, mapping from revision number to branch name and close.
    This is a low level cache, independent of filtering.

    Branch names are stored in rbc-names in internal encoding separated by 0.
    rbc-names is append-only, and each branch name is only stored once and will
    thus have a unique index.

    The branch info for each revision is stored in rbc-revs as constant size
    records. The whole file is read into memory, but it is only 'parsed' on
    demand. The file is usually append-only but will be truncated if repo
    modification is detected.
    The record for each revision contains the first 4 bytes of the
    corresponding node hash, and the record is only used if it still matches.
    Even a completely trashed rbc-revs fill thus still give the right result
    while converging towards full recovery ... assuming no incorrectly matching
    node hashes.
    The record also contains 4 bytes where 31 bits contains the index of the
    branch and the last bit indicate that it is a branch close commit.
    The usage pattern for rbc-revs is thus somewhat similar to 00changelog.i
    and will grow with it but be 1/8th of its size.
    """

    def __init__(self, repo, readonly=True):
        assert repo.filtername is None
        self._repo = repo
        self._names = [] # branch names in local encoding with static index
        self._rbcrevs = array('c') # structs of type _rbcrecfmt
        self._rbcsnameslen = 0
        try:
            bndata = repo.vfs.read(_rbcnames)
            self._rbcsnameslen = len(bndata) # for verification before writing
            self._names = [encoding.tolocal(bn) for bn in bndata.split('\0')]
        except (IOError, OSError) as inst:
            if readonly:
                # don't try to use cache - fall back to the slow path
                self.branchinfo = self._branchinfo

        if self._names:
            try:
                data = repo.vfs.read(_rbcrevs)
                self._rbcrevs.fromstring(data)
            except (IOError, OSError) as inst:
                repo.ui.debug("couldn't read revision branch cache: %s\n" %
                              inst)
        # remember number of good records on disk
        self._rbcrevslen = min(len(self._rbcrevs) // _rbcrecsize,
                               len(repo.changelog))
        if self._rbcrevslen == 0:
            self._names = []
        self._rbcnamescount = len(self._names) # number of good names on disk
        self._namesreverse = dict((b, r) for r, b in enumerate(self._names))

    def branchinfo(self, rev):
        """Return branch name and close flag for rev, using and updating
        persistent cache."""
        changelog = self._repo.changelog
        rbcrevidx = rev * _rbcrecsize

        # avoid negative index, changelog.read(nullrev) is fast without cache
        if rev == nullrev:
            return changelog.branchinfo(rev)

        # if requested rev is missing, add and populate all missing revs
        if len(self._rbcrevs) < rbcrevidx + _rbcrecsize:
            self._rbcrevs.extend('\0' * (len(changelog) * _rbcrecsize -
                                         len(self._rbcrevs)))

        # fast path: extract data from cache, use it if node is matching
        reponode = changelog.node(rev)[:_rbcnodelen]
        cachenode, branchidx = unpack(
            _rbcrecfmt, buffer(self._rbcrevs, rbcrevidx, _rbcrecsize))
        close = bool(branchidx & _rbccloseflag)
        if close:
            branchidx &= _rbcbranchidxmask
        if cachenode == '\0\0\0\0':
            pass
        elif cachenode == reponode:
            return self._names[branchidx], close
        else:
            # rev/node map has changed, invalidate the cache from here up
            truncate = rbcrevidx + _rbcrecsize
            del self._rbcrevs[truncate:]
            self._rbcrevslen = min(self._rbcrevslen, truncate)

        # fall back to slow path and make sure it will be written to disk
        return self._branchinfo(rev)

    def _branchinfo(self, rev):
        """Retrieve branch info from changelog and update _rbcrevs"""
        changelog = self._repo.changelog
        b, close = changelog.branchinfo(rev)
        if b in self._namesreverse:
            branchidx = self._namesreverse[b]
        else:
            branchidx = len(self._names)
            self._names.append(b)
            self._namesreverse[b] = branchidx
        reponode = changelog.node(rev)
        if close:
            branchidx |= _rbccloseflag
        self._setcachedata(rev, reponode, branchidx)
        return b, close

    def _setcachedata(self, rev, node, branchidx):
        """Writes the node's branch data to the in-memory cache data."""
        rbcrevidx = rev * _rbcrecsize
        rec = array('c')
        rec.fromstring(pack(_rbcrecfmt, node, branchidx))
        self._rbcrevs[rbcrevidx:rbcrevidx + _rbcrecsize] = rec
        self._rbcrevslen = min(self._rbcrevslen, rev)

        tr = self._repo.currenttransaction()
        if tr:
            tr.addfinalize('write-revbranchcache', self.write)

    def write(self, tr=None):
        """Save branch cache if it is dirty."""
        repo = self._repo
        if self._rbcnamescount < len(self._names):
            try:
                if self._rbcnamescount != 0:
                    f = repo.vfs.open(_rbcnames, 'ab')
                    if f.tell() == self._rbcsnameslen:
                        f.write('\0')
                    else:
                        f.close()
                        repo.ui.debug("%s changed - rewriting it\n" % _rbcnames)
                        self._rbcnamescount = 0
                        self._rbcrevslen = 0
                if self._rbcnamescount == 0:
                    f = repo.vfs.open(_rbcnames, 'wb')
                f.write('\0'.join(encoding.fromlocal(b)
                                  for b in self._names[self._rbcnamescount:]))
                self._rbcsnameslen = f.tell()
                f.close()
            except (IOError, OSError, error.Abort) as inst:
                repo.ui.debug("couldn't write revision branch cache names: "
                              "%s\n" % inst)
                return
            self._rbcnamescount = len(self._names)

        start = self._rbcrevslen * _rbcrecsize
        if start != len(self._rbcrevs):
            revs = min(len(repo.changelog), len(self._rbcrevs) // _rbcrecsize)
            try:
                f = repo.vfs.open(_rbcrevs, 'ab')
                if f.tell() != start:
                    repo.ui.debug("truncating %s to %s\n" % (_rbcrevs, start))
                    f.seek(start)
                    f.truncate()
                end = revs * _rbcrecsize
                f.write(self._rbcrevs[start:end])
                f.close()
            except (IOError, OSError, error.Abort) as inst:
                repo.ui.debug("couldn't write revision branch cache: %s\n" %
                              inst)
                return
            self._rbcrevslen = revs