##// 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:

r37237:7f78de1c default
r37304:9bfcbe4f default
Show More
procutil.py
361 lines | 11.6 KiB | text/x-python | PythonLexer
# procutil.py - utility for managing processes and executable environment
#
# Copyright 2005 K. Thananchayan <thananck@yahoo.com>
# Copyright 2005-2007 Matt Mackall <mpm@selenic.com>
# Copyright 2006 Vadim Gelfer <vadim.gelfer@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 contextlib
import imp
import io
import os
import signal
import subprocess
import sys
import tempfile
import time
from ..i18n import _
from .. import (
encoding,
error,
policy,
pycompat,
)
osutil = policy.importmod(r'osutil')
stderr = pycompat.stderr
stdin = pycompat.stdin
stdout = pycompat.stdout
def isatty(fp):
try:
return fp.isatty()
except AttributeError:
return False
# glibc determines buffering on first write to stdout - if we replace a TTY
# destined stdout with a pipe destined stdout (e.g. pager), we want line
# buffering
if isatty(stdout):
stdout = os.fdopen(stdout.fileno(), r'wb', 1)
if pycompat.iswindows:
from .. import windows as platform
stdout = platform.winstdout(stdout)
else:
from .. import posix as platform
explainexit = platform.explainexit
findexe = platform.findexe
_gethgcmd = platform.gethgcmd
getuser = platform.getuser
getpid = os.getpid
hidewindow = platform.hidewindow
popen = platform.popen
quotecommand = platform.quotecommand
readpipe = platform.readpipe
setbinary = platform.setbinary
setsignalhandler = platform.setsignalhandler
shellquote = platform.shellquote
shellsplit = platform.shellsplit
spawndetached = platform.spawndetached
sshargs = platform.sshargs
testpid = platform.testpid
try:
setprocname = osutil.setprocname
except AttributeError:
pass
try:
unblocksignal = osutil.unblocksignal
except AttributeError:
pass
closefds = pycompat.isposix
def popen2(cmd, env=None, newlines=False):
# Setting bufsize to -1 lets the system decide the buffer size.
# The default for bufsize is 0, meaning unbuffered. This leads to
# poor performance on Mac OS X: http://bugs.python.org/issue4194
p = subprocess.Popen(cmd, shell=True, bufsize=-1,
close_fds=closefds,
stdin=subprocess.PIPE, stdout=subprocess.PIPE,
universal_newlines=newlines,
env=env)
return p.stdin, p.stdout
def popen3(cmd, env=None, newlines=False):
stdin, stdout, stderr, p = popen4(cmd, env, newlines)
return stdin, stdout, stderr
def popen4(cmd, env=None, newlines=False, bufsize=-1):
p = subprocess.Popen(cmd, shell=True, bufsize=bufsize,
close_fds=closefds,
stdin=subprocess.PIPE, stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
universal_newlines=newlines,
env=env)
return p.stdin, p.stdout, p.stderr, p
def pipefilter(s, cmd):
'''filter string S through command CMD, returning its output'''
p = subprocess.Popen(cmd, shell=True, close_fds=closefds,
stdin=subprocess.PIPE, stdout=subprocess.PIPE)
pout, perr = p.communicate(s)
return pout
def tempfilter(s, cmd):
'''filter string S through a pair of temporary files with CMD.
CMD is used as a template to create the real command to be run,
with the strings INFILE and OUTFILE replaced by the real names of
the temporary files generated.'''
inname, outname = None, None
try:
infd, inname = tempfile.mkstemp(prefix='hg-filter-in-')
fp = os.fdopen(infd, r'wb')
fp.write(s)
fp.close()
outfd, outname = tempfile.mkstemp(prefix='hg-filter-out-')
os.close(outfd)
cmd = cmd.replace('INFILE', inname)
cmd = cmd.replace('OUTFILE', outname)
code = os.system(cmd)
if pycompat.sysplatform == 'OpenVMS' and code & 1:
code = 0
if code:
raise error.Abort(_("command '%s' failed: %s") %
(cmd, explainexit(code)))
with open(outname, 'rb') as fp:
return fp.read()
finally:
try:
if inname:
os.unlink(inname)
except OSError:
pass
try:
if outname:
os.unlink(outname)
except OSError:
pass
_filtertable = {
'tempfile:': tempfilter,
'pipe:': pipefilter,
}
def filter(s, cmd):
"filter a string through a command that transforms its input to its output"
for name, fn in _filtertable.iteritems():
if cmd.startswith(name):
return fn(s, cmd[len(name):].lstrip())
return pipefilter(s, cmd)
def mainfrozen():
"""return True if we are a frozen executable.
The code supports py2exe (most common, Windows only) and tools/freeze
(portable, not much used).
"""
return (pycompat.safehasattr(sys, "frozen") or # new py2exe
pycompat.safehasattr(sys, "importers") or # old py2exe
imp.is_frozen(u"__main__")) # tools/freeze
_hgexecutable = None
def hgexecutable():
"""return location of the 'hg' executable.
Defaults to $HG or 'hg' in the search path.
"""
if _hgexecutable is None:
hg = encoding.environ.get('HG')
mainmod = sys.modules[r'__main__']
if hg:
_sethgexecutable(hg)
elif mainfrozen():
if getattr(sys, 'frozen', None) == 'macosx_app':
# Env variable set by py2app
_sethgexecutable(encoding.environ['EXECUTABLEPATH'])
else:
_sethgexecutable(pycompat.sysexecutable)
elif (os.path.basename(
pycompat.fsencode(getattr(mainmod, '__file__', ''))) == 'hg'):
_sethgexecutable(pycompat.fsencode(mainmod.__file__))
else:
exe = findexe('hg') or os.path.basename(sys.argv[0])
_sethgexecutable(exe)
return _hgexecutable
def _sethgexecutable(path):
"""set location of the 'hg' executable"""
global _hgexecutable
_hgexecutable = path
def _testfileno(f, stdf):
fileno = getattr(f, 'fileno', None)
try:
return fileno and fileno() == stdf.fileno()
except io.UnsupportedOperation:
return False # fileno() raised UnsupportedOperation
def isstdin(f):
return _testfileno(f, sys.__stdin__)
def isstdout(f):
return _testfileno(f, sys.__stdout__)
def protectstdio(uin, uout):
"""Duplicate streams and redirect original if (uin, uout) are stdio
If uin is stdin, it's redirected to /dev/null. If uout is stdout, it's
redirected to stderr so the output is still readable.
Returns (fin, fout) which point to the original (uin, uout) fds, but
may be copy of (uin, uout). The returned streams can be considered
"owned" in that print(), exec(), etc. never reach to them.
"""
uout.flush()
fin, fout = uin, uout
if uin is stdin:
newfd = os.dup(uin.fileno())
nullfd = os.open(os.devnull, os.O_RDONLY)
os.dup2(nullfd, uin.fileno())
os.close(nullfd)
fin = os.fdopen(newfd, r'rb')
if uout is stdout:
newfd = os.dup(uout.fileno())
os.dup2(stderr.fileno(), uout.fileno())
fout = os.fdopen(newfd, r'wb')
return fin, fout
def restorestdio(uin, uout, fin, fout):
"""Restore (uin, uout) streams from possibly duplicated (fin, fout)"""
uout.flush()
for f, uif in [(fin, uin), (fout, uout)]:
if f is not uif:
os.dup2(f.fileno(), uif.fileno())
f.close()
@contextlib.contextmanager
def protectedstdio(uin, uout):
"""Run code block with protected standard streams"""
fin, fout = protectstdio(uin, uout)
try:
yield fin, fout
finally:
restorestdio(uin, uout, fin, fout)
def shellenviron(environ=None):
"""return environ with optional override, useful for shelling out"""
def py2shell(val):
'convert python object into string that is useful to shell'
if val is None or val is False:
return '0'
if val is True:
return '1'
return pycompat.bytestr(val)
env = dict(encoding.environ)
if environ:
env.update((k, py2shell(v)) for k, v in environ.iteritems())
env['HG'] = hgexecutable()
return env
def system(cmd, environ=None, cwd=None, out=None):
'''enhanced shell command execution.
run with environment maybe modified, maybe in different dir.
if out is specified, it is assumed to be a file-like object that has a
write() method. stdout and stderr will be redirected to out.'''
try:
stdout.flush()
except Exception:
pass
cmd = quotecommand(cmd)
env = shellenviron(environ)
if out is None or isstdout(out):
rc = subprocess.call(cmd, shell=True, close_fds=closefds,
env=env, cwd=cwd)
else:
proc = subprocess.Popen(cmd, shell=True, close_fds=closefds,
env=env, cwd=cwd, stdout=subprocess.PIPE,
stderr=subprocess.STDOUT)
for line in iter(proc.stdout.readline, ''):
out.write(line)
proc.wait()
rc = proc.returncode
if pycompat.sysplatform == 'OpenVMS' and rc & 1:
rc = 0
return rc
def gui():
'''Are we running in a GUI?'''
if pycompat.isdarwin:
if 'SSH_CONNECTION' in encoding.environ:
# handle SSH access to a box where the user is logged in
return False
elif getattr(osutil, 'isgui', None):
# check if a CoreGraphics session is available
return osutil.isgui()
else:
# pure build; use a safe default
return True
else:
return pycompat.iswindows or encoding.environ.get("DISPLAY")
def hgcmd():
"""Return the command used to execute current hg
This is different from hgexecutable() because on Windows we want
to avoid things opening new shell windows like batch files, so we
get either the python call or current executable.
"""
if mainfrozen():
if getattr(sys, 'frozen', None) == 'macosx_app':
# Env variable set by py2app
return [encoding.environ['EXECUTABLEPATH']]
else:
return [pycompat.sysexecutable]
return _gethgcmd()
def rundetached(args, condfn):
"""Execute the argument list in a detached process.
condfn is a callable which is called repeatedly and should return
True once the child process is known to have started successfully.
At this point, the child process PID is returned. If the child
process fails to start or finishes before condfn() evaluates to
True, return -1.
"""
# Windows case is easier because the child process is either
# successfully starting and validating the condition or exiting
# on failure. We just poll on its PID. On Unix, if the child
# process fails to start, it will be left in a zombie state until
# the parent wait on it, which we cannot do since we expect a long
# running process on success. Instead we listen for SIGCHLD telling
# us our child process terminated.
terminated = set()
def handler(signum, frame):
terminated.add(os.wait())
prevhandler = None
SIGCHLD = getattr(signal, 'SIGCHLD', None)
if SIGCHLD is not None:
prevhandler = signal.signal(SIGCHLD, handler)
try:
pid = spawndetached(args)
while not condfn():
if ((pid in terminated or not testpid(pid))
and not condfn()):
return -1
time.sleep(0.1)
return pid
finally:
if prevhandler is not None:
signal.signal(signal.SIGCHLD, prevhandler)