##// END OF EJS Templates
procutil: avoid using os.fork() to implement runbgcommand...
procutil: avoid using os.fork() to implement runbgcommand We ran into the following deadlock: - some command creates an ssh peer, then raises without explicitly closing the peer (hg id + extension in our case) - dispatch catches the exception, calls ui.log('commandfinish', ..) (the sshpeer is still not closed), which calls logtoprocess, which calls procutil.runbgcommand. - in the child of runbgcommand's fork(), between the fork and the exec, the opening of file descriptors triggers a gc which runs the destructor for sshpeer, which waits on ssh's stderr being closed, which never happens since ssh's stderr is held open by the parent of the fork where said destructor hasn't run Remotefilelog appears to have a hack around this deadlock as well. I don't know if there's more subtlety to it, because even though the problem is determistic, it is very fragile, so I didn't manage to reduce it. I can imagine three ways of tackling this problem: 1. don't run any python between fork and exec in runbgcommand 2. make the finalizer harmless after the fork 3. close the peer without relying on gc behavior This commit goes with 1, as forking without exec'ing is tricky in general in a language with gc finalizers. And maybe it's better in the presence of rust threads. A future commit will try 2 or 3. Performance wise: at low memory usage, it's an improvement. At higher memory usage, it's about 2x faster than before when ensurestart=True, but 2x slower when ensurestart=False. Not sure if that matters. The reason for that last bit is that the subprocess.Popen always waits for the execve to finish, and at high memory usage, execve is slow because it deallocates the large page table. Numbers and script: before after mem=1.0GB, ensurestart=True 52.1ms 26.0ms mem=1.0GB, ensurestart=False 14.7ms 26.0ms mem=0.5GB, ensurestart=True 23.2ms 11.2ms mem=0.5GB, ensurestart=False 6.2ms 11.3ms mem=0.2GB, ensurestart=True 15.7ms 7.4ms mem=0.2GB, ensurestart=False 4.3ms 8.1ms mem=0.0GB, ensurestart=True 2.3ms 0.7ms mem=0.0GB, ensurestart=False 0.8ms 0.8ms import time for memsize in [1_000_000_000, 500_000_000, 250_000_000, 0]: mem = 'a' * memsize for ensurestart in [True, False]: now = time.time() n = 100 for i in range(n): procutil.runbgcommand([b'true'], {}, ensurestart=ensurestart) after = time.time() ms = (after - now) / float(n) * 1000 print(f'mem={memsize / 1e9:.1f}GB, ensurestart={ensurestart} -> {ms:.1f}ms') Differential Revision: https://phab.mercurial-scm.org/D9019

File last commit:

r47575:d4ba4d51 default
r47651:8759e22f default
Show More
server.py
441 lines | 13.7 KiB | text/x-python | PythonLexer
# hgweb/server.py - The standalone hg web server.
#
# Copyright 21 May 2005 - (c) 2005 Jake Edge <jake@edge2.net>
# Copyright 2005-2007 Olivia Mackall <olivia@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 errno
import importlib
import os
import socket
import sys
import traceback
import wsgiref.validate
from ..i18n import _
from ..pycompat import (
getattr,
open,
)
from .. import (
encoding,
error,
pycompat,
util,
)
httpservermod = util.httpserver
socketserver = util.socketserver
urlerr = util.urlerr
urlreq = util.urlreq
from . import common
def _splitURI(uri):
"""Return path and query that has been split from uri
Just like CGI environment, the path is unquoted, the query is
not.
"""
if '?' in uri:
path, query = uri.split('?', 1)
else:
path, query = uri, r''
return urlreq.unquote(path), query
class _error_logger(object):
def __init__(self, handler):
self.handler = handler
def flush(self):
pass
def write(self, str):
self.writelines(str.split(b'\n'))
def writelines(self, seq):
for msg in seq:
self.handler.log_error("HG error: %s", encoding.strfromlocal(msg))
class _httprequesthandler(httpservermod.basehttprequesthandler):
url_scheme = b'http'
@staticmethod
def preparehttpserver(httpserver, ui):
"""Prepare .socket of new HTTPServer instance"""
def __init__(self, *args, **kargs):
self.protocol_version = r'HTTP/1.1'
httpservermod.basehttprequesthandler.__init__(self, *args, **kargs)
def _log_any(self, fp, format, *args):
fp.write(
pycompat.sysbytes(
r"%s - - [%s] %s"
% (
self.client_address[0],
self.log_date_time_string(),
format % args,
)
)
+ b'\n'
)
fp.flush()
def log_error(self, format, *args):
self._log_any(self.server.errorlog, format, *args)
def log_message(self, format, *args):
self._log_any(self.server.accesslog, format, *args)
def log_request(self, code='-', size='-'):
xheaders = []
if util.safehasattr(self, b'headers'):
xheaders = [
h for h in self.headers.items() if h[0].startswith('x-')
]
self.log_message(
'"%s" %s %s%s',
self.requestline,
str(code),
str(size),
''.join([' %s:%s' % h for h in sorted(xheaders)]),
)
def do_write(self):
try:
self.do_hgweb()
except socket.error as inst:
if inst.errno != errno.EPIPE:
raise
def do_POST(self):
try:
self.do_write()
except Exception as e:
# I/O below could raise another exception. So log the original
# exception first to ensure it is recorded.
if not (
isinstance(e, (OSError, socket.error))
and e.errno == errno.ECONNRESET
):
tb = "".join(traceback.format_exception(*sys.exc_info()))
# We need a native-string newline to poke in the log
# message, because we won't get a newline when using an
# r-string. This is the easy way out.
newline = chr(10)
self.log_error(
r"Exception happened during processing "
"request '%s':%s%s",
self.path,
newline,
tb,
)
self._start_response("500 Internal Server Error", [])
self._write(b"Internal Server Error")
self._done()
def do_PUT(self):
self.do_POST()
def do_GET(self):
self.do_POST()
def do_hgweb(self):
self.sent_headers = False
path, query = _splitURI(self.path)
# Ensure the slicing of path below is valid
if path != self.server.prefix and not path.startswith(
self.server.prefix + b'/'
):
self._start_response(pycompat.strurl(common.statusmessage(404)), [])
if self.command == 'POST':
# Paranoia: tell the client we're going to close the
# socket so they don't try and reuse a socket that
# might have a POST body waiting to confuse us. We do
# this by directly munging self.saved_headers because
# self._start_response ignores Connection headers.
self.saved_headers = [('Connection', 'Close')]
self._write(b"Not Found")
self._done()
return
env = {}
env['GATEWAY_INTERFACE'] = 'CGI/1.1'
env['REQUEST_METHOD'] = self.command
env['SERVER_NAME'] = self.server.server_name
env['SERVER_PORT'] = str(self.server.server_port)
env['REQUEST_URI'] = self.path
env['SCRIPT_NAME'] = pycompat.sysstr(self.server.prefix)
env['PATH_INFO'] = pycompat.sysstr(path[len(self.server.prefix) :])
env['REMOTE_HOST'] = self.client_address[0]
env['REMOTE_ADDR'] = self.client_address[0]
env['QUERY_STRING'] = query or ''
if pycompat.ispy3:
if self.headers.get_content_type() is None:
env['CONTENT_TYPE'] = self.headers.get_default_type()
else:
env['CONTENT_TYPE'] = self.headers.get_content_type()
length = self.headers.get('content-length')
else:
if self.headers.typeheader is None:
env['CONTENT_TYPE'] = self.headers.type
else:
env['CONTENT_TYPE'] = self.headers.typeheader
length = self.headers.getheader('content-length')
if length:
env['CONTENT_LENGTH'] = length
for header in [
h
for h in self.headers.keys()
if h.lower() not in ('content-type', 'content-length')
]:
hkey = 'HTTP_' + header.replace('-', '_').upper()
hval = self.headers.get(header)
hval = hval.replace('\n', '').strip()
if hval:
env[hkey] = hval
env['SERVER_PROTOCOL'] = self.request_version
env['wsgi.version'] = (1, 0)
env['wsgi.url_scheme'] = pycompat.sysstr(self.url_scheme)
if env.get('HTTP_EXPECT', b'').lower() == b'100-continue':
self.rfile = common.continuereader(self.rfile, self.wfile.write)
env['wsgi.input'] = self.rfile
env['wsgi.errors'] = _error_logger(self)
env['wsgi.multithread'] = isinstance(
self.server, socketserver.ThreadingMixIn
)
if util.safehasattr(socketserver, b'ForkingMixIn'):
env['wsgi.multiprocess'] = isinstance(
self.server, socketserver.ForkingMixIn
)
else:
env['wsgi.multiprocess'] = False
env['wsgi.run_once'] = 0
wsgiref.validate.check_environ(env)
self.saved_status = None
self.saved_headers = []
self.length = None
self._chunked = None
for chunk in self.server.application(env, self._start_response):
self._write(chunk)
if not self.sent_headers:
self.send_headers()
self._done()
def send_headers(self):
if not self.saved_status:
raise AssertionError(
b"Sending headers before start_response() called"
)
saved_status = self.saved_status.split(None, 1)
saved_status[0] = int(saved_status[0])
self.send_response(*saved_status)
self.length = None
self._chunked = False
for h in self.saved_headers:
self.send_header(*h)
if h[0].lower() == 'content-length':
self.length = int(h[1])
if self.length is None and saved_status[0] != common.HTTP_NOT_MODIFIED:
self._chunked = (
not self.close_connection and self.request_version == 'HTTP/1.1'
)
if self._chunked:
self.send_header('Transfer-Encoding', 'chunked')
else:
self.send_header('Connection', 'close')
self.end_headers()
self.sent_headers = True
def _start_response(self, http_status, headers, exc_info=None):
assert isinstance(http_status, str)
code, msg = http_status.split(None, 1)
code = int(code)
self.saved_status = http_status
bad_headers = ('connection', 'transfer-encoding')
self.saved_headers = [
h for h in headers if h[0].lower() not in bad_headers
]
return self._write
def _write(self, data):
if not self.saved_status:
raise AssertionError(b"data written before start_response() called")
elif not self.sent_headers:
self.send_headers()
if self.length is not None:
if len(data) > self.length:
raise AssertionError(
b"Content-length header sent, but more "
b"bytes than specified are being written."
)
self.length = self.length - len(data)
elif self._chunked and data:
data = b'%x\r\n%s\r\n' % (len(data), data)
self.wfile.write(data)
self.wfile.flush()
def _done(self):
if self._chunked:
self.wfile.write(b'0\r\n\r\n')
self.wfile.flush()
def version_string(self):
if self.server.serverheader:
return encoding.strfromlocal(self.server.serverheader)
return httpservermod.basehttprequesthandler.version_string(self)
class _httprequesthandlerssl(_httprequesthandler):
"""HTTPS handler based on Python's ssl module"""
url_scheme = b'https'
@staticmethod
def preparehttpserver(httpserver, ui):
try:
from .. import sslutil
sslutil.wrapserversocket
except ImportError:
raise error.Abort(_(b"SSL support is unavailable"))
certfile = ui.config(b'web', b'certificate')
# These config options are currently only meant for testing. Use
# at your own risk.
cafile = ui.config(b'devel', b'servercafile')
reqcert = ui.configbool(b'devel', b'serverrequirecert')
httpserver.socket = sslutil.wrapserversocket(
httpserver.socket,
ui,
certfile=certfile,
cafile=cafile,
requireclientcert=reqcert,
)
def setup(self):
self.connection = self.request
self.rfile = self.request.makefile("rb", self.rbufsize)
self.wfile = self.request.makefile("wb", self.wbufsize)
try:
import threading
threading.activeCount() # silence pyflakes and bypass demandimport
_mixin = socketserver.ThreadingMixIn
except ImportError:
if util.safehasattr(os, b"fork"):
_mixin = socketserver.ForkingMixIn
else:
class _mixin(object):
pass
def openlog(opt, default):
if opt and opt != b'-':
return open(opt, b'ab')
return default
class MercurialHTTPServer(_mixin, httpservermod.httpserver, object):
# SO_REUSEADDR has broken semantics on windows
if pycompat.iswindows:
allow_reuse_address = 0
def __init__(self, ui, app, addr, handler, **kwargs):
httpservermod.httpserver.__init__(self, addr, handler, **kwargs)
self.daemon_threads = True
self.application = app
handler.preparehttpserver(self, ui)
prefix = ui.config(b'web', b'prefix')
if prefix:
prefix = b'/' + prefix.strip(b'/')
self.prefix = prefix
alog = openlog(ui.config(b'web', b'accesslog'), ui.fout)
elog = openlog(ui.config(b'web', b'errorlog'), ui.ferr)
self.accesslog = alog
self.errorlog = elog
self.addr, self.port = self.socket.getsockname()[0:2]
self.fqaddr = self.server_name
self.serverheader = ui.config(b'web', b'server-header')
class IPv6HTTPServer(MercurialHTTPServer):
address_family = getattr(socket, 'AF_INET6', None)
def __init__(self, *args, **kwargs):
if self.address_family is None:
raise error.RepoError(_(b'IPv6 is not available on this system'))
super(IPv6HTTPServer, self).__init__(*args, **kwargs)
def create_server(ui, app):
if ui.config(b'web', b'certificate'):
handler = _httprequesthandlerssl
else:
handler = _httprequesthandler
if ui.configbool(b'web', b'ipv6'):
cls = IPv6HTTPServer
else:
cls = MercurialHTTPServer
# ugly hack due to python issue5853 (for threaded use)
try:
import mimetypes
mimetypes.init()
except UnicodeDecodeError:
# Python 2.x's mimetypes module attempts to decode strings
# from Windows' ANSI APIs as ascii (fail), then re-encode them
# as ascii (clown fail), because the default Python Unicode
# codec is hardcoded as ascii.
sys.argv # unwrap demand-loader so that reload() works
# resurrect sys.setdefaultencoding()
try:
importlib.reload(sys)
except AttributeError:
reload(sys)
oldenc = sys.getdefaultencoding()
sys.setdefaultencoding(b"latin1") # or any full 8-bit encoding
mimetypes.init()
sys.setdefaultencoding(oldenc)
address = ui.config(b'web', b'address')
port = util.getport(ui.config(b'web', b'port'))
try:
return cls(ui, app, (address, port), handler)
except socket.error as inst:
raise error.Abort(
_(b"cannot start server at '%s:%d': %s")
% (address, port, encoding.strtolocal(inst.args[1]))
)