##// END OF EJS Templates
wireproto: add streams to frame-based protocol...
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

File last commit:

r36732:613954a1 default
r37304:9bfcbe4f default
Show More
discovery.py
525 lines | 20.4 KiB | text/x-python | PythonLexer
# discovery.py - protocol changeset discovery functions
#
# Copyright 2010 Matt Mackall <mpm@selenic.com>
#
# 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 functools
from .i18n import _
from .node import (
hex,
nullid,
short,
)
from . import (
bookmarks,
branchmap,
error,
phases,
scmutil,
setdiscovery,
treediscovery,
util,
)
def findcommonincoming(repo, remote, heads=None, force=False, ancestorsof=None):
"""Return a tuple (common, anyincoming, heads) used to identify the common
subset of nodes between repo and remote.
"common" is a list of (at least) the heads of the common subset.
"anyincoming" is testable as a boolean indicating if any nodes are missing
locally. If remote does not support getbundle, this actually is a list of
roots of the nodes that would be incoming, to be supplied to
changegroupsubset. No code except for pull should be relying on this fact
any longer.
"heads" is either the supplied heads, or else the remote's heads.
"ancestorsof" if not None, restrict the discovery to a subset defined by
these nodes. Changeset outside of this set won't be considered (and
won't appears in "common")
If you pass heads and they are all known locally, the response lists just
these heads in "common" and in "heads".
Please use findcommonoutgoing to compute the set of outgoing nodes to give
extensions a good hook into outgoing.
"""
if not remote.capable('getbundle'):
return treediscovery.findcommonincoming(repo, remote, heads, force)
if heads:
knownnode = repo.changelog.hasnode # no nodemap until it is filtered
if all(knownnode(h) for h in heads):
return (heads, False, heads)
res = setdiscovery.findcommonheads(repo.ui, repo, remote,
abortwhenunrelated=not force,
ancestorsof=ancestorsof)
common, anyinc, srvheads = res
return (list(common), anyinc, heads or list(srvheads))
class outgoing(object):
'''Represents the set of nodes present in a local repo but not in a
(possibly) remote one.
Members:
missing is a list of all nodes present in local but not in remote.
common is a list of all nodes shared between the two repos.
excluded is the list of missing changeset that shouldn't be sent remotely.
missingheads is the list of heads of missing.
commonheads is the list of heads of common.
The sets are computed on demand from the heads, unless provided upfront
by discovery.'''
def __init__(self, repo, commonheads=None, missingheads=None,
missingroots=None):
# at least one of them must not be set
assert None in (commonheads, missingroots)
cl = repo.changelog
if missingheads is None:
missingheads = cl.heads()
if missingroots:
discbases = []
for n in missingroots:
discbases.extend([p for p in cl.parents(n) if p != nullid])
# TODO remove call to nodesbetween.
# TODO populate attributes on outgoing instance instead of setting
# discbases.
csets, roots, heads = cl.nodesbetween(missingroots, missingheads)
included = set(csets)
missingheads = heads
commonheads = [n for n in discbases if n not in included]
elif not commonheads:
commonheads = [nullid]
self.commonheads = commonheads
self.missingheads = missingheads
self._revlog = cl
self._common = None
self._missing = None
self.excluded = []
def _computecommonmissing(self):
sets = self._revlog.findcommonmissing(self.commonheads,
self.missingheads)
self._common, self._missing = sets
@util.propertycache
def common(self):
if self._common is None:
self._computecommonmissing()
return self._common
@util.propertycache
def missing(self):
if self._missing is None:
self._computecommonmissing()
return self._missing
def findcommonoutgoing(repo, other, onlyheads=None, force=False,
commoninc=None, portable=False):
'''Return an outgoing instance to identify the nodes present in repo but
not in other.
If onlyheads is given, only nodes ancestral to nodes in onlyheads
(inclusive) are included. If you already know the local repo's heads,
passing them in onlyheads is faster than letting them be recomputed here.
If commoninc is given, it must be the result of a prior call to
findcommonincoming(repo, other, force) to avoid recomputing it here.
If portable is given, compute more conservative common and missingheads,
to make bundles created from the instance more portable.'''
# declare an empty outgoing object to be filled later
og = outgoing(repo, None, None)
# get common set if not provided
if commoninc is None:
commoninc = findcommonincoming(repo, other, force=force,
ancestorsof=onlyheads)
og.commonheads, _any, _hds = commoninc
# compute outgoing
mayexclude = (repo._phasecache.phaseroots[phases.secret] or repo.obsstore)
if not mayexclude:
og.missingheads = onlyheads or repo.heads()
elif onlyheads is None:
# use visible heads as it should be cached
og.missingheads = repo.filtered("served").heads()
og.excluded = [ctx.node() for ctx in repo.set('secret() or extinct()')]
else:
# compute common, missing and exclude secret stuff
sets = repo.changelog.findcommonmissing(og.commonheads, onlyheads)
og._common, allmissing = sets
og._missing = missing = []
og.excluded = excluded = []
for node in allmissing:
ctx = repo[node]
if ctx.phase() >= phases.secret or ctx.extinct():
excluded.append(node)
else:
missing.append(node)
if len(missing) == len(allmissing):
missingheads = onlyheads
else: # update missing heads
missingheads = phases.newheads(repo, onlyheads, excluded)
og.missingheads = missingheads
if portable:
# recompute common and missingheads as if -r<rev> had been given for
# each head of missing, and --base <rev> for each head of the proper
# ancestors of missing
og._computecommonmissing()
cl = repo.changelog
missingrevs = set(cl.rev(n) for n in og._missing)
og._common = set(cl.ancestors(missingrevs)) - missingrevs
commonheads = set(og.commonheads)
og.missingheads = [h for h in og.missingheads if h not in commonheads]
return og
def _headssummary(pushop):
"""compute a summary of branch and heads status before and after push
return {'branch': ([remoteheads], [newheads],
[unsyncedheads], [discardedheads])} mapping
- branch: the branch name,
- remoteheads: the list of remote heads known locally
None if the branch is new,
- newheads: the new remote heads (known locally) with outgoing pushed,
- unsyncedheads: the list of remote heads unknown locally,
- discardedheads: the list of heads made obsolete by the push.
"""
repo = pushop.repo.unfiltered()
remote = pushop.remote
outgoing = pushop.outgoing
cl = repo.changelog
headssum = {}
# A. Create set of branches involved in the push.
branches = set(repo[n].branch() for n in outgoing.missing)
remotemap = remote.branchmap()
newbranches = branches - set(remotemap)
branches.difference_update(newbranches)
# A. register remote heads
remotebranches = set()
for branch, heads in remote.branchmap().iteritems():
remotebranches.add(branch)
known = []
unsynced = []
knownnode = cl.hasnode # do not use nodemap until it is filtered
for h in heads:
if knownnode(h):
known.append(h)
else:
unsynced.append(h)
headssum[branch] = (known, list(known), unsynced)
# B. add new branch data
missingctx = list(repo[n] for n in outgoing.missing)
touchedbranches = set()
for ctx in missingctx:
branch = ctx.branch()
touchedbranches.add(branch)
if branch not in headssum:
headssum[branch] = (None, [], [])
# C drop data about untouched branches:
for branch in remotebranches - touchedbranches:
del headssum[branch]
# D. Update newmap with outgoing changes.
# This will possibly add new heads and remove existing ones.
newmap = branchmap.branchcache((branch, heads[1])
for branch, heads in headssum.iteritems()
if heads[0] is not None)
newmap.update(repo, (ctx.rev() for ctx in missingctx))
for branch, newheads in newmap.iteritems():
headssum[branch][1][:] = newheads
for branch, items in headssum.iteritems():
for l in items:
if l is not None:
l.sort()
headssum[branch] = items + ([],)
# If there are no obsstore, no post processing are needed.
if repo.obsstore:
torev = repo.changelog.rev
futureheads = set(torev(h) for h in outgoing.missingheads)
futureheads |= set(torev(h) for h in outgoing.commonheads)
allfuturecommon = repo.changelog.ancestors(futureheads, inclusive=True)
for branch, heads in sorted(headssum.iteritems()):
remoteheads, newheads, unsyncedheads, placeholder = heads
result = _postprocessobsolete(pushop, allfuturecommon, newheads)
headssum[branch] = (remoteheads, sorted(result[0]), unsyncedheads,
sorted(result[1]))
return headssum
def _oldheadssummary(repo, remoteheads, outgoing, inc=False):
"""Compute branchmapsummary for repo without branchmap support"""
# 1-4b. old servers: Check for new topological heads.
# Construct {old,new}map with branch = None (topological branch).
# (code based on update)
knownnode = repo.changelog.hasnode # no nodemap until it is filtered
oldheads = sorted(h for h in remoteheads if knownnode(h))
# all nodes in outgoing.missing are children of either:
# - an element of oldheads
# - another element of outgoing.missing
# - nullrev
# This explains why the new head are very simple to compute.
r = repo.set('heads(%ln + %ln)', oldheads, outgoing.missing)
newheads = sorted(c.node() for c in r)
# set some unsynced head to issue the "unsynced changes" warning
if inc:
unsynced = [None]
else:
unsynced = []
return {None: (oldheads, newheads, unsynced, [])}
def _nowarnheads(pushop):
# Compute newly pushed bookmarks. We don't warn about bookmarked heads.
repo = pushop.repo.unfiltered()
remote = pushop.remote
localbookmarks = repo._bookmarks
remotebookmarks = remote.listkeys('bookmarks')
bookmarkedheads = set()
# internal config: bookmarks.pushing
newbookmarks = [localbookmarks.expandname(b)
for b in pushop.ui.configlist('bookmarks', 'pushing')]
for bm in localbookmarks:
rnode = remotebookmarks.get(bm)
if rnode and rnode in repo:
lctx, rctx = repo[bm], repo[rnode]
if bookmarks.validdest(repo, rctx, lctx):
bookmarkedheads.add(lctx.node())
else:
if bm in newbookmarks and bm not in remotebookmarks:
bookmarkedheads.add(repo[bm].node())
return bookmarkedheads
def checkheads(pushop):
"""Check that a push won't add any outgoing head
raise Abort error and display ui message as needed.
"""
repo = pushop.repo.unfiltered()
remote = pushop.remote
outgoing = pushop.outgoing
remoteheads = pushop.remoteheads
newbranch = pushop.newbranch
inc = bool(pushop.incoming)
# Check for each named branch if we're creating new remote heads.
# To be a remote head after push, node must be either:
# - unknown locally
# - a local outgoing head descended from update
# - a remote head that's known locally and not
# ancestral to an outgoing head
if remoteheads == [nullid]:
# remote is empty, nothing to check.
return
if remote.capable('branchmap'):
headssum = _headssummary(pushop)
else:
headssum = _oldheadssummary(repo, remoteheads, outgoing, inc)
pushop.pushbranchmap = headssum
newbranches = [branch for branch, heads in headssum.iteritems()
if heads[0] is None]
# 1. Check for new branches on the remote.
if newbranches and not newbranch: # new branch requires --new-branch
branchnames = ', '.join(sorted(newbranches))
raise error.Abort(_("push creates new remote branches: %s!")
% branchnames,
hint=_("use 'hg push --new-branch' to create"
" new remote branches"))
# 2. Find heads that we need not warn about
nowarnheads = _nowarnheads(pushop)
# 3. Check for new heads.
# If there are more heads after the push than before, a suitable
# error message, depending on unsynced status, is displayed.
errormsg = None
for branch, heads in sorted(headssum.iteritems()):
remoteheads, newheads, unsyncedheads, discardedheads = heads
# add unsynced data
if remoteheads is None:
oldhs = set()
else:
oldhs = set(remoteheads)
oldhs.update(unsyncedheads)
dhs = None # delta heads, the new heads on branch
newhs = set(newheads)
newhs.update(unsyncedheads)
if unsyncedheads:
if None in unsyncedheads:
# old remote, no heads data
heads = None
else:
heads = scmutil.nodesummaries(repo, unsyncedheads)
if heads is None:
repo.ui.status(_("remote has heads that are "
"not known locally\n"))
elif branch is None:
repo.ui.status(_("remote has heads that are "
"not known locally: %s\n") % heads)
else:
repo.ui.status(_("remote has heads on branch '%s' that are "
"not known locally: %s\n") % (branch, heads))
if remoteheads is None:
if len(newhs) > 1:
dhs = list(newhs)
if errormsg is None:
errormsg = (_("push creates new branch '%s' "
"with multiple heads") % (branch))
hint = _("merge or"
" see 'hg help push' for details about"
" pushing new heads")
elif len(newhs) > len(oldhs):
# remove bookmarked or existing remote heads from the new heads list
dhs = sorted(newhs - nowarnheads - oldhs)
if dhs:
if errormsg is None:
if branch not in ('default', None):
errormsg = _("push creates new remote head %s "
"on branch '%s'!") % (short(dhs[0]), branch)
elif repo[dhs[0]].bookmarks():
errormsg = _("push creates new remote head %s "
"with bookmark '%s'!") % (
short(dhs[0]), repo[dhs[0]].bookmarks()[0])
else:
errormsg = _("push creates new remote head %s!"
) % short(dhs[0])
if unsyncedheads:
hint = _("pull and merge or"
" see 'hg help push' for details about"
" pushing new heads")
else:
hint = _("merge or"
" see 'hg help push' for details about"
" pushing new heads")
if branch is None:
repo.ui.note(_("new remote heads:\n"))
else:
repo.ui.note(_("new remote heads on branch '%s':\n") % branch)
for h in dhs:
repo.ui.note((" %s\n") % short(h))
if errormsg:
raise error.Abort(errormsg, hint=hint)
def _postprocessobsolete(pushop, futurecommon, candidate_newhs):
"""post process the list of new heads with obsolescence information
Exists as a sub-function to contain the complexity and allow extensions to
experiment with smarter logic.
Returns (newheads, discarded_heads) tuple
"""
# known issue
#
# * We "silently" skip processing on all changeset unknown locally
#
# * if <nh> is public on the remote, it won't be affected by obsolete
# marker and a new is created
# define various utilities and containers
repo = pushop.repo
unfi = repo.unfiltered()
tonode = unfi.changelog.node
torev = unfi.changelog.nodemap.get
public = phases.public
getphase = unfi._phasecache.phase
ispublic = (lambda r: getphase(unfi, r) == public)
ispushed = (lambda n: torev(n) in futurecommon)
hasoutmarker = functools.partial(pushingmarkerfor, unfi.obsstore, ispushed)
successorsmarkers = unfi.obsstore.successors
newhs = set() # final set of new heads
discarded = set() # new head of fully replaced branch
localcandidate = set() # candidate heads known locally
unknownheads = set() # candidate heads unknown locally
for h in candidate_newhs:
if h in unfi:
localcandidate.add(h)
else:
if successorsmarkers.get(h) is not None:
msg = ('checkheads: remote head unknown locally has'
' local marker: %s\n')
repo.ui.debug(msg % hex(h))
unknownheads.add(h)
# fast path the simple case
if len(localcandidate) == 1:
return unknownheads | set(candidate_newhs), set()
# actually process branch replacement
while localcandidate:
nh = localcandidate.pop()
# run this check early to skip the evaluation of the whole branch
if (torev(nh) in futurecommon or ispublic(torev(nh))):
newhs.add(nh)
continue
# Get all revs/nodes on the branch exclusive to this head
# (already filtered heads are "ignored"))
branchrevs = unfi.revs('only(%n, (%ln+%ln))',
nh, localcandidate, newhs)
branchnodes = [tonode(r) for r in branchrevs]
# The branch won't be hidden on the remote if
# * any part of it is public,
# * any part of it is considered part of the result by previous logic,
# * if we have no markers to push to obsolete it.
if (any(ispublic(r) for r in branchrevs)
or any(torev(n) in futurecommon for n in branchnodes)
or any(not hasoutmarker(n) for n in branchnodes)):
newhs.add(nh)
else:
# note: there is a corner case if there is a merge in the branch.
# we might end up with -more- heads. However, these heads are not
# "added" by the push, but more by the "removal" on the remote so I
# think is a okay to ignore them,
discarded.add(nh)
newhs |= unknownheads
return newhs, discarded
def pushingmarkerfor(obsstore, ispushed, node):
"""true if some markers are to be pushed for node
We cannot just look in to the pushed obsmarkers from the pushop because
discovery might have filtered relevant markers. In addition listing all
markers relevant to all changesets in the pushed set would be too expensive
(O(len(repo)))
(note: There are cache opportunity in this function. but it would requires
a two dimensional stack.)
"""
successorsmarkers = obsstore.successors
stack = [node]
seen = set(stack)
while stack:
current = stack.pop()
if ispushed(current):
return True
markers = successorsmarkers.get(current, ())
# markers fields = ('prec', 'succs', 'flag', 'meta', 'date', 'parents')
for m in markers:
nexts = m[1] # successors
if not nexts: # this is a prune marker
nexts = m[5] or () # parents
for n in nexts:
if n not in seen:
seen.add(n)
stack.append(n)
return False