~launchpad-pqm/launchpad/devel

#!/usr/bin/python -S

#

# Copyright 2009, 2010 Canonical Ltd.  This software is licensed under the

# GNU Affero General Public License version 3 (see the file LICENSE).

import _pythonpath

import logging

import os

import sys

import time

import traceback

from paste import httpserver

from paste.deploy.config import PrefixMiddleware

from paste.httpexceptions import HTTPExceptionHandler

from paste.request import construct_url

from paste.translogger import TransLogger

from paste.wsgilib import catch_errors

import lp.codehosting

from lp.services.config import config

LISTEN_HOST = config.codebrowse.listen_host

LISTEN_PORT = config.codebrowse.port

THREADPOOL_WORKERS = 10

class NoLockingFileHandler(logging.FileHandler):

    """A version of logging.FileHandler that doesn't do it's own locking.

    We experienced occasional hangs in production where gdb-ery on the server

    revealed that we sometimes end up with many threads blocking on the RLock

    held by the logging file handler, and log reading finds that an exception

    managed to kill a thread in an unsafe window for RLock's.

    Luckily, there's no real reason for us to take a lock during logging as

    each log message translates to one call to .write on a file object, which

    translates to one fwrite call, and it seems that this does enough locking

    itself for our purposes.

    So this handler just doesn't lock in log message handling.

    """

    def acquire(self):

        pass

    def release(self):

        pass

def setup_logging(home, foreground):

    # i hate that stupid logging config format, so just set up logging here.

    log_folder = config.codebrowse.log_folder

    if not log_folder:

        log_folder = os.path.join(home, 'logs')

    if not os.path.exists(log_folder):

        os.makedirs(log_folder)

    f = logging.Formatter(

        '%(levelname)-.3s [%(asctime)s.%(msecs)03d] [%(thread)d] %(name)s: %(message)s',

        '%Y%m%d-%H:%M:%S')

    debug_log = NoLockingFileHandler(os.path.join(log_folder, 'debug.log'))

    debug_log.setLevel(logging.DEBUG)

    debug_log.setFormatter(f)

    if foreground:

        stdout_log = logging.StreamHandler(sys.stdout)

        stdout_log.setLevel(logging.DEBUG)

        stdout_log.setFormatter(f)

    f = logging.Formatter('[%(asctime)s.%(msecs)03d] %(message)s',

                          '%Y%m%d-%H:%M:%S')

    access_log = NoLockingFileHandler(os.path.join(log_folder, 'access.log'))

    access_log.setLevel(logging.INFO)

    access_log.setFormatter(f)

    logging.getLogger('').setLevel(logging.DEBUG)

    logging.getLogger('').addHandler(debug_log)

    logging.getLogger('wsgi').addHandler(access_log)

    if foreground:

        logging.getLogger('').addHandler(stdout_log)

    else:

        class S(object):

            def write(self, str):

                logging.getLogger().error(str.rstrip('\n'))

            def flush(self):

                pass

        sys.stderr = S()

foreground = False

if len(sys.argv) > 1:

    if sys.argv[1] == '-f':

        foreground = True

home = os.path.realpath(os.path.dirname(__file__))

pidfile = os.path.join(home, 'loggerhead.pid')

if not foreground:

    sys.stderr.write('\n')

    sys.stderr.write('Launching loggerhead into the background.\n')

    sys.stderr.write('PID file: %s\n' % (pidfile,))

    sys.stderr.write('\n')

    from loggerhead.daemon import daemonize

    daemonize(pidfile, home)

setup_logging(home, foreground=foreground)

log = logging.getLogger('loggerhead')

log.info('Starting up...')

log.info('Loading the bzr plugins...')

from bzrlib.plugin import load_plugins

load_plugins()

import bzrlib.plugins

if getattr(bzrlib.plugins, 'loom', None) is None:

    log.error('Loom plugin loading failed.')

from launchpad_loggerhead.debug import (

    change_kill_thread_criteria, threadpool_debug)

from launchpad_loggerhead.app import RootApp, oops_middleware

from launchpad_loggerhead.session import SessionHandler

SESSION_VAR = 'lh.session'

secret = open(os.path.join(config.root, config.codebrowse.secret_path)).read()

app = RootApp(SESSION_VAR)

app = HTTPExceptionHandler(app)

app = SessionHandler(app, SESSION_VAR, secret)

def log_request_start_and_stop(app):

    def wrapped(environ, start_response):

        log = logging.getLogger('loggerhead')

        url = construct_url(environ)

        log.info("Starting to process %s", url)

        start_time = time.time()

        def request_done_ok():

            log.info("Processed ok %s [%0.3f seconds]", url, time.time() -

                    start_time)

        def request_done_err(exc_info):

            log.info("Processed err %s [%0.3f seconds]: %s", url, time.time() -

                    start_time, traceback.format_exception_only(*exc_info[:2]))

        return catch_errors(app, environ, start_response, request_done_err,

                request_done_ok)

    return wrapped

app = log_request_start_and_stop(app)

app = PrefixMiddleware(app)

app = TransLogger(app)

app = threadpool_debug(app)

def set_scheme(app):

    """Set wsgi.url_scheme in the environment correctly.

    We serve requests that originated from both http and https, and

    distinguish between them by adding a header in the https Apache config.

    """

    def wrapped(environ, start_response):

        environ['wsgi.url_scheme'] = environ.pop(

            'HTTP_X_FORWARDED_SCHEME', 'http')

        return app(environ, start_response)

    return wrapped

app = set_scheme(app)

app = change_kill_thread_criteria(app)

app = oops_middleware(app)

try:

    httpserver.serve(

        app, host=LISTEN_HOST, port=LISTEN_PORT,

        threadpool_workers=THREADPOOL_WORKERS,

        threadpool_options={

            # Kill threads after 300 seconds.  This is insanely high, but

            # lower enough than the default (1800 seconds!) that evidence

            # suggests it will be hit occasionally, and there's very little

            # chance of it having negative consequences.

            'kill_thread_limit': 300,

            # Check for threads that should be killed every 10 requests.  The

            # default is every 100, which is easily long enough for things to

            # gum up completely in between checks.

            'hung_check_period': 10,

            })

finally:

    log.info('Shutdown.')

    try:

        os.remove(pidfile)

    except OSError:

        pass