##// 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:

r35305:f77121b6 default
r37304:9bfcbe4f default
Show More
dagutil.py
287 lines | 8.2 KiB | text/x-python | PythonLexer
# dagutil.py - dag utilities for mercurial
#
# Copyright 2010 Benoit Boissinot <bboissin@gmail.com>
# and Peter Arrenbrecht <peter@arrenbrecht.ch>
#
# 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
from .i18n import _
from .node import nullrev
class basedag(object):
'''generic interface for DAGs
terms:
"ix" (short for index) identifies a nodes internally,
"id" identifies one externally.
All params are ixs unless explicitly suffixed otherwise.
Pluralized params are lists or sets.
'''
def __init__(self):
self._inverse = None
def nodeset(self):
'''set of all node ixs'''
raise NotImplementedError
def heads(self):
'''list of head ixs'''
raise NotImplementedError
def parents(self, ix):
'''list of parents ixs of ix'''
raise NotImplementedError
def inverse(self):
'''inverse DAG, where parents becomes children, etc.'''
raise NotImplementedError
def ancestorset(self, starts, stops=None):
'''
set of all ancestors of starts (incl), but stop walk at stops (excl)
'''
raise NotImplementedError
def descendantset(self, starts, stops=None):
'''
set of all descendants of starts (incl), but stop walk at stops (excl)
'''
return self.inverse().ancestorset(starts, stops)
def headsetofconnecteds(self, ixs):
'''
subset of connected list of ixs so that no node has a descendant in it
By "connected list" we mean that if an ancestor and a descendant are in
the list, then so is at least one path connecting them.
'''
raise NotImplementedError
def externalize(self, ix):
'''return a node id'''
return self._externalize(ix)
def externalizeall(self, ixs):
'''return a list of (or set if given a set) of node ids'''
ids = self._externalizeall(ixs)
if isinstance(ixs, set):
return set(ids)
return list(ids)
def internalize(self, id):
'''return a node ix'''
return self._internalize(id)
def internalizeall(self, ids, filterunknown=False):
'''return a list of (or set if given a set) of node ixs'''
ixs = self._internalizeall(ids, filterunknown)
if isinstance(ids, set):
return set(ixs)
return list(ixs)
class genericdag(basedag):
'''generic implementations for DAGs'''
def ancestorset(self, starts, stops=None):
if stops:
stops = set(stops)
else:
stops = set()
seen = set()
pending = list(starts)
while pending:
n = pending.pop()
if n not in seen and n not in stops:
seen.add(n)
pending.extend(self.parents(n))
return seen
def headsetofconnecteds(self, ixs):
hds = set(ixs)
if not hds:
return hds
for n in ixs:
for p in self.parents(n):
hds.discard(p)
assert hds
return hds
class revlogbaseddag(basedag):
'''generic dag interface to a revlog'''
def __init__(self, revlog, nodeset):
basedag.__init__(self)
self._revlog = revlog
self._heads = None
self._nodeset = nodeset
def nodeset(self):
return self._nodeset
def heads(self):
if self._heads is None:
self._heads = self._getheads()
return self._heads
def _externalize(self, ix):
return self._revlog.index[ix][7]
def _externalizeall(self, ixs):
idx = self._revlog.index
return [idx[i][7] for i in ixs]
def _internalize(self, id):
ix = self._revlog.rev(id)
if ix == nullrev:
raise LookupError(id, self._revlog.indexfile, _('nullid'))
return ix
def _internalizeall(self, ids, filterunknown):
rl = self._revlog
if filterunknown:
return [r for r in map(rl.nodemap.get, ids)
if (r is not None
and r != nullrev
and r not in rl.filteredrevs)]
return [self._internalize(i) for i in ids]
class revlogdag(revlogbaseddag):
'''dag interface to a revlog'''
def __init__(self, revlog, localsubset=None):
revlogbaseddag.__init__(self, revlog, set(revlog))
self._heads = localsubset
def _getheads(self):
return [r for r in self._revlog.headrevs() if r != nullrev]
def parents(self, ix):
rlog = self._revlog
idx = rlog.index
revdata = idx[ix]
prev = revdata[5]
if prev != nullrev:
prev2 = revdata[6]
if prev2 == nullrev:
return [prev]
return [prev, prev2]
prev2 = revdata[6]
if prev2 != nullrev:
return [prev2]
return []
def inverse(self):
if self._inverse is None:
self._inverse = inverserevlogdag(self)
return self._inverse
def ancestorset(self, starts, stops=None):
rlog = self._revlog
idx = rlog.index
if stops:
stops = set(stops)
else:
stops = set()
seen = set()
pending = list(starts)
while pending:
rev = pending.pop()
if rev not in seen and rev not in stops:
seen.add(rev)
revdata = idx[rev]
for i in [5, 6]:
prev = revdata[i]
if prev != nullrev:
pending.append(prev)
return seen
def headsetofconnecteds(self, ixs):
if not ixs:
return set()
rlog = self._revlog
idx = rlog.index
headrevs = set(ixs)
for rev in ixs:
revdata = idx[rev]
for i in [5, 6]:
prev = revdata[i]
if prev != nullrev:
headrevs.discard(prev)
assert headrevs
return headrevs
def linearize(self, ixs):
'''linearize and topologically sort a list of revisions
The linearization process tries to create long runs of revs where
a child rev comes immediately after its first parent. This is done by
visiting the heads of the given revs in inverse topological order,
and for each visited rev, visiting its second parent, then its first
parent, then adding the rev itself to the output list.
'''
sorted = []
visit = list(self.headsetofconnecteds(ixs))
visit.sort(reverse=True)
finished = set()
while visit:
cur = visit.pop()
if cur < 0:
cur = -cur - 1
if cur not in finished:
sorted.append(cur)
finished.add(cur)
else:
visit.append(-cur - 1)
visit += [p for p in self.parents(cur)
if p in ixs and p not in finished]
assert len(sorted) == len(ixs)
return sorted
class inverserevlogdag(revlogbaseddag, genericdag):
'''inverse of an existing revlog dag; see revlogdag.inverse()'''
def __init__(self, orig):
revlogbaseddag.__init__(self, orig._revlog, orig._nodeset)
self._orig = orig
self._children = {}
self._roots = []
self._walkfrom = len(self._revlog) - 1
def _walkto(self, walkto):
rev = self._walkfrom
cs = self._children
roots = self._roots
idx = self._revlog.index
while rev >= walkto:
data = idx[rev]
isroot = True
for prev in [data[5], data[6]]: # parent revs
if prev != nullrev:
cs.setdefault(prev, []).append(rev)
isroot = False
if isroot:
roots.append(rev)
rev -= 1
self._walkfrom = rev
def _getheads(self):
self._walkto(nullrev)
return self._roots
def parents(self, ix):
if ix is None:
return []
if ix <= self._walkfrom:
self._walkto(ix)
return self._children.get(ix, [])
def inverse(self):
return self._orig