view mercurial/sshpeer.py @ 35793:4fb2bb61597c

bundle2: increase payload part chunk size to 32kb

Bundle2 payload parts are framed chunks. Esentially, we obtain
data in equal size chunks of size `preferedchunksize` and emit those
to a generator. That generator is fed into a compressor (which can
be the no-op compressor, which just re-emits the generator). And
the output from the compressor likely goes to a file descriptor
or socket.

What this means is that small chunk sizes create more Python objects
and Python function calls than larger chunk sizes. And as we know,
Python object and function call overhead in performance sensitive
code matters (at least with CPython).

This commit increases the bundle2 part payload chunk size from 4k
to 32k. Practically speaking, this means that the chunks we feed
into a compressor (implemented in C code) or feed directly into a
file handle or socket write() are larger. It's possible the chunks
might be larger than what the receiver can handle in one logical
operation. But at that point, we're in C code, which is much more
efficient at dealing with splitting up the chunk and making multiple
function calls than Python is.

A downside to larger chunks is that the receiver has to wait for that
much data to arrive (either raw or from a decompressor) before it
can process the chunk. But 32kb still feels like a small buffer to
have to wait for. And in many cases, the client will convert from
8 read(4096) to 1 read(32768). That's happening in Python land. So
we cut down on the number of Python objects and function calls,
making the client faster as well. I don't think there are any
significant concerns to increasing the payload chunk size to 32kb.

The impact of this change on performance significant. Using `curl`
to obtain a stream clone bundle2 payload from a server on localhost
serving the mozilla-unified repository:

before: 20.78 user; 7.71 system;  80.5 MB/s
after:  13.90 user; 3.51 system; 132 MB/s
legacy:  9.72 user; 8.16 system; 132 MB/s

bundle2 stream clone generation is still more resource intensive than
legacy stream clone (that's likely because of the use of a
util.chunkbuffer). But the throughput is the same. We might
be in territory we're this is effectively a benchmark of the
networking stack or Python's syscall throughput.

From the client perspective, `hg clone -U --stream`:

before: 33.50 user; 7.95 system; 53.3 MB/s
after:  22.82 user; 7.33 system; 72.7 MB/s
legacy: 29.96 user; 7.94 system; 58.0 MB/s

And for `hg clone --stream` with a working directory update of
~230k files:

after:  119.55 user; 26.47 system; 0:57.08 wall
legacy: 126.98 user; 26.94 system; 1:05.56 wall

So, it appears that bundle2's stream clone is now definitively faster
than legacy stream clone!

Differential Revision: https://phab.mercurial-scm.org/D1932

# sshpeer.py - ssh repository proxy class for mercurial
#
# Copyright 2005, 2006 Matt Mackall <[e-mail]>
#
# 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

from .i18n import _
from . import (
    error,
    pycompat,
    util,
    wireproto,
)

def _serverquote(s):
    """quote a string for the remote shell ... which we assume is sh"""
    if not s:
        return s
    if re.match('[a-zA-Z0-9@%_+=:,./-]*$', s):
        return s
    return "'%s'" % s.replace("'", "'\\''")

def _forwardoutput(ui, pipe):
    """display all data currently available on pipe as remote output.

    This is non blocking."""
    s = util.readpipe(pipe)
    if s:
        for l in s.splitlines():
            ui.status(_("remote: "), l, '\n')

class doublepipe(object):
    """Operate a side-channel pipe in addition of a main one

    The side-channel pipe contains server output to be forwarded to the user
    input. The double pipe will behave as the "main" pipe, but will ensure the
    content of the "side" pipe is properly processed while we wait for blocking
    call on the "main" pipe.

    If large amounts of data are read from "main", the forward will cease after
    the first bytes start to appear. This simplifies the implementation
    without affecting actual output of sshpeer too much as we rarely issue
    large read for data not yet emitted by the server.

    The main pipe is expected to be a 'bufferedinputpipe' from the util module
    that handle all the os specific bits. This class lives in this module
    because it focus on behavior specific to the ssh protocol."""

    def __init__(self, ui, main, side):
        self._ui = ui
        self._main = main
        self._side = side

    def _wait(self):
        """wait until some data are available on main or side

        return a pair of boolean (ismainready, issideready)

        (This will only wait for data if the setup is supported by `util.poll`)
        """
        if getattr(self._main, 'hasbuffer', False): # getattr for classic pipe
            return (True, True) # main has data, assume side is worth poking at.
        fds = [self._main.fileno(), self._side.fileno()]
        try:
            act = util.poll(fds)
        except NotImplementedError:
            # non supported yet case, assume all have data.
            act = fds
        return (self._main.fileno() in act, self._side.fileno() in act)

    def write(self, data):
        return self._call('write', data)

    def read(self, size):
        r = self._call('read', size)
        if size != 0 and not r:
            # We've observed a condition that indicates the
            # stdout closed unexpectedly. Check stderr one
            # more time and snag anything that's there before
            # letting anyone know the main part of the pipe
            # closed prematurely.
            _forwardoutput(self._ui, self._side)
        return r

    def readline(self):
        return self._call('readline')

    def _call(self, methname, data=None):
        """call <methname> on "main", forward output of "side" while blocking
        """
        # data can be '' or 0
        if (data is not None and not data) or self._main.closed:
            _forwardoutput(self._ui, self._side)
            return ''
        while True:
            mainready, sideready = self._wait()
            if sideready:
                _forwardoutput(self._ui, self._side)
            if mainready:
                meth = getattr(self._main, methname)
                if data is None:
                    return meth()
                else:
                    return meth(data)

    def close(self):
        return self._main.close()

    def flush(self):
        return self._main.flush()

class sshpeer(wireproto.wirepeer):
    def __init__(self, ui, path, create=False):
        self._url = path
        self._ui = ui
        self._pipeo = self._pipei = self._pipee = None

        u = util.url(path, parsequery=False, parsefragment=False)
        if u.scheme != 'ssh' or not u.host or u.path is None:
            self._abort(error.RepoError(_("couldn't parse location %s") % path))

        util.checksafessh(path)

        if u.passwd is not None:
            self._abort(error.RepoError(_("password in URL not supported")))

        self._user = u.user
        self._host = u.host
        self._port = u.port
        self._path = u.path or '.'

        sshcmd = self.ui.config("ui", "ssh")
        remotecmd = self.ui.config("ui", "remotecmd")
        sshaddenv = dict(self.ui.configitems("sshenv"))
        sshenv = util.shellenviron(sshaddenv)

        args = util.sshargs(sshcmd, self._host, self._user, self._port)

        if create:
            cmd = '%s %s %s' % (sshcmd, args,
                util.shellquote("%s init %s" %
                    (_serverquote(remotecmd), _serverquote(self._path))))
            ui.debug('running %s\n' % cmd)
            res = ui.system(cmd, blockedtag='sshpeer', environ=sshenv)
            if res != 0:
                self._abort(error.RepoError(_("could not create remote repo")))

        self._validaterepo(sshcmd, args, remotecmd, sshenv)

    # Begin of _basepeer interface.

    @util.propertycache
    def ui(self):
        return self._ui

    def url(self):
        return self._url

    def local(self):
        return None

    def peer(self):
        return self

    def canpush(self):
        return True

    def close(self):
        pass

    # End of _basepeer interface.

    # Begin of _basewirecommands interface.

    def capabilities(self):
        return self._caps

    # End of _basewirecommands interface.

    def _validaterepo(self, sshcmd, args, remotecmd, sshenv=None):
        # cleanup up previous run
        self._cleanup()

        cmd = '%s %s %s' % (sshcmd, args,
            util.shellquote("%s -R %s serve --stdio" %
                (_serverquote(remotecmd), _serverquote(self._path))))
        self.ui.debug('running %s\n' % cmd)
        cmd = util.quotecommand(cmd)

        # while self._subprocess isn't used, having it allows the subprocess to
        # to clean up correctly later
        #
        # no buffer allow the use of 'select'
        # feel free to remove buffering and select usage when we ultimately
        # move to threading.
        sub = util.popen4(cmd, bufsize=0, env=sshenv)
        self._pipeo, self._pipei, self._pipee, self._subprocess = sub

        self._pipei = util.bufferedinputpipe(self._pipei)
        self._pipei = doublepipe(self.ui, self._pipei, self._pipee)
        self._pipeo = doublepipe(self.ui, self._pipeo, self._pipee)

        def badresponse():
            msg = _("no suitable response from remote hg")
            hint = self.ui.config("ui", "ssherrorhint")
            self._abort(error.RepoError(msg, hint=hint))

        try:
            # skip any noise generated by remote shell
            self._callstream("hello")
            r = self._callstream("between", pairs=("%s-%s" % ("0"*40, "0"*40)))
        except IOError:
            badresponse()

        lines = ["", "dummy"]
        max_noise = 500
        while lines[-1] and max_noise:
            try:
                l = r.readline()
                self._readerr()
                if lines[-1] == "1\n" and l == "\n":
                    break
                if l:
                    self.ui.debug("remote: ", l)
                lines.append(l)
                max_noise -= 1
            except IOError:
                badresponse()
        else:
            badresponse()

        self._caps = set()
        for l in reversed(lines):
            if l.startswith("capabilities:"):
                self._caps.update(l[:-1].split(":")[1].split())
                break

    def _readerr(self):
        _forwardoutput(self.ui, self._pipee)

    def _abort(self, exception):
        self._cleanup()
        raise exception

    def _cleanup(self):
        if self._pipeo is None:
            return
        self._pipeo.close()
        self._pipei.close()
        try:
            # read the error descriptor until EOF
            for l in self._pipee:
                self.ui.status(_("remote: "), l)
        except (IOError, ValueError):
            pass
        self._pipee.close()

    __del__ = _cleanup

    def _submitbatch(self, req):
        rsp = self._callstream("batch", cmds=wireproto.encodebatchcmds(req))
        available = self._getamount()
        # TODO this response parsing is probably suboptimal for large
        # batches with large responses.
        toread = min(available, 1024)
        work = rsp.read(toread)
        available -= toread
        chunk = work
        while chunk:
            while ';' in work:
                one, work = work.split(';', 1)
                yield wireproto.unescapearg(one)
            toread = min(available, 1024)
            chunk = rsp.read(toread)
            available -= toread
            work += chunk
        yield wireproto.unescapearg(work)

    def _callstream(self, cmd, **args):
        args = pycompat.byteskwargs(args)
        if (self.ui.debugflag
            and self.ui.configbool('devel', 'debug.peer-request')):
            dbg = self.ui.debug
            line = 'devel-peer-request: %s\n'
            dbg(line % cmd)
            for key, value in sorted(args.items()):
                if not isinstance(value, dict):
                    dbg(line % '  %s: %d bytes' % (key, len(value)))
                else:
                    for dk, dv in sorted(value.items()):
                        dbg(line % '  %s-%s: %d' % (key, dk, len(dv)))
        self.ui.debug("sending %s command\n" % cmd)
        self._pipeo.write("%s\n" % cmd)
        _func, names = wireproto.commands[cmd]
        keys = names.split()
        wireargs = {}
        for k in keys:
            if k == '*':
                wireargs['*'] = args
                break
            else:
                wireargs[k] = args[k]
                del args[k]
        for k, v in sorted(wireargs.iteritems()):
            self._pipeo.write("%s %d\n" % (k, len(v)))
            if isinstance(v, dict):
                for dk, dv in v.iteritems():
                    self._pipeo.write("%s %d\n" % (dk, len(dv)))
                    self._pipeo.write(dv)
            else:
                self._pipeo.write(v)
        self._pipeo.flush()

        return self._pipei

    def _callcompressable(self, cmd, **args):
        return self._callstream(cmd, **args)

    def _call(self, cmd, **args):
        self._callstream(cmd, **args)
        return self._recv()

    def _callpush(self, cmd, fp, **args):
        r = self._call(cmd, **args)
        if r:
            return '', r
        for d in iter(lambda: fp.read(4096), ''):
            self._send(d)
        self._send("", flush=True)
        r = self._recv()
        if r:
            return '', r
        return self._recv(), ''

    def _calltwowaystream(self, cmd, fp, **args):
        r = self._call(cmd, **args)
        if r:
            # XXX needs to be made better
            raise error.Abort(_('unexpected remote reply: %s') % r)
        for d in iter(lambda: fp.read(4096), ''):
            self._send(d)
        self._send("", flush=True)
        return self._pipei

    def _getamount(self):
        l = self._pipei.readline()
        if l == '\n':
            self._readerr()
            msg = _('check previous remote output')
            self._abort(error.OutOfBandError(hint=msg))
        self._readerr()
        try:
            return int(l)
        except ValueError:
            self._abort(error.ResponseError(_("unexpected response:"), l))

    def _recv(self):
        return self._pipei.read(self._getamount())

    def _send(self, data, flush=False):
        self._pipeo.write("%d\n" % len(data))
        if data:
            self._pipeo.write(data)
        if flush:
            self._pipeo.flush()
        self._readerr()

instance = sshpeer