def main(args):
syslog.openlog("config-manager", syslog.LOG_PID)
pdlogs.STARTUP.log()
try:
arguments = docopt(__doc__, argv=args)
except DocoptExit:
pdlogs.EXITING_BAD_CONFIG.log()
raise
local_ip = arguments['--local-ip']
local_site = arguments['--local-site']
log_dir = arguments['--log-directory']
log_level = LOG_LEVELS.get(arguments['--log-level'], logging.DEBUG)
stdout_err_log = os.path.join(log_dir, "config-manager.output.log")
if not arguments['--foreground']:
utils.daemonize(stdout_err_log)
logging_config.configure_logging(log_level, log_dir, "config-manager", show_thread=True)
# urllib3 logs a WARNING log whenever it recreates a connection, but our
# etcd usage does this frequently (to allow watch timeouts), so deliberately
# ignore this log
urllib_logger = logging.getLogger('urllib3')
urllib_logger.setLevel(logging.ERROR)
utils.install_sigusr1_handler("config-manager")
# Drop a pidfile.
pid = os.getpid()
with open(arguments['--pidfile'], "w") as pidfile:
pidfile.write(str(pid) + "\n")
plugins_dir = "/usr/share/clearwater/clearwater-config-manager/plugins/"
plugins = load_plugins_in_dir(plugins_dir)
plugins.sort(key=lambda x: x.key())
threads = []
files = [p.file() for p in plugins]
alarm = ConfigAlarm(files)
for plugin in plugins:
syncer = EtcdSynchronizer(plugin, local_ip, local_site, alarm)
thread = Thread(target=syncer.main, name=plugin.__class__.__name__)
thread.start()
threads.append(thread)
_log.info("Loaded plugin %s" % plugin)
while any([thread.isAlive() for thread in threads]):
for thread in threads:
if thread.isAlive():
thread.join(1)
_log.info("Clearwater Configuration Manager shutting down")
pdlogs.EXITING.log()
syslog.closelog()
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Parse arguments
parser = argparse.ArgumentParser(description="Homer web server")
parser.add_argument("--background", action="store_true", help="Detach and run server in background")
parser.add_argument("--worker-processes", default=1, type=int)
parser.add_argument("--shared-http-fd", default=None, type=int)
parser.add_argument("--process-id", default=0, type=int)
args = parser.parse_args()
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR, settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
# Drop a pidfile.
pid = os.getpid()
with open(settings.PID_FILE, "w") as pidfile:
pidfile.write(str(pid) + "\n")
utils.install_sigusr1_handler(settings.LOG_FILE_PREFIX)
# Setup logging
logging_config.configure_logging(args.process_id)
# setup accumulators and counters for statistics gathering
api.base.setupStats(args.process_id, args.worker_processes)
if args.shared_http_fd:
reactor.adoptStreamPort(args.shared_http_fd, AF_INET, application)
else:
# Cyclone
_log.info("Going to listen for HTTP on port %s", settings.HTTP_PORT)
http_port = reactor.listenTCP(settings.HTTP_PORT, application, interface=settings.LOCAL_IP)
for process_id in range(1, args.worker_processes):
reactor.spawnProcess(None, executable, [executable, __file__,
"--shared-http-fd", str(http_port.fileno()),
"--process-id", str(process_id)],
childFDs={0: 0, 1: 1, 2: 2, http_port.fileno(): http_port.fileno()},
env = os.environ)
# Kick off the reactor to start listening on configured ports
reactor.run()
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Parse arguments
parser = argparse.ArgumentParser(description="Ellis web server")
parser.add_argument("--background", action="store_true", help="Detach and run server in background")
args = parser.parse_args()
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR, settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
utils.install_sigusr1_handler(settings.LOG_FILE_PREFIX)
# Drop a pidfile.
pid = os.getpid()
with open(settings.PID_FILE, "w") as pidfile:
pidfile.write(str(pid) + "\n")
# Fork off a child process per core. In the parent process, the
# fork_processes call blocks until the children exit.
num_processes = settings.TORNADO_PROCESSES_PER_CORE * tornado.process.cpu_count()
task_id = tornado.process.fork_processes(num_processes)
if task_id is not None:
logging_config.configure_logging(settings.LOG_LEVEL, settings.LOGS_DIR, settings.LOG_FILE_PREFIX, task_id)
# We're a child process, start up.
_log.info("Process %s starting up", task_id)
connection.init_connection()
http_server = httpserver.HTTPServer(application)
unix_socket = bind_unix_socket(settings.HTTP_UNIX + "-" + str(task_id),
0666);
http_server.add_socket(unix_socket)
homestead.ping()
background.start_background_worker_io_loop()
io_loop = tornado.ioloop.IOLoop.instance()
io_loop.start()
else:
# This shouldn't happen since the children should run their IOLoops
# forever.
_log.critical("Children all exited")
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Parse arguments
parser = argparse.ArgumentParser(description="Homer web server")
parser.add_argument("--background",
action="store_true",
help="Detach and run server in background")
parser.add_argument("--worker-processes", default=1, type=int)
parser.add_argument("--shared-http-fd", default=None, type=int)
args = parser.parse_args()
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR,
settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
# Drop a pidfile.
pid = os.getpid()
with open(settings.PID_FILE, "w") as pidfile:
pidfile.write(str(pid) + "\n")
if args.shared_http_fd:
reactor.adoptStreamPort(args.shared_http_fd, AF_INET, application)
else:
# Cyclone
_log.info("Going to listen for HTTP on port %s", settings.HTTP_PORT)
http_port = reactor.listenTCP(settings.HTTP_PORT,
application,
interface="0.0.0.0")
for _ in range(1, args.worker_processes):
reactor.spawnProcess(None,
executable, [
executable, __file__, "--shared-http-fd",
str(http_port.fileno())
],
childFDs={
0: 0,
1: 1,
2: 2,
http_port.fileno(): http_port.fileno()
},
env=os.environ)
# Kick off the reactor to start listening on configured ports
reactor.run()
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Parse arguments
parser = argparse.ArgumentParser(description="Ellis web server")
parser.add_argument("--background", action="store_true", help="Detach and run server in background")
args = parser.parse_args()
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
listening_on_some_port = False
http_sockets = None
https_sockets = None
if settings.ALLOW_HTTP:
http_sockets = bind_sockets(settings.HTTP_PORT, address=settings.LOCAL_IP)
listening_on_some_port = True
if (os.path.exists(settings.TLS_CERTIFICATE) and
os.path.exists(settings.TLS_PRIVATE_KEY)):
https_sockets = bind_sockets(settings.HTTPS_PORT, address=settings.LOCAL_IP)
listening_on_some_port = True
if not listening_on_some_port:
# We usually don't configure logging until after we fork but since
# we're about to crash...
logging_config.configure_logging("parent")
_log.critical("Failed to listen on any ports.")
raise Exception("Failed to listen on any ports")
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR, settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
utils.install_sigusr1_handler(settings.LOG_FILE_PREFIX)
# Drop a pidfile.
pid = os.getpid()
with open(settings.PID_FILE, "w") as pidfile:
pidfile.write(str(pid) + "\n")
# Fork off a child process per core. In the parent process, the
# fork_processes call blocks until the children exit.
num_processes = settings.TORNADO_PROCESSES_PER_CORE * tornado.process.cpu_count()
task_id = tornado.process.fork_processes(num_processes)
if task_id is not None:
logging_config.configure_logging(task_id)
# We're a child process, start up.
_log.info("Process %s starting up", task_id)
connection.init_connection()
if http_sockets:
_log.info("Going to listen for HTTP on port %s", settings.HTTP_PORT)
http_server = httpserver.HTTPServer(application)
http_server.add_sockets(http_sockets)
else:
_log.info("Not starting HTTP, set ALLOW_HTTP in local_settings.py to enable HTTP.")
if https_sockets:
_log.info("Going to listen for HTTPS on port %s", settings.HTTPS_PORT)
https_server = httpserver.HTTPServer(application,
ssl_options={
"certfile": settings.TLS_CERTIFICATE,
"keyfile": settings.TLS_PRIVATE_KEY,
})
https_server.add_sockets(https_sockets)
else:
_log.critical("Not starting HTTPS")
homestead.ping()
background.start_background_worker_io_loop()
io_loop = tornado.ioloop.IOLoop.instance()
io_loop.start()
else:
# This shouldn't happen since the children should run their IOLoops
# forever.
_log.critical("Children all exited")
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Hack to work-around issue with Cyclone and UNIX domain sockets
twisted.internet.address.UNIXAddress.host = "localhost"
# Parse arguments
parser = argparse.ArgumentParser(description="Crest web server")
parser.add_argument("--background", action="store_true", help="Detach and run server in background")
parser.add_argument("--signaling-namespace", action="store_true", help="Server running in signaling namespace")
parser.add_argument("--worker-processes", default=1, type=int)
parser.add_argument("--shared-http-tcp-fd", default=None, type=int)
parser.add_argument("--process-id", default=0, type=int)
args = parser.parse_args()
# Set process name.
prctl.prctl(prctl.NAME, settings.PROCESS_NAME)
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR, settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
utils.install_sigusr1_handler(settings.LOG_FILE_PREFIX)
# Setup logging
syslog.openlog(settings.LOG_FILE_PREFIX, syslog.LOG_PID)
logging_config.configure_logging(settings.LOG_LEVEL, settings.LOGS_DIR, settings.LOG_FILE_PREFIX, args.process_id)
twisted.python.log.addObserver(on_twisted_log)
pdlogs.CREST_STARTING.log()
# setup accumulators and counters for statistics gathering
api.base.setupStats(args.process_id, args.worker_processes)
# Initialize reactor ports and create worker sub-processes
if args.process_id == 0:
# Main process startup, create pidfile.
# We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(settings.PID_FILE) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
# Create UNIX domain socket for nginx front-end (used for
# normal operation and as a bridge from the default namespace to the signaling
# namespace in a multiple interface configuration).
bind_safely(reactor, args.process_id, application)
pdlogs.CREST_UP.log()
if args.signaling_namespace and settings.PROCESS_NAME == "homer":
# Running in signaling namespace as Homer, create TCP socket for XDMS requests
# from signaling interface
_log.info("Going to listen for HTTP on TCP port %s", settings.HTTP_PORT)
http_tcp_port = reactor.listenTCP(settings.HTTP_PORT, application, interface=settings.LOCAL_IP)
# Spin up worker sub-processes, passing TCP file descriptor
for process_id in range(1, args.worker_processes):
reactor.spawnProcess(None, executable, [executable, __file__,
"--shared-http-tcp-fd", str(http_tcp_port.fileno()),
"--process-id", str(process_id)],
childFDs={0: 0, 1: 1, 2: 2, http_tcp_port.fileno(): http_tcp_port.fileno()},
env = os.environ)
else:
# Spin up worker sub-processes
for process_id in range(1, args.worker_processes):
reactor.spawnProcess(None, executable, [executable, __file__,
"--process-id", str(process_id)],
childFDs={0: 0, 1: 1, 2: 2},
env = os.environ)
else:
# Sub-process startup, ensure we die if our parent does.
prctl.prctl(prctl.PDEATHSIG, signal.SIGTERM)
# Create UNIX domain socket for nginx front-end based on process ID.
bind_safely(reactor, args.process_id, application)
# Create TCP socket if file descriptor was passed.
if args.shared_http_tcp_fd:
reactor.adoptStreamPort(args.shared_http_tcp_fd, AF_INET, application)
# We need to catch the shutdown request so that we can properly stop
# the ZMQ interface; otherwise the reactor won't shut down on a SIGTERM
# and will be SIGKILLed when the service is stopped.
reactor.addSystemEventTrigger('before', 'shutdown', on_before_shutdown)
# Kick off the reactor to start listening on configured ports
reactor.run()
def main(args):
syslog.openlog("cluster-manager", syslog.LOG_PID)
pdlogs.STARTUP.log()
try:
arguments = docopt(__doc__, argv=args)
except DocoptExit:
pdlogs.EXITING_BAD_CONFIG.log()
raise
mgmt_ip = arguments['--mgmt-local-ip']
sig_ip = arguments['--sig-local-ip']
local_site_name = arguments['--local-site']
remote_site_name = arguments['--remote-site']
remote_cassandra_seeds = arguments['--remote-cassandra-seeds']
if remote_cassandra_seeds:
remote_cassandra_seeds = remote_cassandra_seeds.split(',')
else:
remote_cassandra_seeds = []
signaling_namespace = arguments.get('--signaling-namespace')
local_uuid = UUID(arguments['--uuid'])
etcd_key = arguments.get('--etcd-key')
etcd_cluster_key = arguments.get('--etcd-cluster-key')
cluster_manager_enabled = arguments['--cluster-manager-enabled']
log_dir = arguments['--log-directory']
log_level = LOG_LEVELS.get(arguments['--log-level'], logging.DEBUG)
stdout_err_log = os.path.join(log_dir, "cluster-manager.output.log")
# Check that there's an etcd_cluster_key value passed to the cluster
# manager
if etcd_cluster_key == "":
# The etcd_cluster_key isn't valid, and possibly get weird entries in
# the etcd database if we allow the cluster_manager to start
pdlogs.EXITING_MISSING_ETCD_CLUSTER_KEY.log()
exit(1)
if not arguments['--foreground']:
utils.daemonize(stdout_err_log)
# Process names are limited to 15 characters, so abbreviate
prctl.prctl(prctl.NAME, "cw-cluster-mgr")
logging_config.configure_logging(log_level, log_dir, "cluster-manager", show_thread=True)
# urllib3 logs a WARNING log whenever it recreates a connection, but our
# etcd usage does this frequently (to allow watch timeouts), so deliberately
# ignore this log
urllib_logger = logging.getLogger('urllib3')
urllib_logger.setLevel(logging.ERROR)
utils.install_sigusr1_handler("cluster-manager")
# Drop a pidfile. We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(arguments['--pidfile']) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
plugins_dir = "/usr/share/clearwater/clearwater-cluster-manager/plugins/"
plugins = load_plugins_in_dir(plugins_dir,
PluginParams(ip=sig_ip,
mgmt_ip=mgmt_ip,
local_site=local_site_name,
remote_site=remote_site_name,
remote_cassandra_seeds=remote_cassandra_seeds,
signaling_namespace=signaling_namespace,
uuid=local_uuid,
etcd_key=etcd_key,
etcd_cluster_key=etcd_cluster_key))
plugins.sort(key=lambda x: x.key())
plugins_to_use = []
files = []
skip = False
for plugin in plugins:
for plugin_file in plugin.files():
if plugin_file in files:
_log.info("Skipping plugin {} because {} "
"is already managed by another plugin"
.format(plugin, plugin_file))
skip = True
if not skip:
plugins_to_use.append(plugin)
files.extend(plugin.files())
synchronizers = []
threads = []
if cluster_manager_enabled == "N":
# Don't start any threads as we don't want the cluster manager to run
pdlogs.DO_NOT_START.log()
elif etcd_cluster_key == "DO_NOT_CLUSTER":
# Don't start any threads as we don't want this box to cluster
pdlogs.DO_NOT_CLUSTER.log()
else:
for plugin in plugins_to_use:
syncer = EtcdSynchronizer(plugin, sig_ip, etcd_ip=mgmt_ip)
syncer.start_thread()
synchronizers.append(syncer)
threads.append(syncer.thread)
_log.info("Loaded plugin %s" % plugin)
install_sigquit_handler(synchronizers)
utils.install_sigterm_handler(synchronizers)
while any([thread.isAlive() for thread in threads]):
for thread in threads:
if thread.isAlive():
thread.join(1)
_log.info("No plugin threads running, waiting for a SIGTERM or SIGQUIT")
while not utils.should_quit and not should_quit:
sleep(1)
_log.info("Quitting")
_log.debug("%d threads outstanding at exit" % activeCount())
pdlogs.EXITING.log()
syslog.closelog()
def main(args):
syslog.openlog("queue-manager", syslog.LOG_PID)
pdlogs.STARTUP.log()
try:
arguments = docopt(__doc__, argv=args)
except DocoptExit:
pdlogs.EXITING_BAD_CONFIG.log()
raise
local_ip = arguments['--local-ip']
local_site = arguments['--local-site']
etcd_key = arguments['--etcd-key']
node_type = arguments['--node-type']
log_dir = arguments['--log-directory']
log_level = LOG_LEVELS.get(arguments['--log-level'], logging.DEBUG)
wait_plugin_complete = arguments['--wait-plugin-complete']
stdout_err_log = os.path.join(log_dir, "queue-manager.output.log")
if not arguments['--foreground']:
utils.daemonize(stdout_err_log)
# Process names are limited to 15 characters, so abbreviate
prctl.prctl(prctl.NAME, "cw-queue-mgr")
logging_config.configure_logging(log_level, log_dir, "queue-manager", show_thread=True)
# urllib3 logs a WARNING log whenever it recreates a connection, but our
# etcd usage does this frequently (to allow watch timeouts), so deliberately
# ignore this log
urllib_logger = logging.getLogger('urllib3')
urllib_logger.setLevel(logging.ERROR)
utils.install_sigusr1_handler("queue-manager")
# Drop a pidfile. We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(arguments['--pidfile']) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
plugins_dir = "/usr/share/clearwater/clearwater-queue-manager/plugins/"
plugins = load_plugins_in_dir(plugins_dir,
PluginParams(wait_plugin_complete=wait_plugin_complete))
plugins.sort(key=lambda x: x.key())
synchronizers = []
threads = []
for plugin in plugins:
syncer = EtcdSynchronizer(plugin, local_ip, local_site, etcd_key, node_type)
syncer.start_thread()
synchronizers.append(syncer)
threads.append(syncer.thread)
_log.info("Loaded plugin %s" % plugin)
utils.install_sigterm_handler(synchronizers)
while any([thr.isAlive() for thr in threads]):
for thr in threads:
if thr.isAlive():
thr.join(1)
while not utils.should_quit:
sleep(1)
_log.info("Clearwater Queue Manager shutting down")
pdlogs.EXITING.log()
syslog.closelog()
def main(args):
syslog.openlog("cluster-manager", syslog.LOG_PID)
pdlogs.STARTUP.log()
try:
arguments = docopt(__doc__, argv=args)
except DocoptExit:
pdlogs.EXITING_BAD_CONFIG.log()
raise
mgmt_ip = arguments['--mgmt-local-ip']
sig_ip = arguments['--sig-local-ip']
local_site_name = arguments['--local-site']
remote_site_name = arguments['--remote-site']
remote_cassandra_seeds = arguments['--remote-cassandra-seeds']
if remote_cassandra_seeds:
remote_cassandra_seeds = remote_cassandra_seeds.split(',')
else:
remote_cassandra_seeds = []
signaling_namespace = arguments.get('--signaling-namespace')
local_uuid = UUID(arguments['--uuid'])
etcd_key = arguments.get('--etcd-key')
etcd_cluster_key = arguments.get('--etcd-cluster-key')
cluster_manager_enabled = arguments['--cluster-manager-enabled']
log_dir = arguments['--log-directory']
log_level = LOG_LEVELS.get(arguments['--log-level'], logging.DEBUG)
stdout_err_log = os.path.join(log_dir, "cluster-manager.output.log")
# Check that there's an etcd_cluster_key value passed to the cluster
# manager
if etcd_cluster_key == "":
# The etcd_cluster_key isn't valid, and possibly get weird entries in
# the etcd database if we allow the cluster_manager to start
pdlogs.EXITING_MISSING_ETCD_CLUSTER_KEY.log()
exit(1)
if not arguments['--foreground']:
utils.daemonize(stdout_err_log)
# Process names are limited to 15 characters, so abbreviate
prctl.prctl(prctl.NAME, "cw-cluster-mgr")
logging_config.configure_logging(log_level,
log_dir,
"cluster-manager",
show_thread=True)
# urllib3 logs a WARNING log whenever it recreates a connection, but our
# etcd usage does this frequently (to allow watch timeouts), so deliberately
# ignore this log
urllib_logger = logging.getLogger('urllib3')
urllib_logger.setLevel(logging.ERROR)
utils.install_sigusr1_handler("cluster-manager")
# Drop a pidfile. We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(
arguments['--pidfile']) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
plugins_dir = "/usr/share/clearwater/clearwater-cluster-manager/plugins/"
plugins = load_plugins_in_dir(
plugins_dir,
PluginParams(ip=sig_ip,
mgmt_ip=mgmt_ip,
local_site=local_site_name,
remote_site=remote_site_name,
remote_cassandra_seeds=remote_cassandra_seeds,
signaling_namespace=signaling_namespace,
uuid=local_uuid,
etcd_key=etcd_key,
etcd_cluster_key=etcd_cluster_key))
plugins.sort(key=lambda x: x.key())
plugins_to_use = []
files = []
skip = False
for plugin in plugins:
for plugin_file in plugin.files():
if plugin_file in files:
_log.info("Skipping plugin {} because {} "
"is already managed by another plugin".format(
plugin, plugin_file))
skip = True
if not skip:
plugins_to_use.append(plugin)
files.extend(plugin.files())
synchronizers = []
threads = []
if cluster_manager_enabled == "N":
# Don't start any threads as we don't want the cluster manager to run
pdlogs.DO_NOT_START.log()
elif etcd_cluster_key == "DO_NOT_CLUSTER":
# Don't start any threads as we don't want this box to cluster
pdlogs.DO_NOT_CLUSTER.log()
else:
for plugin in plugins_to_use:
syncer = EtcdSynchronizer(plugin, sig_ip, etcd_ip=mgmt_ip)
syncer.start_thread()
synchronizers.append(syncer)
threads.append(syncer.thread)
_log.info("Loaded plugin %s" % plugin)
install_sigquit_handler(synchronizers)
install_sigterm_handler(synchronizers)
while any([thread.isAlive() for thread in threads]):
for thread in threads:
if thread.isAlive():
thread.join(1)
_log.info("No plugin threads running, waiting for a SIGTERM or SIGQUIT")
while not should_quit:
sleep(1)
_log.info("Quitting")
_log.debug("%d threads outstanding at exit" % activeCount())
pdlogs.EXITING.log()
syslog.closelog()
def main(args):
syslog.openlog("cluster-manager", syslog.LOG_PID)
pdlogs.STARTUP.log()
try:
arguments = docopt(__doc__, argv=args)
except DocoptExit:
pdlogs.EXITING_BAD_CONFIG.log()
raise
mgmt_ip = arguments['--mgmt-local-ip']
sig_ip = arguments['--sig-local-ip']
local_site_name = arguments['--local-site']
remote_site_name = arguments['--remote-site']
signaling_namespace = arguments.get('--signaling-namespace')
log_dir = arguments['--log-directory']
log_level = LOG_LEVELS.get(arguments['--log-level'], logging.DEBUG)
stdout_err_log = os.path.join(log_dir, "cluster-manager.output.log")
if not arguments['--foreground']:
utils.daemonize(stdout_err_log)
logging_config.configure_logging(log_level, log_dir, "cluster-manager", show_thread=True)
# urllib3 logs a WARNING log whenever it recreates a connection, but our
# etcd usage does this frequently (to allow watch timeouts), so deliberately
# ignore this log
urllib_logger = logging.getLogger('urllib3')
urllib_logger.setLevel(logging.ERROR)
utils.install_sigusr1_handler("cluster-manager")
# Drop a pidfile.
pid = os.getpid()
with open(arguments['--pidfile'], "w") as pidfile:
pidfile.write(str(pid) + "\n")
plugins_dir = "/usr/share/clearwater/clearwater-cluster-manager/plugins/"
plugins = load_plugins_in_dir(plugins_dir,
PluginParams(ip=sig_ip,
mgmt_ip=mgmt_ip,
local_site=local_site_name,
remote_site=remote_site_name,
signaling_namespace=signaling_namespace))
plugins.sort(key=lambda x: x.key())
plugins_to_use = []
files = []
skip = False
for plugin in plugins:
for plugin_file in plugin.files():
if plugin_file in files:
_log.info("Skipping plugin {} because {} "
"is already managed by another plugin"
.format(plugin, plugin_file))
skip = True
if not skip:
plugins_to_use.append(plugin)
files.extend(plugin.files())
synchronizers = []
threads = []
for plugin in plugins_to_use:
syncer = EtcdSynchronizer(plugin, sig_ip, etcd_ip=mgmt_ip)
syncer.start_thread()
synchronizers.append(syncer)
threads.append(syncer.thread)
_log.info("Loaded plugin %s" % plugin)
install_sigquit_handler(synchronizers)
install_sigterm_handler(synchronizers)
while any([thread.isAlive() for thread in threads]):
for thread in threads:
if thread.isAlive():
thread.join(1)
_log.info("No plugin threads running, waiting for a SIGTERM or SIGQUIT")
while not should_quit:
sleep(1)
_log.info("Quitting")
_log.debug("%d threads outstanding at exit" % activeCount())
pdlogs.EXITING.log()
syslog.closelog()
# Use os.exit to skip exit handlers - otherwise the concurrent.futures exit
# handler waits for an infinite wait to end
os._exit(0)
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Hack to work-around issue with Cyclone and UNIX domain sockets
twisted.internet.address.UNIXAddress.host = "localhost"
# Parse arguments
parser = argparse.ArgumentParser(description="Crest web server")
parser.add_argument("--background",
action="store_true",
help="Detach and run server in background")
parser.add_argument("--signaling-namespace",
action="store_true",
help="Server running in signaling namespace")
parser.add_argument("--worker-processes", default=1, type=int)
parser.add_argument("--shared-http-tcp-fd", default=None, type=int)
parser.add_argument("--process-id", default=0, type=int)
parser.add_argument("--log-level", default=2, type=int)
args = parser.parse_args()
# Set process name.
prctl.prctl(prctl.NAME, settings.PROCESS_NAME)
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR,
settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
utils.install_sigusr1_handler(settings.LOG_FILE_PREFIX)
# Setup logging
syslog.openlog(settings.LOG_FILE_PREFIX, syslog.LOG_PID)
logging_config.configure_logging(
utils.map_clearwater_log_level(args.log_level), settings.LOGS_DIR,
settings.LOG_FILE_PREFIX, args.process_id)
twisted.python.log.addObserver(on_twisted_log)
pdlogs.CREST_STARTING.log()
# setup accumulators and counters for statistics gathering
api.base.setupStats(args.process_id, args.worker_processes)
# Initialize reactor ports and create worker sub-processes
if args.process_id == 0:
# Main process startup, create pidfile.
# We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(
settings.PID_FILE) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
# Create UNIX domain socket for nginx front-end (used for
# normal operation and as a bridge from the default namespace to the signaling
# namespace in a multiple interface configuration).
bind_safely(reactor, args.process_id, application)
pdlogs.CREST_UP.log()
if args.signaling_namespace and settings.PROCESS_NAME == "homer":
# Running in signaling namespace as Homer, create TCP socket for XDMS requests
# from signaling interface
_log.info("Going to listen for HTTP on TCP port %s",
settings.HTTP_PORT)
http_tcp_port = reactor.listenTCP(settings.HTTP_PORT,
application,
interface=settings.LOCAL_IP)
# Spin up worker sub-processes, passing TCP file descriptor
for process_id in range(1, args.worker_processes):
reactor.spawnProcess(
None,
executable, [
executable, __file__, "--shared-http-tcp-fd",
str(http_tcp_port.fileno()), "--process-id",
str(process_id)
],
childFDs={
0: 0,
1: 1,
2: 2,
http_tcp_port.fileno(): http_tcp_port.fileno()
},
env=os.environ)
else:
# Spin up worker sub-processes
for process_id in range(1, args.worker_processes):
reactor.spawnProcess(
None,
executable,
[executable, __file__, "--process-id",
str(process_id)],
childFDs={
0: 0,
1: 1,
2: 2
},
env=os.environ)
else:
# Sub-process startup, ensure we die if our parent does.
prctl.prctl(prctl.PDEATHSIG, signal.SIGTERM)
# Create UNIX domain socket for nginx front-end based on process ID.
bind_safely(reactor, args.process_id, application)
# Create TCP socket if file descriptor was passed.
if args.shared_http_tcp_fd:
reactor.adoptStreamPort(args.shared_http_tcp_fd, AF_INET,
application)
# We need to catch the shutdown request so that we can properly stop
# the ZMQ interface; otherwise the reactor won't shut down on a SIGTERM
# and will be SIGKILLed when the service is stopped.
reactor.addSystemEventTrigger('before', 'shutdown', on_before_shutdown)
# Kick off the reactor to start listening on configured ports
reactor.run()
def main(args):
syslog.openlog("queue-manager", syslog.LOG_PID)
pdlogs.STARTUP.log()
try:
arguments = docopt(__doc__, argv=args)
except DocoptExit:
pdlogs.EXITING_BAD_CONFIG.log()
raise
local_ip = arguments['--local-ip']
local_site = arguments['--local-site']
etcd_key = arguments['--etcd-key']
node_type = arguments['--node-type']
log_dir = arguments['--log-directory']
log_level = LOG_LEVELS.get(arguments['--log-level'], logging.DEBUG)
wait_plugin_complete = arguments['--wait-plugin-complete']
stdout_err_log = os.path.join(log_dir, "queue-manager.output.log")
if not arguments['--foreground']:
utils.daemonize(stdout_err_log)
# Process names are limited to 15 characters, so abbreviate
prctl.prctl(prctl.NAME, "cw-queue-mgr")
logging_config.configure_logging(log_level,
log_dir,
"queue-manager",
show_thread=True)
# urllib3 logs a WARNING log whenever it recreates a connection, but our
# etcd usage does this frequently (to allow watch timeouts), so deliberately
# ignore this log
urllib_logger = logging.getLogger('urllib3')
urllib_logger.setLevel(logging.ERROR)
utils.install_sigusr1_handler("queue-manager")
# Drop a pidfile. We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(
arguments['--pidfile']) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
plugins_dir = "/usr/share/clearwater/clearwater-queue-manager/plugins/"
plugins = load_plugins_in_dir(
plugins_dir, PluginParams(wait_plugin_complete=wait_plugin_complete))
plugins.sort(key=lambda x: x.key())
synchronizers = []
threads = []
# Load the plugins, but don't start them until we've installed the SIGTERM
# handler, as that handler will gracefully shut down any running
# synchronizers on receiving a SIGTERM
for plugin in plugins:
syncer = EtcdSynchronizer(plugin, local_ip, local_site, etcd_key,
node_type)
synchronizers.append(syncer)
threads.append(syncer.thread)
_log.info("Loaded plugin %s" % plugin)
utils.install_sigterm_handler(synchronizers)
# Now start the plugin threads
for syncer in synchronizers:
syncer.start_thread()
_log.info("Started thread for plugin %s" % syncer._plugin)
while any([thr.isAlive() for thr in threads]):
for thr in threads:
if thr.isAlive():
thr.join(1)
while not utils.should_quit:
sleep(1)
_log.info("Clearwater Queue Manager shutting down")
pdlogs.EXITING.log()
syslog.closelog()
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Hack to work-around issue with Cyclone and UNIX domain sockets
twisted.internet.address.UNIXAddress.host = "localhost"
# Parse arguments
parser = argparse.ArgumentParser(description="Homer web server")
parser.add_argument("--background", action="store_true", help="Detach and run server in background")
parser.add_argument("--signaling-namespace", action="store_true", help="Server running in signaling namespace")
parser.add_argument("--worker-processes", default=1, type=int)
parser.add_argument("--shared-http-tcp-fd", default=None, type=int)
parser.add_argument("--process-id", default=0, type=int)
args = parser.parse_args()
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR, settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
# Drop a pidfile.
pid = os.getpid()
with open(settings.PID_FILE, "w") as pidfile:
pidfile.write(str(pid) + "\n")
utils.install_sigusr1_handler(settings.LOG_FILE_PREFIX)
# Setup logging
logging_config.configure_logging(settings.LOG_LEVEL, settings.LOGS_DIR, settings.LOG_FILE_PREFIX, args.process_id)
# setup accumulators and counters for statistics gathering
api.base.setupStats(args.process_id, args.worker_processes)
# Initialize reactor ports and create worker sub-processes
if args.process_id == 0:
# Main process startup, create UNIX domain socket for nginx front-end (used for
# normal operation and as a bridge from the default namespace to the signaling
# namespace in a multiple interface configuration).
_lock_fd = bind_safely(reactor, args.process_id, application)
if args.signaling_namespace and settings.PROCESS_NAME == "homer":
# Running in signaling namespace as Homer, create TCP socket for XDMS requests
# from signaling interface
_log.info("Going to listen for HTTP on TCP port %s", settings.HTTP_PORT)
http_tcp_port = reactor.listenTCP(settings.HTTP_PORT, application, interface=settings.LOCAL_IP)
# Spin up worker sub-processes, passing TCP file descriptor
for process_id in range(1, args.worker_processes):
reactor.spawnProcess(None, executable, [executable, __file__,
"--shared-http-tcp-fd", str(http_tcp_port.fileno()),
"--process-id", str(process_id)],
childFDs={0: 0, 1: 1, 2: 2, http_tcp_port.fileno(): http_tcp_port.fileno()},
env = os.environ)
else:
# Spin up worker sub-processes
for process_id in range(1, args.worker_processes):
reactor.spawnProcess(None, executable, [executable, __file__,
"--process-id", str(process_id)],
childFDs={0: 0, 1: 1, 2: 2},
env = os.environ)
else:
# Sub-process startup, create UNIX domain socket for nginx front-end based on
# process ID.
_lock_fd = bind_safely(reactor, args.process_id, application)
# Create TCP socket if file descriptor was passed.
if args.shared_http_tcp_fd:
reactor.adoptStreamPort(args.shared_http_tcp_fd, AF_INET, application)
# Kick off the reactor to start listening on configured ports
reactor.run()
def standalone():
"""
Initializes Tornado and our application. Forks worker processes to handle
requests. Does not return until all child processes exit normally.
"""
# Parse arguments
parser = argparse.ArgumentParser(description="Ellis web server")
parser.add_argument("--background",
action="store_true",
help="Detach and run server in background")
parser.add_argument("--log-level", default=2, type=int)
args = parser.parse_args()
prctl.prctl(prctl.NAME, "ellis")
# We don't initialize logging until we fork because we want each child to
# have its own logging and it's awkward to reconfigure logging that is
# defined by the parent.
application = create_application()
if args.background:
# Get a new logfile, rotating the old one if present.
err_log_name = os.path.join(settings.LOGS_DIR,
settings.LOG_FILE_PREFIX + "-err.log")
try:
os.rename(err_log_name, err_log_name + ".old")
except OSError:
pass
# Fork into background.
utils.daemonize(err_log_name)
utils.install_sigusr1_handler(settings.LOG_FILE_PREFIX)
# Drop a pidfile. We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(settings.PID_FILE) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
# Only run one process, not one per core - we don't need the performance
# and this keeps everything in one log file
prctl.prctl(prctl.NAME, "ellis")
logging_config.configure_logging(
utils.map_clearwater_log_level(args.log_level), settings.LOGS_DIR,
settings.LOG_FILE_PREFIX)
_log.info("Ellis process starting up")
connection.init_connection()
http_server = httpserver.HTTPServer(application)
unix_socket = bind_unix_socket(settings.HTTP_UNIX, 0666)
http_server.add_socket(unix_socket)
homestead.ping()
background.start_background_worker_io_loop()
io_loop = tornado.ioloop.IOLoop.instance()
io_loop.start()
def main(args):
syslog.openlog("config-manager", syslog.LOG_PID)
pdlogs.STARTUP.log()
try:
arguments = docopt(__doc__, argv=args)
except DocoptExit:
pdlogs.EXITING_BAD_CONFIG.log()
raise
local_ip = arguments['--local-ip']
local_site = arguments['--local-site']
etcd_key = arguments['--etcd-key']
log_dir = arguments['--log-directory']
log_level = LOG_LEVELS.get(arguments['--log-level'], logging.DEBUG)
stdout_err_log = os.path.join(log_dir, "config-manager.output.log")
if not arguments['--foreground']:
utils.daemonize(stdout_err_log)
# Process names are limited to 15 characters, so abbreviate
prctl.prctl(prctl.NAME, "cw-config-mgr")
logging_config.configure_logging(log_level, log_dir, "config-manager", show_thread=True)
# urllib3 logs a WARNING log whenever it recreates a connection, but our
# etcd usage does this frequently (to allow watch timeouts), so deliberately
# ignore this log
urllib_logger = logging.getLogger('urllib3')
urllib_logger.setLevel(logging.ERROR)
utils.install_sigusr1_handler("config-manager")
# Drop a pidfile. We must keep a reference to the file object here, as this keeps
# the file locked and provides extra protection against two processes running at
# once.
pidfile_lock = None
try:
pidfile_lock = utils.lock_and_write_pid_file(arguments['--pidfile']) # noqa
except IOError:
# We failed to take the lock - another process is already running
exit(1)
plugins_dir = "/usr/share/clearwater/clearwater-config-manager/plugins/"
plugins = load_plugins_in_dir(plugins_dir)
plugins.sort(key=lambda x: x.key())
threads = []
files = [p.file() for p in plugins]
alarm = ConfigAlarm(files)
for plugin in plugins:
syncer = EtcdSynchronizer(plugin, local_ip, local_site, alarm, etcd_key)
syncer.start_thread()
threads.append(syncer.thread)
_log.info("Loaded plugin %s" % plugin)
while any([thr.isAlive() for thr in threads]):
for thr in threads:
if thr.isAlive():
thr.join(1)
_log.info("Clearwater Configuration Manager shutting down")
pdlogs.EXITING.log()
syslog.closelog()
