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
# repoview.py - Filtered view of a localrepo object
#
# Copyright 2012 Pierre-Yves David <pierre-yves.david@ens-lyon.org>
# Logilab SA <contact@logilab.fr>
#
# 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 copy
import weakref
from .node import nullrev
from . import (
obsolete,
phases,
pycompat,
tags as tagsmod,
)
def hideablerevs(repo):
"""Revision candidates to be hidden
This is a standalone function to allow extensions to wrap it.
Because we use the set of immutable changesets as a fallback subset in
branchmap (see mercurial.branchmap.subsettable), you cannot set "public"
changesets as "hideable". Doing so would break multiple code assertions and
lead to crashes."""
return obsolete.getrevs(repo, 'obsolete')
def pinnedrevs(repo):
"""revisions blocking hidden changesets from being filtered
"""
cl = repo.changelog
pinned = set()
pinned.update([par.rev() for par in repo[None].parents()])
pinned.update([cl.rev(bm) for bm in repo._bookmarks.values()])
tags = {}
tagsmod.readlocaltags(repo.ui, repo, tags, {})
if tags:
rev, nodemap = cl.rev, cl.nodemap
pinned.update(rev(t[0]) for t in tags.values() if t[0] in nodemap)
return pinned
def _revealancestors(pfunc, hidden, revs):
"""reveals contiguous chains of hidden ancestors of 'revs' by removing them
from 'hidden'
- pfunc(r): a funtion returning parent of 'r',
- hidden: the (preliminary) hidden revisions, to be updated
- revs: iterable of revnum,
(Ancestors are revealed exclusively, i.e. the elements in 'revs' are
*not* revealed)
"""
stack = list(revs)
while stack:
for p in pfunc(stack.pop()):
if p != nullrev and p in hidden:
hidden.remove(p)
stack.append(p)
def computehidden(repo, visibilityexceptions=None):
"""compute the set of hidden revision to filter
During most operation hidden should be filtered."""
assert not repo.changelog.filteredrevs
hidden = hideablerevs(repo)
if hidden:
hidden = set(hidden - pinnedrevs(repo))
if visibilityexceptions:
hidden -= visibilityexceptions
pfunc = repo.changelog.parentrevs
mutablephases = (phases.draft, phases.secret)
mutable = repo._phasecache.getrevset(repo, mutablephases)
visible = mutable - hidden
_revealancestors(pfunc, hidden, visible)
return frozenset(hidden)
def computeunserved(repo, visibilityexceptions=None):
"""compute the set of revision that should be filtered when used a server
Secret and hidden changeset should not pretend to be here."""
assert not repo.changelog.filteredrevs
# fast path in simple case to avoid impact of non optimised code
hiddens = filterrevs(repo, 'visible')
if phases.hassecret(repo):
cl = repo.changelog
secret = phases.secret
getphase = repo._phasecache.phase
first = min(cl.rev(n) for n in repo._phasecache.phaseroots[secret])
revs = cl.revs(start=first)
secrets = set(r for r in revs if getphase(repo, r) >= secret)
return frozenset(hiddens | secrets)
else:
return hiddens
def computemutable(repo, visibilityexceptions=None):
assert not repo.changelog.filteredrevs
# fast check to avoid revset call on huge repo
if any(repo._phasecache.phaseroots[1:]):
getphase = repo._phasecache.phase
maymutable = filterrevs(repo, 'base')
return frozenset(r for r in maymutable if getphase(repo, r))
return frozenset()
def computeimpactable(repo, visibilityexceptions=None):
"""Everything impactable by mutable revision
The immutable filter still have some chance to get invalidated. This will
happen when:
- you garbage collect hidden changeset,
- public phase is moved backward,
- something is changed in the filtering (this could be fixed)
This filter out any mutable changeset and any public changeset that may be
impacted by something happening to a mutable revision.
This is achieved by filtered everything with a revision number egal or
higher than the first mutable changeset is filtered."""
assert not repo.changelog.filteredrevs
cl = repo.changelog
firstmutable = len(cl)
for roots in repo._phasecache.phaseroots[1:]:
if roots:
firstmutable = min(firstmutable, min(cl.rev(r) for r in roots))
# protect from nullrev root
firstmutable = max(0, firstmutable)
return frozenset(xrange(firstmutable, len(cl)))
# function to compute filtered set
#
# When adding a new filter you MUST update the table at:
# mercurial.branchmap.subsettable
# Otherwise your filter will have to recompute all its branches cache
# from scratch (very slow).
filtertable = {'visible': computehidden,
'visible-hidden': computehidden,
'served': computeunserved,
'immutable': computemutable,
'base': computeimpactable}
def filterrevs(repo, filtername, visibilityexceptions=None):
"""returns set of filtered revision for this filter name
visibilityexceptions is a set of revs which must are exceptions for
hidden-state and must be visible. They are dynamic and hence we should not
cache it's result"""
if filtername not in repo.filteredrevcache:
func = filtertable[filtername]
if visibilityexceptions:
return func(repo.unfiltered, visibilityexceptions)
repo.filteredrevcache[filtername] = func(repo.unfiltered())
return repo.filteredrevcache[filtername]
class repoview(object):
"""Provide a read/write view of a repo through a filtered changelog
This object is used to access a filtered version of a repository without
altering the original repository object itself. We can not alter the
original object for two main reasons:
- It prevents the use of a repo with multiple filters at the same time. In
particular when multiple threads are involved.
- It makes scope of the filtering harder to control.
This object behaves very closely to the original repository. All attribute
operations are done on the original repository:
- An access to `repoview.someattr` actually returns `repo.someattr`,
- A write to `repoview.someattr` actually sets value of `repo.someattr`,
- A deletion of `repoview.someattr` actually drops `someattr`
from `repo.__dict__`.
The only exception is the `changelog` property. It is overridden to return
a (surface) copy of `repo.changelog` with some revisions filtered. The
`filtername` attribute of the view control the revisions that need to be
filtered. (the fact the changelog is copied is an implementation detail).
Unlike attributes, this object intercepts all method calls. This means that
all methods are run on the `repoview` object with the filtered `changelog`
property. For this purpose the simple `repoview` class must be mixed with
the actual class of the repository. This ensures that the resulting
`repoview` object have the very same methods than the repo object. This
leads to the property below.
repoview.method() --> repo.__class__.method(repoview)
The inheritance has to be done dynamically because `repo` can be of any
subclasses of `localrepo`. Eg: `bundlerepo` or `statichttprepo`.
"""
def __init__(self, repo, filtername, visibilityexceptions=None):
object.__setattr__(self, r'_unfilteredrepo', repo)
object.__setattr__(self, r'filtername', filtername)
object.__setattr__(self, r'_clcachekey', None)
object.__setattr__(self, r'_clcache', None)
# revs which are exceptions and must not be hidden
object.__setattr__(self, r'_visibilityexceptions',
visibilityexceptions)
# not a propertycache on purpose we shall implement a proper cache later
@property
def changelog(self):
"""return a filtered version of the changeset
this changelog must not be used for writing"""
# some cache may be implemented later
unfi = self._unfilteredrepo
unfichangelog = unfi.changelog
# bypass call to changelog.method
unfiindex = unfichangelog.index
unfilen = len(unfiindex) - 1
unfinode = unfiindex[unfilen - 1][7]
revs = filterrevs(unfi, self.filtername, self._visibilityexceptions)
cl = self._clcache
newkey = (unfilen, unfinode, hash(revs), unfichangelog._delayed)
# if cl.index is not unfiindex, unfi.changelog would be
# recreated, and our clcache refers to garbage object
if (cl is not None and
(cl.index is not unfiindex or newkey != self._clcachekey)):
cl = None
# could have been made None by the previous if
if cl is None:
cl = copy.copy(unfichangelog)
cl.filteredrevs = revs
object.__setattr__(self, r'_clcache', cl)
object.__setattr__(self, r'_clcachekey', newkey)
return cl
def unfiltered(self):
"""Return an unfiltered version of a repo"""
return self._unfilteredrepo
def filtered(self, name, visibilityexceptions=None):
"""Return a filtered version of a repository"""
if name == self.filtername and not visibilityexceptions:
return self
return self.unfiltered().filtered(name, visibilityexceptions)
def __repr__(self):
return r'<%s:%s %r>' % (self.__class__.__name__,
pycompat.sysstr(self.filtername),
self.unfiltered())
# everything access are forwarded to the proxied repo
def __getattr__(self, attr):
return getattr(self._unfilteredrepo, attr)
def __setattr__(self, attr, value):
return setattr(self._unfilteredrepo, attr, value)
def __delattr__(self, attr):
return delattr(self._unfilteredrepo, attr)
# Python <3.4 easily leaks types via __mro__. See
# https://bugs.python.org/issue17950. We cache dynamically created types
# so they won't be leaked on every invocation of repo.filtered().
_filteredrepotypes = weakref.WeakKeyDictionary()
def newtype(base):
"""Create a new type with the repoview mixin and the given base class"""
if base not in _filteredrepotypes:
class filteredrepo(repoview, base):
pass
_filteredrepotypes[base] = filteredrepo
return _filteredrepotypes[base]