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
# __init__.py - Startup and module loading logic for Mercurial.
#
# Copyright 2015 Gregory Szorc <gregory.szorc@gmail.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 sys
# Allow 'from mercurial import demandimport' to keep working.
import hgdemandimport
demandimport = hgdemandimport
__all__ = []
# Python 3 uses a custom module loader that transforms source code between
# source file reading and compilation. This is done by registering a custom
# finder that changes the spec for Mercurial modules to use a custom loader.
if sys.version_info[0] >= 3:
import importlib
import importlib.abc
import io
import token
import tokenize
class hgpathentryfinder(importlib.abc.MetaPathFinder):
"""A sys.meta_path finder that uses a custom module loader."""
def find_spec(self, fullname, path, target=None):
# Only handle Mercurial-related modules.
if not fullname.startswith(('mercurial.', 'hgext.', 'hgext3rd.')):
return None
# don't try to parse binary
if fullname.startswith('mercurial.cext.'):
return None
# third-party packages are expected to be dual-version clean
if fullname.startswith('mercurial.thirdparty'):
return None
# zstd is already dual-version clean, don't try and mangle it
if fullname.startswith('mercurial.zstd'):
return None
# pywatchman is already dual-version clean, don't try and mangle it
if fullname.startswith('hgext.fsmonitor.pywatchman'):
return None
# Try to find the module using other registered finders.
spec = None
for finder in sys.meta_path:
if finder == self:
continue
spec = finder.find_spec(fullname, path, target=target)
if spec:
break
# This is a Mercurial-related module but we couldn't find it
# using the previously-registered finders. This likely means
# the module doesn't exist.
if not spec:
return None
# TODO need to support loaders from alternate specs, like zip
# loaders.
loader = hgloader(spec.name, spec.origin)
# Can't use util.safehasattr here because that would require
# importing util, and we're in import code.
if hasattr(spec.loader, 'loader'): # hasattr-py3-only
# This is a nested loader (maybe a lazy loader?)
spec.loader.loader = loader
else:
spec.loader = loader
return spec
def replacetokens(tokens, fullname):
"""Transform a stream of tokens from raw to Python 3.
It is called by the custom module loading machinery to rewrite
source/tokens between source decoding and compilation.
Returns a generator of possibly rewritten tokens.
The input token list may be mutated as part of processing. However,
its changes do not necessarily match the output token stream.
REMEMBER TO CHANGE ``BYTECODEHEADER`` WHEN CHANGING THIS FUNCTION
OR CACHED FILES WON'T GET INVALIDATED PROPERLY.
"""
futureimpline = False
# The following utility functions access the tokens list and i index of
# the for i, t enumerate(tokens) loop below
def _isop(j, *o):
"""Assert that tokens[j] is an OP with one of the given values"""
try:
return tokens[j].type == token.OP and tokens[j].string in o
except IndexError:
return False
def _findargnofcall(n):
"""Find arg n of a call expression (start at 0)
Returns index of the first token of that argument, or None if
there is not that many arguments.
Assumes that token[i + 1] is '('.
"""
nested = 0
for j in range(i + 2, len(tokens)):
if _isop(j, ')', ']', '}'):
# end of call, tuple, subscription or dict / set
nested -= 1
if nested < 0:
return None
elif n == 0:
# this is the starting position of arg
return j
elif _isop(j, '(', '[', '{'):
nested += 1
elif _isop(j, ',') and nested == 0:
n -= 1
return None
def _ensureunicode(j):
"""Make sure the token at j is a unicode string
This rewrites a string token to include the unicode literal prefix
so the string transformer won't add the byte prefix.
Ignores tokens that are not strings. Assumes bounds checking has
already been done.
"""
st = tokens[j]
if st.type == token.STRING and st.string.startswith(("'", '"')):
tokens[j] = st._replace(string='u%s' % st.string)
for i, t in enumerate(tokens):
# Convert most string literals to byte literals. String literals
# in Python 2 are bytes. String literals in Python 3 are unicode.
# Most strings in Mercurial are bytes and unicode strings are rare.
# Rather than rewrite all string literals to use ``b''`` to indicate
# byte strings, we apply this token transformer to insert the ``b``
# prefix nearly everywhere.
if t.type == token.STRING:
s = t.string
# Preserve docstrings as string literals. This is inconsistent
# with regular unprefixed strings. However, the
# "from __future__" parsing (which allows a module docstring to
# exist before it) doesn't properly handle the docstring if it
# is b''' prefixed, leading to a SyntaxError. We leave all
# docstrings as unprefixed to avoid this. This means Mercurial
# components touching docstrings need to handle unicode,
# unfortunately.
if s[0:3] in ("'''", '"""'):
yield t
continue
# If the first character isn't a quote, it is likely a string
# prefixing character (such as 'b', 'u', or 'r'. Ignore.
if s[0] not in ("'", '"'):
yield t
continue
# String literal. Prefix to make a b'' string.
yield t._replace(string='b%s' % t.string)
continue
# Insert compatibility imports at "from __future__ import" line.
# No '\n' should be added to preserve line numbers.
if (t.type == token.NAME and t.string == 'import' and
all(u.type == token.NAME for u in tokens[i - 2:i]) and
[u.string for u in tokens[i - 2:i]] == ['from', '__future__']):
futureimpline = True
if t.type == token.NEWLINE and futureimpline:
futureimpline = False
if fullname == 'mercurial.pycompat':
yield t
continue
r, c = t.start
l = (b'; from mercurial.pycompat import '
b'delattr, getattr, hasattr, setattr, xrange, '
b'open, unicode\n')
for u in tokenize.tokenize(io.BytesIO(l).readline):
if u.type in (tokenize.ENCODING, token.ENDMARKER):
continue
yield u._replace(
start=(r, c + u.start[1]), end=(r, c + u.end[1]))
continue
# This looks like a function call.
if t.type == token.NAME and _isop(i + 1, '('):
fn = t.string
# *attr() builtins don't accept byte strings to 2nd argument.
if (fn in ('getattr', 'setattr', 'hasattr', 'safehasattr') and
not _isop(i - 1, '.')):
arg1idx = _findargnofcall(1)
if arg1idx is not None:
_ensureunicode(arg1idx)
# .encode() and .decode() on str/bytes/unicode don't accept
# byte strings on Python 3.
elif fn in ('encode', 'decode') and _isop(i - 1, '.'):
for argn in range(2):
argidx = _findargnofcall(argn)
if argidx is not None:
_ensureunicode(argidx)
# It changes iteritems/values to items/values as they are not
# present in Python 3 world.
elif fn in ('iteritems', 'itervalues'):
yield t._replace(string=fn[4:])
continue
# Emit unmodified token.
yield t
# Header to add to bytecode files. This MUST be changed when
# ``replacetoken`` or any mechanism that changes semantics of module
# loading is changed. Otherwise cached bytecode may get loaded without
# the new transformation mechanisms applied.
BYTECODEHEADER = b'HG\x00\x0a'
class hgloader(importlib.machinery.SourceFileLoader):
"""Custom module loader that transforms source code.
When the source code is converted to a code object, we transform
certain patterns to be Python 3 compatible. This allows us to write code
that is natively Python 2 and compatible with Python 3 without
making the code excessively ugly.
We do this by transforming the token stream between parse and compile.
Implementing transformations invalidates caching assumptions made
by the built-in importer. The built-in importer stores a header on
saved bytecode files indicating the Python/bytecode version. If the
version changes, the cached bytecode is ignored. The Mercurial
transformations could change at any time. This means we need to check
that cached bytecode was generated with the current transformation
code or there could be a mismatch between cached bytecode and what
would be generated from this class.
We supplement the bytecode caching layer by wrapping ``get_data``
and ``set_data``. These functions are called when the
``SourceFileLoader`` retrieves and saves bytecode cache files,
respectively. We simply add an additional header on the file. As
long as the version in this file is changed when semantics change,
cached bytecode should be invalidated when transformations change.
The added header has the form ``HG<VERSION>``. That is a literal
``HG`` with 2 binary bytes indicating the transformation version.
"""
def get_data(self, path):
data = super(hgloader, self).get_data(path)
if not path.endswith(tuple(importlib.machinery.BYTECODE_SUFFIXES)):
return data
# There should be a header indicating the Mercurial transformation
# version. If it doesn't exist or doesn't match the current version,
# we raise an OSError because that is what
# ``SourceFileLoader.get_code()`` expects when loading bytecode
# paths to indicate the cached file is "bad."
if data[0:2] != b'HG':
raise OSError('no hg header')
if data[0:4] != BYTECODEHEADER:
raise OSError('hg header version mismatch')
return data[4:]
def set_data(self, path, data, *args, **kwargs):
if path.endswith(tuple(importlib.machinery.BYTECODE_SUFFIXES)):
data = BYTECODEHEADER + data
return super(hgloader, self).set_data(path, data, *args, **kwargs)
def source_to_code(self, data, path):
"""Perform token transformation before compilation."""
buf = io.BytesIO(data)
tokens = tokenize.tokenize(buf.readline)
data = tokenize.untokenize(replacetokens(list(tokens), self.name))
# Python's built-in importer strips frames from exceptions raised
# for this code. Unfortunately, that mechanism isn't extensible
# and our frame will be blamed for the import failure. There
# are extremely hacky ways to do frame stripping. We haven't
# implemented them because they are very ugly.
return super(hgloader, self).source_to_code(data, path)
# We automagically register our custom importer as a side-effect of
# loading. This is necessary to ensure that any entry points are able
# to import mercurial.* modules without having to perform this
# registration themselves.
if not any(isinstance(x, hgpathentryfinder) for x in sys.meta_path):
# meta_path is used before any implicit finders and before sys.path.
sys.meta_path.insert(0, hgpathentryfinder())