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
# mdiff.py - diff and patch routines for mercurial
#
# Copyright 2005, 2006 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 re
import struct
import zlib
from .i18n import _
from . import (
encoding,
error,
policy,
pycompat,
util,
)
from .utils import dateutil
_missing_newline_marker = "\\ No newline at end of file\n"
bdiff = policy.importmod(r'bdiff')
mpatch = policy.importmod(r'mpatch')
blocks = bdiff.blocks
fixws = bdiff.fixws
patches = mpatch.patches
patchedsize = mpatch.patchedsize
textdiff = bdiff.bdiff
splitnewlines = bdiff.splitnewlines
class diffopts(object):
'''context is the number of context lines
text treats all files as text
showfunc enables diff -p output
git enables the git extended patch format
nodates removes dates from diff headers
nobinary ignores binary files
noprefix disables the 'a/' and 'b/' prefixes (ignored in plain mode)
ignorews ignores all whitespace changes in the diff
ignorewsamount ignores changes in the amount of whitespace
ignoreblanklines ignores changes whose lines are all blank
upgrade generates git diffs to avoid data loss
'''
defaults = {
'context': 3,
'text': False,
'showfunc': False,
'git': False,
'nodates': False,
'nobinary': False,
'noprefix': False,
'index': 0,
'ignorews': False,
'ignorewsamount': False,
'ignorewseol': False,
'ignoreblanklines': False,
'upgrade': False,
'showsimilarity': False,
'worddiff': False,
'xdiff': False,
}
def __init__(self, **opts):
opts = pycompat.byteskwargs(opts)
for k in self.defaults.keys():
v = opts.get(k)
if v is None:
v = self.defaults[k]
setattr(self, k, v)
try:
self.context = int(self.context)
except ValueError:
raise error.Abort(_('diff context lines count must be '
'an integer, not %r') % self.context)
def copy(self, **kwargs):
opts = dict((k, getattr(self, k)) for k in self.defaults)
opts = pycompat.strkwargs(opts)
opts.update(kwargs)
return diffopts(**opts)
defaultopts = diffopts()
def wsclean(opts, text, blank=True):
if opts.ignorews:
text = bdiff.fixws(text, 1)
elif opts.ignorewsamount:
text = bdiff.fixws(text, 0)
if blank and opts.ignoreblanklines:
text = re.sub('\n+', '\n', text).strip('\n')
if opts.ignorewseol:
text = re.sub(br'[ \t\r\f]+\n', r'\n', text)
return text
def splitblock(base1, lines1, base2, lines2, opts):
# The input lines matches except for interwoven blank lines. We
# transform it into a sequence of matching blocks and blank blocks.
lines1 = [(wsclean(opts, l) and 1 or 0) for l in lines1]
lines2 = [(wsclean(opts, l) and 1 or 0) for l in lines2]
s1, e1 = 0, len(lines1)
s2, e2 = 0, len(lines2)
while s1 < e1 or s2 < e2:
i1, i2, btype = s1, s2, '='
if (i1 >= e1 or lines1[i1] == 0
or i2 >= e2 or lines2[i2] == 0):
# Consume the block of blank lines
btype = '~'
while i1 < e1 and lines1[i1] == 0:
i1 += 1
while i2 < e2 and lines2[i2] == 0:
i2 += 1
else:
# Consume the matching lines
while i1 < e1 and lines1[i1] == 1 and lines2[i2] == 1:
i1 += 1
i2 += 1
yield [base1 + s1, base1 + i1, base2 + s2, base2 + i2], btype
s1 = i1
s2 = i2
def hunkinrange(hunk, linerange):
"""Return True if `hunk` defined as (start, length) is in `linerange`
defined as (lowerbound, upperbound).
>>> hunkinrange((5, 10), (2, 7))
True
>>> hunkinrange((5, 10), (6, 12))
True
>>> hunkinrange((5, 10), (13, 17))
True
>>> hunkinrange((5, 10), (3, 17))
True
>>> hunkinrange((5, 10), (1, 3))
False
>>> hunkinrange((5, 10), (18, 20))
False
>>> hunkinrange((5, 10), (1, 5))
False
>>> hunkinrange((5, 10), (15, 27))
False
"""
start, length = hunk
lowerbound, upperbound = linerange
return lowerbound < start + length and start < upperbound
def blocksinrange(blocks, rangeb):
"""filter `blocks` like (a1, a2, b1, b2) from items outside line range
`rangeb` from ``(b1, b2)`` point of view.
Return `filteredblocks, rangea` where:
* `filteredblocks` is list of ``block = (a1, a2, b1, b2), stype`` items of
`blocks` that are inside `rangeb` from ``(b1, b2)`` point of view; a
block ``(b1, b2)`` being inside `rangeb` if
``rangeb[0] < b2 and b1 < rangeb[1]``;
* `rangea` is the line range w.r.t. to ``(a1, a2)`` parts of `blocks`.
"""
lbb, ubb = rangeb
lba, uba = None, None
filteredblocks = []
for block in blocks:
(a1, a2, b1, b2), stype = block
if lbb >= b1 and ubb <= b2 and stype == '=':
# rangeb is within a single "=" hunk, restrict back linerange1
# by offsetting rangeb
lba = lbb - b1 + a1
uba = ubb - b1 + a1
else:
if b1 <= lbb < b2:
if stype == '=':
lba = a2 - (b2 - lbb)
else:
lba = a1
if b1 < ubb <= b2:
if stype == '=':
uba = a1 + (ubb - b1)
else:
uba = a2
if hunkinrange((b1, (b2 - b1)), rangeb):
filteredblocks.append(block)
if lba is None or uba is None or uba < lba:
raise error.Abort(_('line range exceeds file size'))
return filteredblocks, (lba, uba)
def chooseblocksfunc(opts=None):
if (opts is None or not opts.xdiff
or not util.safehasattr(bdiff, 'xdiffblocks')):
return bdiff.blocks
else:
return bdiff.xdiffblocks
def allblocks(text1, text2, opts=None, lines1=None, lines2=None):
"""Return (block, type) tuples, where block is an mdiff.blocks
line entry. type is '=' for blocks matching exactly one another
(bdiff blocks), '!' for non-matching blocks and '~' for blocks
matching only after having filtered blank lines.
line1 and line2 are text1 and text2 split with splitnewlines() if
they are already available.
"""
if opts is None:
opts = defaultopts
if opts.ignorews or opts.ignorewsamount or opts.ignorewseol:
text1 = wsclean(opts, text1, False)
text2 = wsclean(opts, text2, False)
diff = chooseblocksfunc(opts)(text1, text2)
for i, s1 in enumerate(diff):
# The first match is special.
# we've either found a match starting at line 0 or a match later
# in the file. If it starts later, old and new below will both be
# empty and we'll continue to the next match.
if i > 0:
s = diff[i - 1]
else:
s = [0, 0, 0, 0]
s = [s[1], s1[0], s[3], s1[2]]
# bdiff sometimes gives huge matches past eof, this check eats them,
# and deals with the special first match case described above
if s[0] != s[1] or s[2] != s[3]:
type = '!'
if opts.ignoreblanklines:
if lines1 is None:
lines1 = splitnewlines(text1)
if lines2 is None:
lines2 = splitnewlines(text2)
old = wsclean(opts, "".join(lines1[s[0]:s[1]]))
new = wsclean(opts, "".join(lines2[s[2]:s[3]]))
if old == new:
type = '~'
yield s, type
yield s1, '='
def unidiff(a, ad, b, bd, fn1, fn2, binary, opts=defaultopts):
"""Return a unified diff as a (headers, hunks) tuple.
If the diff is not null, `headers` is a list with unified diff header
lines "--- <original>" and "+++ <new>" and `hunks` is a generator yielding
(hunkrange, hunklines) coming from _unidiff().
Otherwise, `headers` and `hunks` are empty.
Set binary=True if either a or b should be taken as a binary file.
"""
def datetag(date, fn=None):
if not opts.git and not opts.nodates:
return '\t%s' % date
if fn and ' ' in fn:
return '\t'
return ''
sentinel = [], ()
if not a and not b:
return sentinel
if opts.noprefix:
aprefix = bprefix = ''
else:
aprefix = 'a/'
bprefix = 'b/'
epoch = dateutil.datestr((0, 0))
fn1 = util.pconvert(fn1)
fn2 = util.pconvert(fn2)
if binary:
if a and b and len(a) == len(b) and a == b:
return sentinel
headerlines = []
hunks = (None, ['Binary file %s has changed\n' % fn1]),
elif not a:
without_newline = not b.endswith('\n')
b = splitnewlines(b)
if a is None:
l1 = '--- /dev/null%s' % datetag(epoch)
else:
l1 = "--- %s%s%s" % (aprefix, fn1, datetag(ad, fn1))
l2 = "+++ %s%s" % (bprefix + fn2, datetag(bd, fn2))
headerlines = [l1, l2]
size = len(b)
hunkrange = (0, 0, 1, size)
hunklines = ["@@ -0,0 +1,%d @@\n" % size] + ["+" + e for e in b]
if without_newline:
hunklines[-1] += '\n'
hunklines.append(_missing_newline_marker)
hunks = (hunkrange, hunklines),
elif not b:
without_newline = not a.endswith('\n')
a = splitnewlines(a)
l1 = "--- %s%s%s" % (aprefix, fn1, datetag(ad, fn1))
if b is None:
l2 = '+++ /dev/null%s' % datetag(epoch)
else:
l2 = "+++ %s%s%s" % (bprefix, fn2, datetag(bd, fn2))
headerlines = [l1, l2]
size = len(a)
hunkrange = (1, size, 0, 0)
hunklines = ["@@ -1,%d +0,0 @@\n" % size] + ["-" + e for e in a]
if without_newline:
hunklines[-1] += '\n'
hunklines.append(_missing_newline_marker)
hunks = (hunkrange, hunklines),
else:
hunks = _unidiff(a, b, opts=opts)
if not next(hunks):
return sentinel
headerlines = [
"--- %s%s%s" % (aprefix, fn1, datetag(ad, fn1)),
"+++ %s%s%s" % (bprefix, fn2, datetag(bd, fn2)),
]
return headerlines, hunks
def _unidiff(t1, t2, opts=defaultopts):
"""Yield hunks of a headerless unified diff from t1 and t2 texts.
Each hunk consists of a (hunkrange, hunklines) tuple where `hunkrange` is a
tuple (s1, l1, s2, l2) representing the range information of the hunk to
form the '@@ -s1,l1 +s2,l2 @@' header and `hunklines` is a list of lines
of the hunk combining said header followed by line additions and
deletions.
The hunks are prefixed with a bool.
"""
l1 = splitnewlines(t1)
l2 = splitnewlines(t2)
def contextend(l, len):
ret = l + opts.context
if ret > len:
ret = len
return ret
def contextstart(l):
ret = l - opts.context
if ret < 0:
return 0
return ret
lastfunc = [0, '']
def yieldhunk(hunk):
(astart, a2, bstart, b2, delta) = hunk
aend = contextend(a2, len(l1))
alen = aend - astart
blen = b2 - bstart + aend - a2
func = ""
if opts.showfunc:
lastpos, func = lastfunc
# walk backwards from the start of the context up to the start of
# the previous hunk context until we find a line starting with an
# alphanumeric char.
for i in xrange(astart - 1, lastpos - 1, -1):
if l1[i][0:1].isalnum():
func = b' ' + l1[i].rstrip()
# split long function name if ASCII. otherwise we have no
# idea where the multi-byte boundary is, so just leave it.
if encoding.isasciistr(func):
func = func[:41]
lastfunc[1] = func
break
# by recording this hunk's starting point as the next place to
# start looking for function lines, we avoid reading any line in
# the file more than once.
lastfunc[0] = astart
# zero-length hunk ranges report their start line as one less
if alen:
astart += 1
if blen:
bstart += 1
hunkrange = astart, alen, bstart, blen
hunklines = (
["@@ -%d,%d +%d,%d @@%s\n" % (hunkrange + (func,))]
+ delta
+ [' ' + l1[x] for x in xrange(a2, aend)]
)
# If either file ends without a newline and the last line of
# that file is part of a hunk, a marker is printed. If the
# last line of both files is identical and neither ends in
# a newline, print only one marker. That's the only case in
# which the hunk can end in a shared line without a newline.
skip = False
if not t1.endswith('\n') and astart + alen == len(l1) + 1:
for i in xrange(len(hunklines) - 1, -1, -1):
if hunklines[i].startswith(('-', ' ')):
if hunklines[i].startswith(' '):
skip = True
hunklines[i] += '\n'
hunklines.insert(i + 1, _missing_newline_marker)
break
if not skip and not t2.endswith('\n') and bstart + blen == len(l2) + 1:
for i in xrange(len(hunklines) - 1, -1, -1):
if hunklines[i].startswith('+'):
hunklines[i] += '\n'
hunklines.insert(i + 1, _missing_newline_marker)
break
yield hunkrange, hunklines
# bdiff.blocks gives us the matching sequences in the files. The loop
# below finds the spaces between those matching sequences and translates
# them into diff output.
#
hunk = None
ignoredlines = 0
has_hunks = False
for s, stype in allblocks(t1, t2, opts, l1, l2):
a1, a2, b1, b2 = s
if stype != '!':
if stype == '~':
# The diff context lines are based on t1 content. When
# blank lines are ignored, the new lines offsets must
# be adjusted as if equivalent blocks ('~') had the
# same sizes on both sides.
ignoredlines += (b2 - b1) - (a2 - a1)
continue
delta = []
old = l1[a1:a2]
new = l2[b1:b2]
b1 -= ignoredlines
b2 -= ignoredlines
astart = contextstart(a1)
bstart = contextstart(b1)
prev = None
if hunk:
# join with the previous hunk if it falls inside the context
if astart < hunk[1] + opts.context + 1:
prev = hunk
astart = hunk[1]
bstart = hunk[3]
else:
if not has_hunks:
has_hunks = True
yield True
for x in yieldhunk(hunk):
yield x
if prev:
# we've joined the previous hunk, record the new ending points.
hunk[1] = a2
hunk[3] = b2
delta = hunk[4]
else:
# create a new hunk
hunk = [astart, a2, bstart, b2, delta]
delta[len(delta):] = [' ' + x for x in l1[astart:a1]]
delta[len(delta):] = ['-' + x for x in old]
delta[len(delta):] = ['+' + x for x in new]
if hunk:
if not has_hunks:
has_hunks = True
yield True
for x in yieldhunk(hunk):
yield x
elif not has_hunks:
yield False
def b85diff(to, tn):
'''print base85-encoded binary diff'''
def fmtline(line):
l = len(line)
if l <= 26:
l = pycompat.bytechr(ord('A') + l - 1)
else:
l = pycompat.bytechr(l - 26 + ord('a') - 1)
return '%c%s\n' % (l, util.b85encode(line, True))
def chunk(text, csize=52):
l = len(text)
i = 0
while i < l:
yield text[i:i + csize]
i += csize
if to is None:
to = ''
if tn is None:
tn = ''
if to == tn:
return ''
# TODO: deltas
ret = []
ret.append('GIT binary patch\n')
ret.append('literal %d\n' % len(tn))
for l in chunk(zlib.compress(tn)):
ret.append(fmtline(l))
ret.append('\n')
return ''.join(ret)
def patchtext(bin):
pos = 0
t = []
while pos < len(bin):
p1, p2, l = struct.unpack(">lll", bin[pos:pos + 12])
pos += 12
t.append(bin[pos:pos + l])
pos += l
return "".join(t)
def patch(a, bin):
if len(a) == 0:
# skip over trivial delta header
return util.buffer(bin, 12)
return mpatch.patches(a, [bin])
# similar to difflib.SequenceMatcher.get_matching_blocks
def get_matching_blocks(a, b):
return [(d[0], d[2], d[1] - d[0]) for d in bdiff.blocks(a, b)]
def trivialdiffheader(length):
return struct.pack(">lll", 0, 0, length) if length else ''
def replacediffheader(oldlen, newlen):
return struct.pack(">lll", 0, oldlen, newlen)