##// 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:

r35128:8287df8b default
r37304:9bfcbe4f default
Show More
hbisect.py
294 lines | 10.3 KiB | text/x-python | PythonLexer
# changelog bisection for mercurial
#
# Copyright 2007 Matt Mackall
# Copyright 2005, 2006 Benoit Boissinot <benoit.boissinot@ens-lyon.org>
#
# Inspired by git bisect, extension skeleton taken from mq.py.
#
# 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 collections
from .i18n import _
from .node import (
hex,
short,
)
from . import (
error,
)
def bisect(repo, state):
"""find the next node (if any) for testing during a bisect search.
returns a (nodes, number, good) tuple.
'nodes' is the final result of the bisect if 'number' is 0.
Otherwise 'number' indicates the remaining possible candidates for
the search and 'nodes' contains the next bisect target.
'good' is True if bisect is searching for a first good changeset, False
if searching for a first bad one.
"""
changelog = repo.changelog
clparents = changelog.parentrevs
skip = set([changelog.rev(n) for n in state['skip']])
def buildancestors(bad, good):
badrev = min([changelog.rev(n) for n in bad])
ancestors = collections.defaultdict(lambda: None)
for rev in repo.revs("descendants(%ln) - ancestors(%ln)", good, good):
ancestors[rev] = []
if ancestors[badrev] is None:
return badrev, None
return badrev, ancestors
good = False
badrev, ancestors = buildancestors(state['bad'], state['good'])
if not ancestors: # looking for bad to good transition?
good = True
badrev, ancestors = buildancestors(state['good'], state['bad'])
bad = changelog.node(badrev)
if not ancestors: # now we're confused
if (len(state['bad']) == 1 and len(state['good']) == 1 and
state['bad'] != state['good']):
raise error.Abort(_("starting revisions are not directly related"))
raise error.Abort(_("inconsistent state, %d:%s is good and bad")
% (badrev, short(bad)))
# build children dict
children = {}
visit = collections.deque([badrev])
candidates = []
while visit:
rev = visit.popleft()
if ancestors[rev] == []:
candidates.append(rev)
for prev in clparents(rev):
if prev != -1:
if prev in children:
children[prev].append(rev)
else:
children[prev] = [rev]
visit.append(prev)
candidates.sort()
# have we narrowed it down to one entry?
# or have all other possible candidates besides 'bad' have been skipped?
tot = len(candidates)
unskipped = [c for c in candidates if (c not in skip) and (c != badrev)]
if tot == 1 or not unskipped:
return ([changelog.node(c) for c in candidates], 0, good)
perfect = tot // 2
# find the best node to test
best_rev = None
best_len = -1
poison = set()
for rev in candidates:
if rev in poison:
# poison children
poison.update(children.get(rev, []))
continue
a = ancestors[rev] or [rev]
ancestors[rev] = None
x = len(a) # number of ancestors
y = tot - x # number of non-ancestors
value = min(x, y) # how good is this test?
if value > best_len and rev not in skip:
best_len = value
best_rev = rev
if value == perfect: # found a perfect candidate? quit early
break
if y < perfect and rev not in skip: # all downhill from here?
# poison children
poison.update(children.get(rev, []))
continue
for c in children.get(rev, []):
if ancestors[c]:
ancestors[c] = list(set(ancestors[c] + a))
else:
ancestors[c] = a + [c]
assert best_rev is not None
best_node = changelog.node(best_rev)
return ([best_node], tot, good)
def extendrange(repo, state, nodes, good):
# bisect is incomplete when it ends on a merge node and
# one of the parent was not checked.
parents = repo[nodes[0]].parents()
if len(parents) > 1:
if good:
side = state['bad']
else:
side = state['good']
num = len(set(i.node() for i in parents) & set(side))
if num == 1:
return parents[0].ancestor(parents[1])
return None
def load_state(repo):
state = {'current': [], 'good': [], 'bad': [], 'skip': []}
for l in repo.vfs.tryreadlines("bisect.state"):
kind, node = l[:-1].split()
node = repo.lookup(node)
if kind not in state:
raise error.Abort(_("unknown bisect kind %s") % kind)
state[kind].append(node)
return state
def save_state(repo, state):
f = repo.vfs("bisect.state", "w", atomictemp=True)
with repo.wlock():
for kind in sorted(state):
for node in state[kind]:
f.write("%s %s\n" % (kind, hex(node)))
f.close()
def resetstate(repo):
"""remove any bisect state from the repository"""
if repo.vfs.exists("bisect.state"):
repo.vfs.unlink("bisect.state")
def checkstate(state):
"""check we have both 'good' and 'bad' to define a range
Raise Abort exception otherwise."""
if state['good'] and state['bad']:
return True
if not state['good']:
raise error.Abort(_('cannot bisect (no known good revisions)'))
else:
raise error.Abort(_('cannot bisect (no known bad revisions)'))
def get(repo, status):
"""
Return a list of revision(s) that match the given status:
- ``good``, ``bad``, ``skip``: csets explicitly marked as good/bad/skip
- ``goods``, ``bads`` : csets topologically good/bad
- ``range`` : csets taking part in the bisection
- ``pruned`` : csets that are goods, bads or skipped
- ``untested`` : csets whose fate is yet unknown
- ``ignored`` : csets ignored due to DAG topology
- ``current`` : the cset currently being bisected
"""
state = load_state(repo)
if status in ('good', 'bad', 'skip', 'current'):
return map(repo.changelog.rev, state[status])
else:
# In the following sets, we do *not* call 'bisect()' with more
# than one level of recursion, because that can be very, very
# time consuming. Instead, we always develop the expression as
# much as possible.
# 'range' is all csets that make the bisection:
# - have a good ancestor and a bad descendant, or conversely
# that's because the bisection can go either way
range = '( bisect(bad)::bisect(good) | bisect(good)::bisect(bad) )'
_t = repo.revs('bisect(good)::bisect(bad)')
# The sets of topologically good or bad csets
if len(_t) == 0:
# Goods are topologically after bads
goods = 'bisect(good)::' # Pruned good csets
bads = '::bisect(bad)' # Pruned bad csets
else:
# Goods are topologically before bads
goods = '::bisect(good)' # Pruned good csets
bads = 'bisect(bad)::' # Pruned bad csets
# 'pruned' is all csets whose fate is already known: good, bad, skip
skips = 'bisect(skip)' # Pruned skipped csets
pruned = '( (%s) | (%s) | (%s) )' % (goods, bads, skips)
# 'untested' is all cset that are- in 'range', but not in 'pruned'
untested = '( (%s) - (%s) )' % (range, pruned)
# 'ignored' is all csets that were not used during the bisection
# due to DAG topology, but may however have had an impact.
# E.g., a branch merged between bads and goods, but whose branch-
# point is out-side of the range.
iba = '::bisect(bad) - ::bisect(good)' # Ignored bads' ancestors
iga = '::bisect(good) - ::bisect(bad)' # Ignored goods' ancestors
ignored = '( ( (%s) | (%s) ) - (%s) )' % (iba, iga, range)
if status == 'range':
return repo.revs(range)
elif status == 'pruned':
return repo.revs(pruned)
elif status == 'untested':
return repo.revs(untested)
elif status == 'ignored':
return repo.revs(ignored)
elif status == "goods":
return repo.revs(goods)
elif status == "bads":
return repo.revs(bads)
else:
raise error.ParseError(_('invalid bisect state'))
def label(repo, node):
rev = repo.changelog.rev(node)
# Try explicit sets
if rev in get(repo, 'good'):
# i18n: bisect changeset status
return _('good')
if rev in get(repo, 'bad'):
# i18n: bisect changeset status
return _('bad')
if rev in get(repo, 'skip'):
# i18n: bisect changeset status
return _('skipped')
if rev in get(repo, 'untested') or rev in get(repo, 'current'):
# i18n: bisect changeset status
return _('untested')
if rev in get(repo, 'ignored'):
# i18n: bisect changeset status
return _('ignored')
# Try implicit sets
if rev in get(repo, 'goods'):
# i18n: bisect changeset status
return _('good (implicit)')
if rev in get(repo, 'bads'):
# i18n: bisect changeset status
return _('bad (implicit)')
return None
def printresult(ui, repo, state, displayer, nodes, good):
if len(nodes) == 1:
# narrowed it down to a single revision
if good:
ui.write(_("The first good revision is:\n"))
else:
ui.write(_("The first bad revision is:\n"))
displayer.show(repo[nodes[0]])
extendnode = extendrange(repo, state, nodes, good)
if extendnode is not None:
ui.write(_('Not all ancestors of this changeset have been'
' checked.\nUse bisect --extend to continue the '
'bisection from\nthe common ancestor, %s.\n')
% extendnode)
else:
# multiple possible revisions
if good:
ui.write(_("Due to skipped revisions, the first "
"good revision could be any of:\n"))
else:
ui.write(_("Due to skipped revisions, the first "
"bad revision could be any of:\n"))
for n in nodes:
displayer.show(repo[n])
displayer.close()