wireproto: add streams to frame-based protocol
Previously, the frame-based protocol was just a series of frames,
with each frame associated with a request ID.
In order to scale the protocol, we'll want to enable the use of
compression. While it is possible to enable compression at the
socket/pipe level, this has its disadvantages. The big one is it
undermines the point of frames being standalone, atomic units that
can be read and written: if you add compression above the framing
protocol, you are back to having a stream-based protocol as opposed
to something frame-based.
So in order to preserve frames, compression needs to occur at
the frame payload level.
Compressing each frame's payload individually will limit compression
ratios because the window size of the compressor will be limited
by the max frame size, which is 32-64kb as currently defined. It
will also add CPU overhead, as it is more efficient for compressors
to operate on fewer, larger blocks of data than more, smaller blocks.
So compressing each frame independently is out.
This means we need to compress each frame's payload as if it is part
of a larger stream.
The simplest approach is to have 1 stream per connection. This
could certainly work. However, it has disadvantages (documented below).
We could also have 1 stream per RPC/command invocation. (This is the
model HTTP/2 goes with.) This also has disadvantages.
The main disadvantage to one global stream is that it has the very
real potential to create CPU bottlenecks doing compression. Networks
are only getting faster and the performance of single CPU cores has
been relatively flat. Newer compression formats like zstandard offer
better CPU cycle efficiency than predecessors like zlib. But it still
all too common to saturate your CPU with compression overhead long
before you saturate the network pipe.
The main disadvantage with streams per request is that you can't
reap the benefits of the compression context for multiple requests.
For example, if you send 1000 RPC requests (or HTTP/2 requests for
that matter), the response to each would have its own compression
context. The overall size of the raw responses would be larger because
compression contexts wouldn't be able to reference data from another
request or response.
The approach for streams as implemented in this commit is to support
N streams per connection and for streams to potentially span requests
and responses. As explained by the added internals docs, this
facilitates servers and clients delegating independent streams and
compression to independent threads / CPU cores. This helps alleviate
the CPU bottleneck of compression. This design also allows compression
contexts to be reused across requests/responses. This can result in
improved compression ratios and less overhead for compressors and
decompressors having to build new contexts.
Another feature that was defined was the ability for individual frames
within a stream to declare whether that individual frame's payload
uses the content encoding (read: compression) defined by the stream.
The idea here is that some servers may serve data from a combination
of caches and dynamic resolution. Data coming from caches may be
pre-compressed. We want to facilitate servers being able to essentially
stream bytes from caches to the wire with minimal overhead. Being
able to mix and match with frames are compressed within a stream
enables these types of advanced server functionality.
This commit defines the new streams mechanism. Basic code for
supporting streams in frames has been added. But that code is
seriously lacking and doesn't fully conform to the defined protocol.
For example, we don't close any streams. And support for content
encoding within streams is not yet implemented. The change was
rather invasive and I didn't think it would be reasonable to implement
the entire feature in a single commit.
For the record, I would have loved to reuse an existing multiplexing
protocol to build the new wire protocol on top of. However, I couldn't
find a protocol that offers the performance and scaling characteristics
that I desired. Namely, it should support multiple compression
contexts to facilitate scaling out to multiple CPU cores and
compression contexts should be able to live longer than single RPC
requests. HTTP/2 *almost* fits the bill. But the semantics of HTTP
message exchange state that streams can only live for a single
request-response. We /could/ tunnel on top of HTTP/2 streams and
frames with HEADER and DATA frames. But there's no guarantee that
HTTP/2 libraries and proxies would allow us to use HTTP/2 streams
and frames without the HTTP message exchange semantics defined in
RFC 7540 Section 8. Other RPC protocols like gRPC tunnel are built
on top of HTTP/2 and thus preserve its semantics of stream per
RPC invocation. Even QUIC does this. We could attempt to invent a
higher-level stream that spans HTTP/2 streams. But this would be
violating HTTP/2 because there is no guarantee that HTTP/2 streams
are routed to the same server. The best we can do - which is what
this protocol does - is shoehorn all request and response data into
a single HTTP message and create streams within. At that point, we've
defined a Content-Type in HTTP parlance. It just so happens our
media type can also work as a standalone, stream-based protocol,
without leaning on HTTP or similar protocol.
Differential Revision: https://phab.mercurial-scm.org/D2907
# bundlerepo.py - repository class for viewing uncompressed bundles
#
# Copyright 2006, 2007 Benoit Boissinot <bboissin@gmail.com>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
"""Repository class for viewing uncompressed bundles.
This provides a read-only repository interface to bundles as if they
were part of the actual repository.
"""
from __future__ import absolute_import
import os
import shutil
import tempfile
from .i18n import _
from .node import nullid
from . import (
bundle2,
changegroup,
changelog,
cmdutil,
discovery,
error,
exchange,
filelog,
localrepo,
manifest,
mdiff,
node as nodemod,
pathutil,
phases,
pycompat,
revlog,
util,
vfs as vfsmod,
)
class bundlerevlog(revlog.revlog):
def __init__(self, opener, indexfile, cgunpacker, linkmapper):
# How it works:
# To retrieve a revision, we need to know the offset of the revision in
# the bundle (an unbundle object). We store this offset in the index
# (start). The base of the delta is stored in the base field.
#
# To differentiate a rev in the bundle from a rev in the revlog, we
# check revision against repotiprev.
opener = vfsmod.readonlyvfs(opener)
revlog.revlog.__init__(self, opener, indexfile)
self.bundle = cgunpacker
n = len(self)
self.repotiprev = n - 1
self.bundlerevs = set() # used by 'bundle()' revset expression
for deltadata in cgunpacker.deltaiter():
node, p1, p2, cs, deltabase, delta, flags = deltadata
size = len(delta)
start = cgunpacker.tell() - size
link = linkmapper(cs)
if node in self.nodemap:
# this can happen if two branches make the same change
self.bundlerevs.add(self.nodemap[node])
continue
for p in (p1, p2):
if p not in self.nodemap:
raise error.LookupError(p, self.indexfile,
_("unknown parent"))
if deltabase not in self.nodemap:
raise LookupError(deltabase, self.indexfile,
_('unknown delta base'))
baserev = self.rev(deltabase)
# start, size, full unc. size, base (unused), link, p1, p2, node
e = (revlog.offset_type(start, flags), size, -1, baserev, link,
self.rev(p1), self.rev(p2), node)
self.index.insert(-1, e)
self.nodemap[node] = n
self.bundlerevs.add(n)
n += 1
def _chunk(self, rev, df=None):
# Warning: in case of bundle, the diff is against what we stored as
# delta base, not against rev - 1
# XXX: could use some caching
if rev <= self.repotiprev:
return revlog.revlog._chunk(self, rev)
self.bundle.seek(self.start(rev))
return self.bundle.read(self.length(rev))
def revdiff(self, rev1, rev2):
"""return or calculate a delta between two revisions"""
if rev1 > self.repotiprev and rev2 > self.repotiprev:
# hot path for bundle
revb = self.index[rev2][3]
if revb == rev1:
return self._chunk(rev2)
elif rev1 <= self.repotiprev and rev2 <= self.repotiprev:
return revlog.revlog.revdiff(self, rev1, rev2)
return mdiff.textdiff(self.revision(rev1, raw=True),
self.revision(rev2, raw=True))
def revision(self, nodeorrev, _df=None, raw=False):
"""return an uncompressed revision of a given node or revision
number.
"""
if isinstance(nodeorrev, int):
rev = nodeorrev
node = self.node(rev)
else:
node = nodeorrev
rev = self.rev(node)
if node == nullid:
return ""
rawtext = None
chain = []
iterrev = rev
# reconstruct the revision if it is from a changegroup
while iterrev > self.repotiprev:
if self._cache and self._cache[1] == iterrev:
rawtext = self._cache[2]
break
chain.append(iterrev)
iterrev = self.index[iterrev][3]
if rawtext is None:
rawtext = self.baserevision(iterrev)
while chain:
delta = self._chunk(chain.pop())
rawtext = mdiff.patches(rawtext, [delta])
text, validatehash = self._processflags(rawtext, self.flags(rev),
'read', raw=raw)
if validatehash:
self.checkhash(text, node, rev=rev)
self._cache = (node, rev, rawtext)
return text
def baserevision(self, nodeorrev):
# Revlog subclasses may override 'revision' method to modify format of
# content retrieved from revlog. To use bundlerevlog with such class one
# needs to override 'baserevision' and make more specific call here.
return revlog.revlog.revision(self, nodeorrev, raw=True)
def addrevision(self, *args, **kwargs):
raise NotImplementedError
def addgroup(self, *args, **kwargs):
raise NotImplementedError
def strip(self, *args, **kwargs):
raise NotImplementedError
def checksize(self):
raise NotImplementedError
class bundlechangelog(bundlerevlog, changelog.changelog):
def __init__(self, opener, cgunpacker):
changelog.changelog.__init__(self, opener)
linkmapper = lambda x: x
bundlerevlog.__init__(self, opener, self.indexfile, cgunpacker,
linkmapper)
def baserevision(self, nodeorrev):
# Although changelog doesn't override 'revision' method, some extensions
# may replace this class with another that does. Same story with
# manifest and filelog classes.
# This bypasses filtering on changelog.node() and rev() because we need
# revision text of the bundle base even if it is hidden.
oldfilter = self.filteredrevs
try:
self.filteredrevs = ()
return changelog.changelog.revision(self, nodeorrev, raw=True)
finally:
self.filteredrevs = oldfilter
class bundlemanifest(bundlerevlog, manifest.manifestrevlog):
def __init__(self, opener, cgunpacker, linkmapper, dirlogstarts=None,
dir=''):
manifest.manifestrevlog.__init__(self, opener, dir=dir)
bundlerevlog.__init__(self, opener, self.indexfile, cgunpacker,
linkmapper)
if dirlogstarts is None:
dirlogstarts = {}
if self.bundle.version == "03":
dirlogstarts = _getfilestarts(self.bundle)
self._dirlogstarts = dirlogstarts
self._linkmapper = linkmapper
def baserevision(self, nodeorrev):
node = nodeorrev
if isinstance(node, int):
node = self.node(node)
if node in self.fulltextcache:
result = '%s' % self.fulltextcache[node]
else:
result = manifest.manifestrevlog.revision(self, nodeorrev, raw=True)
return result
def dirlog(self, d):
if d in self._dirlogstarts:
self.bundle.seek(self._dirlogstarts[d])
return bundlemanifest(
self.opener, self.bundle, self._linkmapper,
self._dirlogstarts, dir=d)
return super(bundlemanifest, self).dirlog(d)
class bundlefilelog(bundlerevlog, filelog.filelog):
def __init__(self, opener, path, cgunpacker, linkmapper):
filelog.filelog.__init__(self, opener, path)
bundlerevlog.__init__(self, opener, self.indexfile, cgunpacker,
linkmapper)
def baserevision(self, nodeorrev):
return filelog.filelog.revision(self, nodeorrev, raw=True)
class bundlepeer(localrepo.localpeer):
def canpush(self):
return False
class bundlephasecache(phases.phasecache):
def __init__(self, *args, **kwargs):
super(bundlephasecache, self).__init__(*args, **kwargs)
if util.safehasattr(self, 'opener'):
self.opener = vfsmod.readonlyvfs(self.opener)
def write(self):
raise NotImplementedError
def _write(self, fp):
raise NotImplementedError
def _updateroots(self, phase, newroots, tr):
self.phaseroots[phase] = newroots
self.invalidate()
self.dirty = True
def _getfilestarts(cgunpacker):
filespos = {}
for chunkdata in iter(cgunpacker.filelogheader, {}):
fname = chunkdata['filename']
filespos[fname] = cgunpacker.tell()
for chunk in iter(lambda: cgunpacker.deltachunk(None), {}):
pass
return filespos
class bundlerepository(localrepo.localrepository):
"""A repository instance that is a union of a local repo and a bundle.
Instances represent a read-only repository composed of a local repository
with the contents of a bundle file applied. The repository instance is
conceptually similar to the state of a repository after an
``hg unbundle`` operation. However, the contents of the bundle are never
applied to the actual base repository.
"""
def __init__(self, ui, repopath, bundlepath):
self._tempparent = None
try:
localrepo.localrepository.__init__(self, ui, repopath)
except error.RepoError:
self._tempparent = tempfile.mkdtemp()
localrepo.instance(ui, self._tempparent, 1)
localrepo.localrepository.__init__(self, ui, self._tempparent)
self.ui.setconfig('phases', 'publish', False, 'bundlerepo')
if repopath:
self._url = 'bundle:' + util.expandpath(repopath) + '+' + bundlepath
else:
self._url = 'bundle:' + bundlepath
self.tempfile = None
f = util.posixfile(bundlepath, "rb")
bundle = exchange.readbundle(ui, f, bundlepath)
if isinstance(bundle, bundle2.unbundle20):
self._bundlefile = bundle
self._cgunpacker = None
cgpart = None
for part in bundle.iterparts(seekable=True):
if part.type == 'changegroup':
if cgpart:
raise NotImplementedError("can't process "
"multiple changegroups")
cgpart = part
self._handlebundle2part(bundle, part)
if not cgpart:
raise error.Abort(_("No changegroups found"))
# This is required to placate a later consumer, which expects
# the payload offset to be at the beginning of the changegroup.
# We need to do this after the iterparts() generator advances
# because iterparts() will seek to end of payload after the
# generator returns control to iterparts().
cgpart.seek(0, os.SEEK_SET)
elif isinstance(bundle, changegroup.cg1unpacker):
if bundle.compressed():
f = self._writetempbundle(bundle.read, '.hg10un',
header='HG10UN')
bundle = exchange.readbundle(ui, f, bundlepath, self.vfs)
self._bundlefile = bundle
self._cgunpacker = bundle
else:
raise error.Abort(_('bundle type %s cannot be read') %
type(bundle))
# dict with the mapping 'filename' -> position in the changegroup.
self._cgfilespos = {}
self.firstnewrev = self.changelog.repotiprev + 1
phases.retractboundary(self, None, phases.draft,
[ctx.node() for ctx in self[self.firstnewrev:]])
def _handlebundle2part(self, bundle, part):
if part.type != 'changegroup':
return
cgstream = part
version = part.params.get('version', '01')
legalcgvers = changegroup.supportedincomingversions(self)
if version not in legalcgvers:
msg = _('Unsupported changegroup version: %s')
raise error.Abort(msg % version)
if bundle.compressed():
cgstream = self._writetempbundle(part.read, '.cg%sun' % version)
self._cgunpacker = changegroup.getunbundler(version, cgstream, 'UN')
def _writetempbundle(self, readfn, suffix, header=''):
"""Write a temporary file to disk
"""
fdtemp, temp = self.vfs.mkstemp(prefix="hg-bundle-",
suffix=suffix)
self.tempfile = temp
with os.fdopen(fdtemp, r'wb') as fptemp:
fptemp.write(header)
while True:
chunk = readfn(2**18)
if not chunk:
break
fptemp.write(chunk)
return self.vfs.open(self.tempfile, mode="rb")
@localrepo.unfilteredpropertycache
def _phasecache(self):
return bundlephasecache(self, self._phasedefaults)
@localrepo.unfilteredpropertycache
def changelog(self):
# consume the header if it exists
self._cgunpacker.changelogheader()
c = bundlechangelog(self.svfs, self._cgunpacker)
self.manstart = self._cgunpacker.tell()
return c
def _constructmanifest(self):
self._cgunpacker.seek(self.manstart)
# consume the header if it exists
self._cgunpacker.manifestheader()
linkmapper = self.unfiltered().changelog.rev
m = bundlemanifest(self.svfs, self._cgunpacker, linkmapper)
self.filestart = self._cgunpacker.tell()
return m
def _consumemanifest(self):
"""Consumes the manifest portion of the bundle, setting filestart so the
file portion can be read."""
self._cgunpacker.seek(self.manstart)
self._cgunpacker.manifestheader()
for delta in self._cgunpacker.deltaiter():
pass
self.filestart = self._cgunpacker.tell()
@localrepo.unfilteredpropertycache
def manstart(self):
self.changelog
return self.manstart
@localrepo.unfilteredpropertycache
def filestart(self):
self.manifestlog
# If filestart was not set by self.manifestlog, that means the
# manifestlog implementation did not consume the manifests from the
# changegroup (ex: it might be consuming trees from a separate bundle2
# part instead). So we need to manually consume it.
if r'filestart' not in self.__dict__:
self._consumemanifest()
return self.filestart
def url(self):
return self._url
def file(self, f):
if not self._cgfilespos:
self._cgunpacker.seek(self.filestart)
self._cgfilespos = _getfilestarts(self._cgunpacker)
if f in self._cgfilespos:
self._cgunpacker.seek(self._cgfilespos[f])
linkmapper = self.unfiltered().changelog.rev
return bundlefilelog(self.svfs, f, self._cgunpacker, linkmapper)
else:
return filelog.filelog(self.svfs, f)
def close(self):
"""Close assigned bundle file immediately."""
self._bundlefile.close()
if self.tempfile is not None:
self.vfs.unlink(self.tempfile)
if self._tempparent:
shutil.rmtree(self._tempparent, True)
def cancopy(self):
return False
def peer(self):
return bundlepeer(self)
def getcwd(self):
return pycompat.getcwd() # always outside the repo
# Check if parents exist in localrepo before setting
def setparents(self, p1, p2=nullid):
p1rev = self.changelog.rev(p1)
p2rev = self.changelog.rev(p2)
msg = _("setting parent to node %s that only exists in the bundle\n")
if self.changelog.repotiprev < p1rev:
self.ui.warn(msg % nodemod.hex(p1))
if self.changelog.repotiprev < p2rev:
self.ui.warn(msg % nodemod.hex(p2))
return super(bundlerepository, self).setparents(p1, p2)
def instance(ui, path, create):
if create:
raise error.Abort(_('cannot create new bundle repository'))
# internal config: bundle.mainreporoot
parentpath = ui.config("bundle", "mainreporoot")
if not parentpath:
# try to find the correct path to the working directory repo
parentpath = cmdutil.findrepo(pycompat.getcwd())
if parentpath is None:
parentpath = ''
if parentpath:
# Try to make the full path relative so we get a nice, short URL.
# In particular, we don't want temp dir names in test outputs.
cwd = pycompat.getcwd()
if parentpath == cwd:
parentpath = ''
else:
cwd = pathutil.normasprefix(cwd)
if parentpath.startswith(cwd):
parentpath = parentpath[len(cwd):]
u = util.url(path)
path = u.localpath()
if u.scheme == 'bundle':
s = path.split("+", 1)
if len(s) == 1:
repopath, bundlename = parentpath, s[0]
else:
repopath, bundlename = s
else:
repopath, bundlename = parentpath, path
return bundlerepository(ui, repopath, bundlename)
class bundletransactionmanager(object):
def transaction(self):
return None
def close(self):
raise NotImplementedError
def release(self):
raise NotImplementedError
def getremotechanges(ui, repo, other, onlyheads=None, bundlename=None,
force=False):
'''obtains a bundle of changes incoming from other
"onlyheads" restricts the returned changes to those reachable from the
specified heads.
"bundlename", if given, stores the bundle to this file path permanently;
otherwise it's stored to a temp file and gets deleted again when you call
the returned "cleanupfn".
"force" indicates whether to proceed on unrelated repos.
Returns a tuple (local, csets, cleanupfn):
"local" is a local repo from which to obtain the actual incoming
changesets; it is a bundlerepo for the obtained bundle when the
original "other" is remote.
"csets" lists the incoming changeset node ids.
"cleanupfn" must be called without arguments when you're done processing
the changes; it closes both the original "other" and the one returned
here.
'''
tmp = discovery.findcommonincoming(repo, other, heads=onlyheads,
force=force)
common, incoming, rheads = tmp
if not incoming:
try:
if bundlename:
os.unlink(bundlename)
except OSError:
pass
return repo, [], other.close
commonset = set(common)
rheads = [x for x in rheads if x not in commonset]
bundle = None
bundlerepo = None
localrepo = other.local()
if bundlename or not localrepo:
# create a bundle (uncompressed if other repo is not local)
# developer config: devel.legacy.exchange
legexc = ui.configlist('devel', 'legacy.exchange')
forcebundle1 = 'bundle2' not in legexc and 'bundle1' in legexc
canbundle2 = (not forcebundle1
and other.capable('getbundle')
and other.capable('bundle2'))
if canbundle2:
kwargs = {}
kwargs[r'common'] = common
kwargs[r'heads'] = rheads
kwargs[r'bundlecaps'] = exchange.caps20to10(repo, role='client')
kwargs[r'cg'] = True
b2 = other.getbundle('incoming', **kwargs)
fname = bundle = changegroup.writechunks(ui, b2._forwardchunks(),
bundlename)
else:
if other.capable('getbundle'):
cg = other.getbundle('incoming', common=common, heads=rheads)
elif onlyheads is None and not other.capable('changegroupsubset'):
# compat with older servers when pulling all remote heads
cg = other.changegroup(incoming, "incoming")
rheads = None
else:
cg = other.changegroupsubset(incoming, rheads, 'incoming')
if localrepo:
bundletype = "HG10BZ"
else:
bundletype = "HG10UN"
fname = bundle = bundle2.writebundle(ui, cg, bundlename,
bundletype)
# keep written bundle?
if bundlename:
bundle = None
if not localrepo:
# use the created uncompressed bundlerepo
localrepo = bundlerepo = bundlerepository(repo.baseui, repo.root,
fname)
# this repo contains local and other now, so filter out local again
common = repo.heads()
if localrepo:
# Part of common may be remotely filtered
# So use an unfiltered version
# The discovery process probably need cleanup to avoid that
localrepo = localrepo.unfiltered()
csets = localrepo.changelog.findmissing(common, rheads)
if bundlerepo:
reponodes = [ctx.node() for ctx in bundlerepo[bundlerepo.firstnewrev:]]
remotephases = other.listkeys('phases')
pullop = exchange.pulloperation(bundlerepo, other, heads=reponodes)
pullop.trmanager = bundletransactionmanager()
exchange._pullapplyphases(pullop, remotephases)
def cleanup():
if bundlerepo:
bundlerepo.close()
if bundle:
os.unlink(bundle)
other.close()
return (localrepo, csets, cleanup)