def _start_native_worker(self, wtype, id, options=None, details=None):
assert (wtype in ['router', 'container', 'websocket-testee'])
# prohibit starting a worker twice
#
if id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(
id)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.worker_already_running',
emsg)
# check worker options
#
options = options or {}
try:
if wtype == 'router':
checkconfig.check_router_options(options)
elif wtype == 'container':
checkconfig.check_container_options(options)
elif wtype == 'websocket-testee':
checkconfig.check_websocket_testee_options(options)
else:
raise Exception("logic error")
except Exception as e:
emsg = "Could not start native worker: invalid configuration ({})".format(
e)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
# allow override Python executable from options
#
if 'python' in options:
exe = options['python']
# the executable must be an absolute path, e.g. /home/oberstet/pypy-2.2.1-linux64/bin/pypy
#
if not os.path.isabs(exe):
emsg = "Invalid worker configuration: python executable '{}' must be an absolute path".format(
exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
# of course the path must exist and actually be executable
#
if not (os.path.isfile(exe) and os.access(exe, os.X_OK)):
emsg = "Invalid worker configuration: python executable '{}' does not exist or isn't an executable".format(
exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
else:
exe = sys.executable
# all native workers (routers and containers for now) start from the same script
#
filename = os.path.abspath(
os.path.join(crossbar.__file__, "..", "worker", "process.py"))
# assemble command line for forking the worker
#
args = [exe, "-u", filename]
args.extend(["--cbdir", self._node._cbdir])
args.extend(["--node", str(self._node_id)])
args.extend(["--worker", str(id)])
args.extend(["--realm", self._realm])
args.extend(["--type", wtype])
args.extend(["--loglevel", _loglevel])
# allow override worker process title from options
#
if options.get('title', None):
args.extend(['--title', options['title']])
# forward explicit reactor selection
#
if 'reactor' in options and sys.platform in options['reactor']:
args.extend(['--reactor', options['reactor'][sys.platform]])
# FIXME
# elif self._node.options.reactor:
# args.extend(['--reactor', self._node.options.reactor])
# create worker process environment
#
worker_env = create_process_env(options)
# We need to use the same PYTHONPATH we were started with, so we can
# find the Crossbar we're working with -- it may not be the same as the
# one on the default path
worker_env["PYTHONPATH"] = os.pathsep.join(sys.path)
# log name of worker
#
worker_logname = {
'router': 'Router',
'container': 'Container',
'websocket-testee': 'WebSocketTestee'
}.get(wtype, 'Worker')
# topic URIs used (later)
#
if wtype == 'router':
starting_topic = 'crossbar.node.{}.on_router_starting'.format(
self._node_id)
started_topic = 'crossbar.node.{}.on_router_started'.format(
self._node_id)
elif wtype == 'container':
starting_topic = 'crossbar.node.{}.on_container_starting'.format(
self._node_id)
started_topic = 'crossbar.node.{}.on_container_started'.format(
self._node_id)
elif wtype == 'websocket-testee':
starting_topic = 'crossbar.node.{}.on_websocket_testee_starting'.format(
self._node_id)
started_topic = 'crossbar.node.{}.on_websocket_testee_started'.format(
self._node_id)
else:
raise Exception("logic error")
# add worker tracking instance to the worker map ..
#
if wtype == 'router':
worker = RouterWorkerProcess(self,
id,
details.caller,
keeplog=options.get(
'traceback', None))
elif wtype == 'container':
worker = ContainerWorkerProcess(self,
id,
details.caller,
keeplog=options.get(
'traceback', None))
elif wtype == 'websocket-testee':
worker = WebSocketTesteeWorkerProcess(self,
id,
details.caller,
keeplog=options.get(
'traceback', None))
else:
raise Exception("logic error")
self._workers[id] = worker
# create a (custom) process endpoint.
#
if platform.isWindows():
childFDs = None # Use the default Twisted ones
else:
# The communication between controller and container workers is
# using WAMP running over 2 pipes.
# For controller->container traffic this runs over FD 0 (`stdin`)
# and for the container->controller traffic, this runs over FD 3.
#
# Note: We use FD 3, not FD 1 (`stdout`) or FD 2 (`stderr`) for
# container->controller traffic, so that components running in the
# container which happen to write to `stdout` or `stderr` do not
# interfere with the container-controller communication.
childFDs = {0: "w", 1: "r", 2: "r", 3: "r"}
ep = WorkerProcessEndpoint(self._node._reactor,
exe,
args,
env=worker_env,
worker=worker,
childFDs=childFDs)
# ready handling
#
def on_ready_success(id):
self.log.info("{worker} with ID '{id}' and PID {pid} started",
worker=worker_logname,
id=worker.id,
pid=worker.pid)
self._node._reactor.addSystemEventTrigger(
'before',
'shutdown',
self._cleanup_worker,
self._node._reactor,
worker,
)
worker.status = 'started'
worker.started = datetime.utcnow()
started_info = {
'id': worker.id,
'status': worker.status,
'started': utcstr(worker.started),
'who': worker.who
}
# FIXME: make start of stats printer dependent on log level ..
worker.log_stats(5.)
self.publish(started_topic,
started_info,
options=PublishOptions(exclude=[details.caller]))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'Failed to start native worker: {}'.format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg,
worker.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(_):
self.log.info("Node worker {} ended successfully".format(
worker.id))
worker.log_stats(0)
del self._workers[worker.id]
return True
def on_exit_error(err):
self.log.info("Node worker {} ended with error ({})".format(
worker.id, err))
worker.log_stats(0)
del self._workers[worker.id]
return False
def check_for_shutdown(was_successful):
shutdown = False
# automatically shutdown node whenever a worker ended (successfully, or with error)
#
if checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT in self._node._node_shutdown_triggers:
self.log.info(
"Node worker ended, and trigger '{}' active".format(
checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT))
shutdown = True
# automatically shutdown node when worker ended with error
#
if not was_successful and checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT_WITH_ERROR in self._node._node_shutdown_triggers:
self.log.info(
"Node worker ended with error, and trigger '{}' active".
format(
checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT_WITH_ERROR))
shutdown = True
# automatically shutdown node when no more workers are left
#
if len(
self._workers
) == 0 and checkconfig.NODE_SHUTDOWN_ON_LAST_WORKER_EXIT in self._node._node_shutdown_triggers:
self.log.info(
"No more node workers running, and trigger '{}' active".
format(checkconfig.NODE_SHUTDOWN_ON_LAST_WORKER_EXIT))
shutdown = True
# initiate shutdown (but only if we are not already shutting down)
#
if shutdown:
if not self._shutdown_requested:
self.log.info("Node shutting down ..")
self.shutdown()
else:
# ignore: shutdown already initiated ..
self.log.info("Node is already shutting down.")
else:
self.log.info(
"Node will continue to run (node shutdown triggers active: {})"
.format(self._node._node_shutdown_triggers))
d_on_exit = worker.exit.addCallbacks(on_exit_success, on_exit_error)
d_on_exit.addBoth(check_for_shutdown)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_native_worker_client_factory(
self._node._router_session_factory, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
'id': id,
'status': worker.status,
'created': utcstr(worker.created),
'who': worker.who
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic,
starting_info,
options=PublishOptions(exclude=[details.caller]))
# now actually fork the worker ..
#
self.log.info("Starting {worker} with ID '{id}'...",
worker=worker_logname,
id=id)
self.log.debug("{worker} '{id}' command line is '{cmdline}'",
worker=worker_logname,
id=id,
cmdline=' '.join(args))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will errback ..
pid = proto.transport.pid
self.log.debug("Native worker process connected with PID {pid}",
pid=pid)
# note the PID of the worker
worker.pid = pid
# proto is an instance of NativeWorkerClientProtocol
worker.proto = proto
worker.status = 'connected'
worker.connected = datetime.utcnow()
def on_connect_error(err):
# not sure when this errback is triggered at all ..
self.log.error(
"Interal error: connection to forked native worker failed ({err})",
err=err)
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def start_guest(self, id, config, details=None):
"""
Start a new guest process on this node.
:param config: The guest process configuration.
:type config: obj
:returns: int -- The PID of the new process.
"""
# prohibit starting a worker twice
#
if id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(
id)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.worker_already_running',
emsg)
try:
checkconfig.check_guest(config)
except Exception as e:
raise ApplicationError(
u'crossbar.error.invalid_configuration',
'invalid guest worker configuration: {}'.format(e))
options = config.get('options', {})
# guest process working directory
#
workdir = self._node._cbdir
if 'workdir' in options:
workdir = os.path.join(workdir, options['workdir'])
workdir = os.path.abspath(workdir)
# guest process executable and command line arguments
#
# first try to configure the fully qualified path for the guest
# executable by joining workdir and configured exectuable ..
exe = os.path.abspath(os.path.join(workdir, config['executable']))
if check_executable(exe):
self.log.info(
"Using guest worker executable '{exe}' (executable path taken from configuration)",
exe=exe)
else:
# try to detect the fully qualified path for the guest
# executable by doing a "which" on the configured executable name
exe = shutil.which(config['executable'])
if exe is not None and check_executable(exe):
self.log.info(
"Using guest worker executable '{exe}' (executable path detected from environment)",
exe=exe)
else:
emsg = "Could not start worker: could not find and executable for '{}'".format(
config['executable'])
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
# guest process command line arguments
#
args = [exe]
args.extend(config.get('arguments', []))
# guest process environment
#
worker_env = create_process_env(options)
# log name of worker
#
worker_logname = 'Guest'
# topic URIs used (later)
#
starting_topic = 'crossbar.node.{}.on_guest_starting'.format(
self._node_id)
started_topic = 'crossbar.node.{}.on_guest_started'.format(
self._node_id)
# add worker tracking instance to the worker map ..
#
worker = GuestWorkerProcess(self,
id,
details.caller,
keeplog=options.get('traceback', None))
self._workers[id] = worker
# create a (custom) process endpoint
#
ep = WorkerProcessEndpoint(self._node._reactor,
exe,
args,
path=workdir,
env=worker_env,
worker=worker)
# ready handling
#
def on_ready_success(proto):
worker.pid = proto.transport.pid
worker.status = 'started'
worker.started = datetime.utcnow()
self.log.info("{worker} with ID '{id}' and PID {pid} started",
worker=worker_logname,
id=worker.id,
pid=worker.pid)
self._node._reactor.addSystemEventTrigger(
'before',
'shutdown',
self._cleanup_worker,
self._node._reactor,
worker,
)
# directory watcher
#
if 'watch' in options:
if HAS_FSNOTIFY:
# assemble list of watched directories
watched_dirs = []
for d in options['watch'].get('directories', []):
watched_dirs.append(
os.path.abspath(os.path.join(self._node._cbdir,
d)))
worker.watch_timeout = options['watch'].get('timeout', 1)
# create a directory watcher
worker.watcher = DirWatcher(dirs=watched_dirs,
notify_once=True)
# make sure to stop the background thread running inside the
# watcher upon Twisted being shut down
def on_shutdown():
worker.watcher.stop()
self._node._reactor.addSystemEventTrigger(
'before', 'shutdown', on_shutdown)
# this handler will get fired by the watcher upon detecting an FS event
def on_fsevent(evt):
worker.watcher.stop()
proto.signal('TERM')
if options['watch'].get('action', None) == 'restart':
self.log.info("Restarting guest ..")
# Add a timeout large enough (perhaps add a config option later)
self._node._reactor.callLater(
worker.watch_timeout, self.start_guest, id,
config, details)
# Shut the worker down, after the restart event is scheduled
worker.stop()
# now run the watcher on a background thread
deferToThread(worker.watcher.loop, on_fsevent)
else:
self.log.warn(
"Warning: cannot watch directory for changes - feature DirWatcher unavailable"
)
# assemble guest worker startup information
#
started_info = {
'id': worker.id,
'status': worker.status,
'started': utcstr(worker.started),
'who': worker.who
}
self.publish(started_topic,
started_info,
options=PublishOptions(exclude=[details.caller]))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'Failed to start guest worker: {}'.format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg,
ep.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
self.log.info("Guest {id} exited with success", id=worker.id)
del self._workers[worker.id]
def on_exit_error(err):
self.log.error("Guest {id} exited with error {err.value}",
id=worker.id,
err=err)
del self._workers[worker.id]
worker.exit.addCallbacks(on_exit_success, on_exit_error)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_guest_worker_client_factory(
config, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
'id': id,
'status': worker.status,
'created': utcstr(worker.created),
'who': worker.who
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic,
starting_info,
options=PublishOptions(exclude=[details.caller]))
# now actually fork the worker ..
#
self.log.info("Starting {worker} with ID '{id}'...",
worker=worker_logname,
id=id)
self.log.debug("{worker} '{id}' using command line '{cli}'...",
worker=worker_logname,
id=id,
cli=' '.join(args))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will
# errback - probably only if the forking of a new process fails
# at OS level due to out of memory conditions or such.
pid = proto.transport.pid
self.log.debug("Guest worker process connected with PID {pid}",
pid=pid)
worker.pid = pid
# proto is an instance of GuestWorkerClientProtocol
worker.proto = proto
worker.status = 'connected'
worker.connected = datetime.utcnow()
def on_connect_error(err):
# not sure when this errback is triggered at all .. see above.
self.log.error(
"Internal error: connection to forked guest worker failed ({})"
.format(err))
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def _start_native_worker(self, wtype, id, options=None, details=None):
assert(wtype in ['router', 'container'])
# prohibit starting a worker twice
#
if id in self._workers:
emsg = "ERROR: could not start worker - a worker with ID '{}'' is already running (or starting)".format(id)
log.msg(emsg)
raise ApplicationError('crossbar.error.worker_already_running', emsg)
# check worker options
#
options = options or {}
try:
if wtype == 'router':
checkconfig.check_router_options(options)
elif wtype == 'container':
checkconfig.check_container_options(options)
else:
raise Exception("logic error")
except Exception as e:
emsg = "ERROR: could not start native worker - invalid configuration ({})".format(e)
log.msg(emsg)
raise ApplicationError('crossbar.error.invalid_configuration', emsg)
# allow override Python executable from options
#
if 'python' in options:
exe = options['python']
# the executable must be an absolute path, e.g. /home/oberstet/pypy-2.2.1-linux64/bin/pypy
#
if not os.path.isabs(exe):
emsg = "ERROR: python '{}' from worker options must be an absolute path".format(exe)
log.msg(emsg)
raise ApplicationError('crossbar.error.invalid_configuration', emsg)
# of course the path must exist and actually be executable
#
if not (os.path.isfile(exe) and os.access(exe, os.X_OK)):
emsg = "ERROR: python '{}' from worker options does not exist or isn't an executable".format(exe)
log.msg(emsg)
raise ApplicationError('crossbar.error.invalid_configuration', emsg)
else:
exe = sys.executable
# all native workers (routers and containers for now) start from the same script
#
filename = pkg_resources.resource_filename('crossbar', 'worker/process.py')
# assemble command line for forking the worker
#
args = [exe, "-u", filename]
args.extend(["--cbdir", self._node._cbdir])
args.extend(["--node", str(self._node_id)])
args.extend(["--worker", str(id)])
args.extend(["--realm", self._realm])
args.extend(["--type", wtype])
# allow override worker process title from options
#
if options.get('title', None):
args.extend(['--title', options['title']])
# allow overriding debug flag from options
#
if options.get('debug', self.debug):
args.append('--debug')
# forward explicit reactor selection
#
if 'reactor' in options and sys.platform in options['reactor']:
args.extend(['--reactor', options['reactor'][sys.platform]])
elif self._node.options.reactor:
args.extend(['--reactor', self._node.options.reactor])
# create worker process environment
#
worker_env = create_process_env(options)
# log name of worker
#
worker_logname = {'router': 'Router', 'container': 'Container'}.get(wtype, 'Worker')
# topic URIs used (later)
#
if wtype == 'router':
starting_topic = 'crossbar.node.{}.on_router_starting'.format(self._node_id)
started_topic = 'crossbar.node.{}.on_router_started'.format(self._node_id)
elif wtype == 'container':
starting_topic = 'crossbar.node.{}.on_container_starting'.format(self._node_id)
started_topic = 'crossbar.node.{}.on_container_started'.format(self._node_id)
else:
raise Exception("logic error")
# add worker tracking instance to the worker map ..
#
if wtype == 'router':
worker = RouterWorkerProcess(self, id, details.caller, keeplog=options.get('traceback', None))
elif wtype == 'container':
worker = ContainerWorkerProcess(self, id, details.caller, keeplog=options.get('traceback', None))
else:
raise Exception("logic error")
self._workers[id] = worker
# create a (custom) process endpoint
#
ep = WorkerProcessEndpoint(self._node._reactor, exe, args, env=worker_env, worker=worker)
# ready handling
#
def on_ready_success(id):
log.msg("{} with ID '{}' and PID {} started".format(worker_logname, worker.id, worker.pid))
def cleanup_worker():
try:
worker.proto.transport.signalProcess('TERM')
except ProcessExitedAlready:
pass # ignore; it's already dead
self._node._reactor.addSystemEventTrigger(
'before', 'shutdown',
cleanup_worker,
)
worker.status = 'started'
worker.started = datetime.utcnow()
started_info = {
'id': worker.id,
'status': worker.status,
'started': utcstr(worker.started),
'who': worker.who
}
self.publish(started_topic, started_info, options=PublishOptions(exclude=[details.caller]))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'ERROR: failed to start native worker - {}'.format(err.value)
log.msg(emsg)
raise ApplicationError("crossbar.error.cannot_start", emsg, worker.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
del self._workers[worker.id]
def on_exit_error(err):
del self._workers[worker.id]
worker.exit.addCallbacks(on_exit_success, on_exit_error)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_native_worker_client_factory(self._node._router_session_factory, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
'id': id,
'status': worker.status,
'created': utcstr(worker.created),
'who': worker.who
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic, starting_info, options=PublishOptions(exclude=[details.caller]))
# now actually fork the worker ..
#
if self.debug:
log.msg("Starting {} with ID '{}' using command line '{}' ..".format(worker_logname, id, ' '.join(args)))
else:
log.msg("Starting {} with ID '{}' ..".format(worker_logname, id))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will errback ..
pid = proto.transport.pid
if self.debug:
log.msg("Native worker process connected with PID {}".format(pid))
# note the PID of the worker
worker.pid = pid
# proto is an instance of NativeWorkerClientProtocol
worker.proto = proto
worker.status = 'connected'
worker.connected = datetime.utcnow()
def on_connect_error(err):
# not sure when this errback is triggered at all ..
if self.debug:
log.msg("ERROR: Connecting forked native worker failed - {}".format(err))
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def start_guest(self, id, config, details=None):
"""
Start a new guest process on this node.
:param config: The guest process configuration.
:type config: obj
:returns: int -- The PID of the new process.
"""
# prohibit starting a worker twice
#
if id in self._workers:
emsg = "ERROR: could not start worker - a worker with ID '{}' is already running (or starting)".format(id)
log.msg(emsg)
raise ApplicationError('crossbar.error.worker_already_running', emsg)
try:
checkconfig.check_guest(config)
except Exception as e:
raise ApplicationError('crossbar.error.invalid_configuration', 'invalid guest worker configuration: {}'.format(e))
options = config.get('options', {})
# guest process working directory
#
workdir = self._node._cbdir
if 'workdir' in options:
workdir = os.path.join(workdir, options['workdir'])
workdir = os.path.abspath(workdir)
# guest process executable and command line arguments
#
# first try to configure the fully qualified path for the guest
# executable by joining workdir and configured exectuable ..
exe = os.path.abspath(os.path.join(workdir, config['executable']))
if check_executable(exe):
log.msg("Using guest worker executable '{}' (executable path taken from configuration)".format(exe))
else:
# try to detect the fully qualified path for the guest
# executable by doing a "which" on the configured executable name
exe = shutil.which(config['executable'])
if exe is not None and check_executable(exe):
log.msg("Using guest worker executable '{}' (executable path detected from environment)".format(exe))
else:
emsg = "ERROR: could not start worker - could not find and executable for '{}'".format(config['executable'])
log.msg(emsg)
raise ApplicationError('crossbar.error.invalid_configuration', emsg)
# guest process command line arguments
#
args = [exe]
args.extend(config.get('arguments', []))
# guest process environment
#
worker_env = create_process_env(options)
# log name of worker
#
worker_logname = 'Guest'
# topic URIs used (later)
#
starting_topic = 'crossbar.node.{}.on_guest_starting'.format(self._node_id)
started_topic = 'crossbar.node.{}.on_guest_started'.format(self._node_id)
# add worker tracking instance to the worker map ..
#
worker = GuestWorkerProcess(self, id, details.caller, keeplog=options.get('traceback', None))
self._workers[id] = worker
# create a (custom) process endpoint
#
ep = WorkerProcessEndpoint(self._node._reactor, exe, args, path=workdir, env=worker_env, worker=worker)
# ready handling
#
def on_ready_success(proto):
worker.pid = proto.transport.pid
worker.status = 'started'
worker.started = datetime.utcnow()
log.msg("{} with ID '{}' and PID {} started".format(worker_logname, worker.id, worker.pid))
# directory watcher
#
if 'watch' in options:
if HAS_FSNOTIFY:
# assemble list of watched directories
watched_dirs = []
for d in options['watch'].get('directories', []):
watched_dirs.append(os.path.abspath(os.path.join(self._node._cbdir, d)))
# create a directory watcher
worker.watcher = DirWatcher(dirs=watched_dirs, notify_once=True)
# make sure to stop the background thread running inside the
# watcher upon Twisted being shut down
def on_shutdown():
worker.watcher.stop()
reactor.addSystemEventTrigger('before', 'shutdown', on_shutdown)
# this handler will get fired by the watcher upon detecting an FS event
def on_fsevent(evt):
worker.watcher.stop()
proto.signal('TERM')
if options['watch'].get('action', None) == 'restart':
log.msg("Restarting guest ..")
reactor.callLater(0.1, self.start_guest, id, config, details)
# now run the watcher on a background thread
deferToThread(worker.watcher.loop, on_fsevent)
else:
log.msg("Warning: cannot watch directory for changes - feature DirWatcher unavailable")
# assemble guest worker startup information
#
started_info = {
'id': worker.id,
'status': worker.status,
'started': utcstr(worker.started),
'who': worker.who
}
self.publish(started_topic, started_info, options=PublishOptions(exclude=[details.caller]))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'ERROR: failed to start guest worker - {}'.format(err.value)
log.msg(emsg)
raise ApplicationError("crossbar.error.cannot_start", emsg, ep.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
log.msg("Guest excited with success")
del self._workers[worker.id]
def on_exit_error(err):
log.msg("Guest excited with error", err)
del self._workers[worker.id]
worker.exit.addCallbacks(on_exit_success, on_exit_error)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_guest_worker_client_factory(config, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
'id': id,
'status': worker.status,
'created': utcstr(worker.created),
'who': worker.who
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic, starting_info, options=PublishOptions(exclude=[details.caller]))
# now actually fork the worker ..
#
if self.debug:
log.msg("Starting {} with ID '{}' using command line '{}' ..".format(worker_logname, id, ' '.join(args)))
else:
log.msg("Starting {} with ID '{}' ..".format(worker_logname, id))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will
# errback - probably only if the forking of a new process fails
# at OS level due to out of memory conditions or such.
pid = proto.transport.pid
if self.debug:
log.msg("Guest worker process connected with PID {}".format(pid))
worker.pid = pid
# proto is an instance of GuestWorkerClientProtocol
worker.proto = proto
worker.status = 'connected'
worker.connected = datetime.utcnow()
def on_connect_error(err):
# not sure when this errback is triggered at all .. see above.
if self.debug:
log.msg("ERROR: Connecting forked guest worker failed - {}".format(err))
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def _start_guest_worker(self, worker_id, worker_config, details=None):
"""
Start a new guest process on this node.
:param config: The guest process configuration.
:type config: obj
:returns: int -- The PID of the new process.
"""
# prohibit starting a worker twice
#
if worker_id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(
worker_id)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.worker_already_running',
emsg)
try:
checkconfig.check_guest(worker_config)
except Exception as e:
raise ApplicationError(
u'crossbar.error.invalid_configuration',
'invalid guest worker configuration: {}'.format(e))
options = worker_config.get('options', {})
# guest process working directory
#
workdir = self._node._cbdir
if 'workdir' in options:
workdir = os.path.join(workdir, options['workdir'])
workdir = os.path.abspath(workdir)
# guest process executable and command line arguments
#
# first try to configure the fully qualified path for the guest
# executable by joining workdir and configured exectuable ..
exe = os.path.abspath(
os.path.join(workdir, worker_config['executable']))
if check_executable(exe):
self.log.info(
"Using guest worker executable '{exe}' (executable path taken from configuration)",
exe=exe)
else:
# try to detect the fully qualified path for the guest
# executable by doing a "which" on the configured executable name
exe = which(worker_config['executable'])
if exe is not None and check_executable(exe):
self.log.info(
"Using guest worker executable '{exe}' (executable path detected from environment)",
exe=exe)
else:
emsg = "Could not start worker: could not find and executable for '{}'".format(
worker_config['executable'])
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
# guest process command line arguments
#
args = [exe]
args.extend(worker_config.get('arguments', []))
# guest process environment
#
worker_env = create_process_env(options)
# log name of worker
#
worker_logname = 'Guest'
# topic URIs used (later)
#
starting_topic = u'{}.on_guest_starting'.format(self._uri_prefix)
started_topic = u'{}.on_guest_started'.format(self._uri_prefix)
# add worker tracking instance to the worker map ..
#
worker = GuestWorkerProcess(self,
worker_id,
details.caller,
keeplog=options.get('traceback', None))
self._workers[worker_id] = worker
# create a (custom) process endpoint
#
ep = WorkerProcessEndpoint(self._node._reactor,
exe,
args,
path=workdir,
env=worker_env,
worker=worker)
# ready handling
#
def on_ready_success(proto):
self.log.info('{worker_logname} worker "{worker_id}" started',
worker_logname=worker_logname,
worker_id=worker.id)
worker.on_worker_started(proto)
self._node._reactor.addSystemEventTrigger(
'before',
'shutdown',
self._cleanup_worker,
self._node._reactor,
worker,
)
# directory watcher
#
if 'watch' in options:
if HAS_FS_WATCHER:
# assemble list of watched directories
watched_dirs = []
for d in options['watch'].get('directories', []):
watched_dirs.append(
os.path.abspath(os.path.join(self._node._cbdir,
d)))
worker.watch_timeout = options['watch'].get('timeout', 1)
# create a filesystem watcher
worker.watcher = FilesystemWatcher(
workdir, watched_dirs=watched_dirs)
# make sure to stop the watch upon Twisted being shut down
def on_shutdown():
worker.watcher.stop()
self._node._reactor.addSystemEventTrigger(
'before', 'shutdown', on_shutdown)
# this handler will get fired by the watcher upon detecting an FS event
def on_filesystem_change(fs_event):
worker.watcher.stop()
proto.signal('TERM')
if options['watch'].get('action', None) == 'restart':
self.log.info(
"Filesystem watcher detected change {fs_event} - restarting guest in {watch_timeout} seconds ..",
fs_event=fs_event,
watch_timeout=worker.watch_timeout)
# Add a timeout large enough (perhaps add a config option later)
self._node._reactor.callLater(
worker.watch_timeout, self.start_guest,
worker_id, worker_config, details)
# Shut the worker down, after the restart event is scheduled
# FIXME: all workers should have a stop() method ..
# -> 'GuestWorkerProcess' object has no attribute 'stop'
# worker.stop()
else:
self.log.info(
"Filesystem watcher detected change {fs_event} - no action taken!",
fs_event=fs_event)
# now start watching ..
worker.watcher.start(on_filesystem_change)
else:
self.log.warn(
"Cannot watch directories for changes - feature not available"
)
# assemble guest worker startup information
#
started_info = {
u'id': worker.id,
u'status': worker.status,
u'started': utcstr(worker.started),
u'who': worker.who,
}
self.publish(started_topic,
started_info,
options=PublishOptions(exclude=details.caller))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'Failed to start guest worker: {}'.format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg,
ep.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
self.log.info("Guest {worker_id} exited with success",
worker_id=worker.id)
del self._workers[worker.id]
def on_exit_error(err):
self.log.error("Guest {worker_id} exited with error {err.value}",
worker_id=worker.id,
err=err)
del self._workers[worker.id]
worker.exit.addCallbacks(on_exit_success, on_exit_error)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_guest_worker_client_factory(
worker_config, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[worker_id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
u'id': worker_id,
u'status': worker.status,
u'created': utcstr(worker.created),
u'who': worker.who,
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic,
starting_info,
options=PublishOptions(exclude=details.caller))
# now actually fork the worker ..
#
self.log.info('{worker_logname} "{worker_id}" process starting ..',
worker_logname=worker_logname,
worker_id=worker_id)
self.log.debug(
'{worker_logname} "{worker_id}" process using command line "{cli}" ..',
worker_logname=worker_logname,
worker_id=worker_id,
cli=' '.join(args))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will
# errback - probably only if the forking of a new process fails
# at OS level due to out of memory conditions or such.
self.log.debug('{worker_logname} "{worker_id}" connected',
worker_logname=worker_logname,
worker_id=worker_id)
# do not comment this: it will lead to on_worker_started being called
# _before_ on_worker_connected, and we don't need it!
# worker.on_worker_connected(proto)
def on_connect_error(err):
# not sure when this errback is triggered at all .. see above.
self.log.failure(
"Internal error: connection to forked guest worker failed ({log_failure.value})",
)
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def _start_native_worker(self, wtype, id, options=None, details=None):
assert (wtype in ['router', 'container'])
## prohibit starting a worker twice
##
if id in self._workers:
emsg = "ERROR: could not start worker - a worker with ID '{}'' is already running (or starting)".format(
id)
log.msg(emsg)
raise ApplicationError('crossbar.error.worker_already_running',
emsg)
## check worker options
##
options = options or {}
try:
if wtype == 'router':
checkconfig.check_router_options(options)
elif wtype == 'container':
checkconfig.check_container_options(options)
else:
raise Exception("logic error")
except Exception as e:
emsg = "ERROR: could not start native worker - invalid configuration ({})".format(
e)
log.msg(emsg)
raise ApplicationError('crossbar.error.invalid_configuration',
emsg)
## allow override Python executable from options
##
if 'python' in options:
exe = options['python']
## the executable must be an absolute path, e.g. /home/oberstet/pypy-2.2.1-linux64/bin/pypy
##
if not os.path.isabs(exe):
emsg = "ERROR: python '{}' from worker options must be an absolute path".format(
exe)
log.msg(emsg)
raise ApplicationError('crossbar.error.invalid_configuration',
emsg)
## of course the path must exist and actually be executable
##
if not (os.path.isfile(exe) and os.access(exe, os.X_OK)):
emsg = "ERROR: python '{}' from worker options does not exist or isn't an executable".format(
exe)
log.msg(emsg)
raise ApplicationError('crossbar.error.invalid_configuration',
emsg)
else:
exe = sys.executable
## all native workers (routers and containers for now) start from the same script
##
filename = pkg_resources.resource_filename('crossbar',
'worker/process.py')
## assemble command line for forking the worker
##
args = [exe, "-u", filename]
args.extend(["--cbdir", self._node._cbdir])
args.extend(["--node", str(self._node_id)])
args.extend(["--worker", str(id)])
args.extend(["--realm", self._realm])
args.extend(["--type", wtype])
## allow override worker process title from options
##
if options.get('title', None):
args.extend(['--title', options['title']])
## allow overriding debug flag from options
##
if options.get('debug', self.debug):
args.append('--debug')
## forward explicit reactor selection
##
if 'reactor' in options and sys.platform in options['reactor']:
args.extend(['--reactor', options['reactor'][sys.platform]])
elif self._node.options.reactor:
args.extend(['--reactor', self._node.options.reactor])
## create worker process environment
##
worker_env = create_process_env(options)
## log name of worker
##
worker_logname = {
'router': 'Router',
'container': 'Container'
}.get(wtype, 'Worker')
## topic URIs used (later)
##
if wtype == 'router':
starting_topic = 'crossbar.node.{}.on_router_starting'.format(
self._node_id)
started_topic = 'crossbar.node.{}.on_router_started'.format(
self._node_id)
elif wtype == 'container':
starting_topic = 'crossbar.node.{}.on_container_starting'.format(
self._node_id)
started_topic = 'crossbar.node.{}.on_container_started'.format(
self._node_id)
else:
raise Exception("logic error")
## add worker tracking instance to the worker map ..
##
if wtype == 'router':
worker = RouterWorkerProcess(self,
id,
details.authid,
keeplog=options.get(
'traceback', None))
elif wtype == 'container':
worker = ContainerWorkerProcess(self,
id,
details.authid,
keeplog=options.get(
'traceback', None))
else:
raise Exception("logic error")
self._workers[id] = worker
## create a (custom) process endpoint
##
ep = WorkerProcessEndpoint(self._node._reactor,
exe,
args,
env=worker_env,
worker=worker)
## ready handling
##
def on_ready_success(id):
log.msg("{} with ID '{}' and PID {} started".format(
worker_logname, worker.id, worker.pid))
worker.status = 'started'
worker.started = datetime.utcnow()
started_info = {
'id': worker.id,
'status': worker.status,
'started': utcstr(worker.started),
'who': worker.who
}
self.publish(started_topic,
started_info,
options=PublishOptions(exclude=[details.caller]))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'ERROR: failed to start native worker - {}'.format(
err.value)
log.msg(emsg)
raise ApplicationError("crossbar.error.cannot_start", emsg,
worker.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
del self._workers[worker.id]
def on_exit_error(err):
del self._workers[worker.id]
worker.exit.addCallbacks(on_exit_success, on_exit_error)
## create a transport factory for talking WAMP to the native worker
##
transport_factory = create_native_worker_client_factory(
self._node._router_session_factory, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
## now (immediately before actually forking) signal the starting of the worker
##
starting_info = {
'id': id,
'status': worker.status,
'created': utcstr(worker.created),
'who': worker.who
}
## the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
## .. while all others get an event
self.publish(starting_topic,
starting_info,
options=PublishOptions(exclude=[details.caller]))
## now actually fork the worker ..
##
if self.debug:
log.msg(
"Starting {} with ID '{}' using command line '{}' ..".format(
worker_logname, id, ' '.join(args)))
else:
log.msg("Starting {} with ID '{}' ..".format(worker_logname, id))
d = ep.connect(transport_factory)
def on_connect_success(proto):
## this seems to be called immediately when the child process
## has been forked. even if it then immediately fails because
## e.g. the executable doesn't even exist. in other words,
## I'm not sure under what conditions the deferred will errback ..
pid = proto.transport.pid
if self.debug:
log.msg(
"Native worker process connected with PID {}".format(pid))
## note the PID of the worker
worker.pid = pid
## proto is an instance of NativeWorkerClientProtocol
worker.proto = proto
worker.status = 'connected'
worker.connected = datetime.utcnow()
def on_connect_error(err):
## not sure when this errback is triggered at all ..
if self.debug:
log.msg("ERROR: Connecting forked native worker failed - {}".
format(err))
## in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def _start_native_worker(self,
worker_type,
worker_id,
worker_options=None,
details=None):
# prohibit starting a worker twice
#
if worker_id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(
worker_id)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.worker_already_running',
emsg)
# check worker options
#
options = worker_options or {}
try:
if worker_type in self._node._native_workers:
if self._node._native_workers[worker_type][
'checkconfig_options']:
self._node._native_workers[worker_type][
'checkconfig_options'](options)
else:
raise Exception(
'No checkconfig_options for worker type "{worker_type}" implemented!'
.format(worker_type=worker_type))
else:
raise Exception('invalid worker type "{}"'.format(worker_type))
except Exception as e:
emsg = "Could not start native worker: invalid configuration ({})".format(
e)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
# the fully qualified worker class as a string
worker_class = qual(
self._node._native_workers[worker_type]['worker_class'])
# allow override Python executable from options
#
if 'python' in options:
exe = options['python']
# the executable must be an absolute path, e.g. /home/oberstet/pypy-2.2.1-linux64/bin/pypy
#
if not os.path.isabs(exe):
emsg = "Invalid worker configuration: python executable '{}' must be an absolute path".format(
exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
# of course the path must exist and actually be executable
#
if not (os.path.isfile(exe) and os.access(exe, os.X_OK)):
emsg = "Invalid worker configuration: python executable '{}' does not exist or isn't an executable".format(
exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration',
emsg)
else:
exe = sys.executable
# allow override default Python module search paths from options
#
if 'pythonpath' in options:
pythonpaths_to_add = [
os.path.abspath(os.path.join(self._node._cbdir, p))
for p in options.get('pythonpath', [])
]
else:
pythonpaths_to_add = []
# assemble command line for forking the worker
#
# all native workers (routers and containers for now) start
# from the same script in crossbar/worker/process.py -- we're
# invoking via "-m" so that .pyc files, __pycache__ etc work
# properly.
#
args = [exe, "-u", "-m", "crossbar.worker.process"]
args.extend(["--cbdir", self._node._cbdir])
args.extend(["--worker", str(worker_id)])
args.extend(["--realm", self._realm])
args.extend(["--klass", worker_class])
args.extend(["--loglevel", get_global_log_level()])
if "shutdown" in options:
args.extend(["--shutdown", options["shutdown"]])
# Node-level callback to inject worker arguments
#
self._node._extend_worker_args(args, options)
# allow override worker process title from options
#
if options.get('title', None):
args.extend(['--title', options['title']])
# forward explicit reactor selection
#
if 'reactor' in options and sys.platform in options['reactor']:
args.extend(['--reactor', options['reactor'][sys.platform]])
# FIXME
# elif self._node.options.reactor:
# args.extend(['--reactor', self._node.options.reactor])
# create worker process environment
#
worker_env = create_process_env(options)
# We need to use the same PYTHONPATH we were started with, so we can
# find the Crossbar we're working with -- it may not be the same as the
# one on the default path
worker_env["PYTHONPATH"] = os.pathsep.join(pythonpaths_to_add +
sys.path)
# log name of worker
#
worker_logname = self._node._native_workers[worker_type]['logname']
# each worker is run under its own dedicated WAMP auth role
#
worker_auth_role = u'crossbar.worker.{}'.format(worker_id)
# topic URIs used (later)
#
starting_topic = self._node._native_workers[worker_type]['topics'][
'starting']
started_topic = self._node._native_workers[worker_type]['topics'][
'started']
# add worker tracking instance to the worker map ..
#
WORKER = self._node._native_workers[worker_type]['class']
worker = WORKER(self,
worker_id,
details.caller,
keeplog=options.get('traceback', None))
self._workers[worker_id] = worker
# create a (custom) process endpoint.
#
if platform.isWindows():
childFDs = None # Use the default Twisted ones
else:
# The communication between controller and container workers is
# using WAMP running over 2 pipes.
# For controller->container traffic this runs over FD 0 (`stdin`)
# and for the container->controller traffic, this runs over FD 3.
#
# Note: We use FD 3, not FD 1 (`stdout`) or FD 2 (`stderr`) for
# container->controller traffic, so that components running in the
# container which happen to write to `stdout` or `stderr` do not
# interfere with the container-controller communication.
childFDs = {0: "w", 1: "r", 2: "r", 3: "r"}
ep = WorkerProcessEndpoint(self._node._reactor,
exe,
args,
env=worker_env,
worker=worker,
childFDs=childFDs)
# ready handling
#
def on_ready_success(worker_id):
self.log.info(
'{worker_type} worker "{worker_id}" process {pid} started',
worker_type=worker_logname,
worker_id=worker.id,
pid=worker.pid)
self._node._reactor.addSystemEventTrigger(
'before',
'shutdown',
self._cleanup_worker,
self._node._reactor,
worker,
)
worker.on_worker_started()
started_info = {
u'id': worker.id,
u'status': worker.status,
u'started': utcstr(worker.started),
u'who': worker.who,
}
# FIXME: make start of stats printer dependent on log level ..
if False:
worker.log_stats(5.)
self.publish(started_topic,
started_info,
options=PublishOptions(exclude=details.caller))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'Failed to start native worker: {}'.format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg,
worker.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(_):
self.log.info("Node worker {worker.id} ended successfully",
worker=worker)
# clear worker log
worker.log_stats(0)
# remove the dedicated node router authrole we dynamically
# added for the worker
self._node._drop_worker_role(worker_auth_role)
# remove our metadata tracking for the worker
del self._workers[worker.id]
# indicate that the worker excited successfully
return True
def on_exit_error(err):
self.log.info("Node worker {worker.id} ended with error ({err})",
worker=worker,
err=err)
# clear worker log
worker.log_stats(0)
# remove the dedicated node router authrole we dynamically
# added for the worker
self._node._drop_worker_role(worker_auth_role)
# remove our metadata tracking for the worker
del self._workers[worker.id]
# indicate that the worker excited with error
return False
def check_for_shutdown(was_successful):
self.log.info(
'Checking for node shutdown: worker_exit_success={worker_exit_success}, shutdown_requested={shutdown_requested}, node_shutdown_triggers={node_shutdown_triggers}',
worker_exit_success=was_successful,
shutdown_requested=self._shutdown_requested,
node_shutdown_triggers=self._node._node_shutdown_triggers)
shutdown = self._shutdown_requested
# automatically shutdown node whenever a worker ended (successfully, or with error)
#
if checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT in self._node._node_shutdown_triggers:
self.log.info(
"Node worker ended, and trigger '{trigger}' active",
trigger=checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT)
shutdown = True
# automatically shutdown node when worker ended with error
#
if not was_successful and checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT_WITH_ERROR in self._node._node_shutdown_triggers:
self.log.info(
"Node worker ended with error, and trigger '{trigger}' active",
trigger=checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT_WITH_ERROR
)
shutdown = True
# automatically shutdown node when no more workers are left
#
if len(
self._workers
) == 0 and checkconfig.NODE_SHUTDOWN_ON_LAST_WORKER_EXIT in self._node._node_shutdown_triggers:
self.log.info(
"No more node workers running, and trigger '{trigger}' active",
trigger=checkconfig.NODE_SHUTDOWN_ON_LAST_WORKER_EXIT)
shutdown = True
# initiate shutdown (but only if we are not already shutting down)
#
if shutdown:
self.shutdown()
else:
self.log.info('Node will continue to run!')
d_on_exit = worker.exit.addCallbacks(on_exit_success, on_exit_error)
d_on_exit.addBoth(check_for_shutdown)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_native_worker_client_factory(
self._node._router_session_factory, worker_auth_role, worker.ready,
worker.exit)
transport_factory.noisy = False
self._workers[worker_id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
u'id': worker_id,
u'status': worker.status,
u'created': utcstr(worker.created),
u'who': worker.who,
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic,
starting_info,
options=PublishOptions(exclude=details.caller))
# now actually fork the worker ..
#
self.log.info('{worker_logname} worker "{worker_id}" starting ..',
worker_logname=worker_logname,
worker_id=worker_id)
self.log.debug(
'{worker_logname} "{worker_id}" command line is "{cmdline}"',
worker_logname=worker_logname,
worker_id=worker_id,
cmdline=' '.join(args))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will errback ..
self.log.debug('Native worker "{worker_id}" connected',
worker_id=worker_id)
worker.on_worker_connected(proto)
# dynamically add a dedicated authrole to the router
# for the worker we've just started
self._node._add_worker_role(worker_auth_role, options)
def on_connect_error(err):
# not sure when this errback is triggered at all ..
self.log.error(
"Interal error: connection to forked native worker failed ({err})",
err=err)
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def _start_native_worker(self, wtype, id, options=None, details=None):
assert(wtype in ['router', 'container', 'websocket-testee'])
# prohibit starting a worker twice
#
if id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(id)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.worker_already_running', emsg)
# check worker options
#
options = options or {}
try:
if wtype == 'router':
checkconfig.check_router_options(options)
elif wtype == 'container':
checkconfig.check_container_options(options)
elif wtype == 'websocket-testee':
checkconfig.check_websocket_testee_options(options)
else:
raise Exception("logic error")
except Exception as e:
emsg = "Could not start native worker: invalid configuration ({})".format(e)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration', emsg)
# allow override Python executable from options
#
if 'python' in options:
exe = options['python']
# the executable must be an absolute path, e.g. /home/oberstet/pypy-2.2.1-linux64/bin/pypy
#
if not os.path.isabs(exe):
emsg = "Invalid worker configuration: python executable '{}' must be an absolute path".format(exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration', emsg)
# of course the path must exist and actually be executable
#
if not (os.path.isfile(exe) and os.access(exe, os.X_OK)):
emsg = "Invalid worker configuration: python executable '{}' does not exist or isn't an executable".format(exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration', emsg)
else:
exe = sys.executable
# all native workers (routers and containers for now) start from the same script
#
filename = FilePath(crossbar.__file__).parent().child("worker").child("process.py").path
# assemble command line for forking the worker
#
args = [exe, "-u", filename]
args.extend(["--cbdir", self._node._cbdir])
args.extend(["--node", str(self._node_id)])
args.extend(["--worker", str(id)])
args.extend(["--realm", self._realm])
args.extend(["--type", wtype])
args.extend(["--loglevel", _loglevel])
# allow override worker process title from options
#
if options.get('title', None):
args.extend(['--title', options['title']])
# forward explicit reactor selection
#
if 'reactor' in options and sys.platform in options['reactor']:
args.extend(['--reactor', options['reactor'][sys.platform]])
elif self._node.options.reactor:
args.extend(['--reactor', self._node.options.reactor])
# create worker process environment
#
worker_env = create_process_env(options)
# We need to use the same PYTHONPATH we were started with, so we can
# find the Crossbar we're working with -- it may not be the same as the
# one on the default path
worker_env["PYTHONPATH"] = os.pathsep.join(sys.path)
# log name of worker
#
worker_logname = {
'router': 'Router',
'container': 'Container',
'websocket-testee': 'WebSocketTestee'
}.get(wtype, 'Worker')
# topic URIs used (later)
#
if wtype == 'router':
starting_topic = 'crossbar.node.{}.on_router_starting'.format(self._node_id)
started_topic = 'crossbar.node.{}.on_router_started'.format(self._node_id)
elif wtype == 'container':
starting_topic = 'crossbar.node.{}.on_container_starting'.format(self._node_id)
started_topic = 'crossbar.node.{}.on_container_started'.format(self._node_id)
elif wtype == 'websocket-testee':
starting_topic = 'crossbar.node.{}.on_websocket_testee_starting'.format(self._node_id)
started_topic = 'crossbar.node.{}.on_websocket_testee_started'.format(self._node_id)
else:
raise Exception("logic error")
# add worker tracking instance to the worker map ..
#
if wtype == 'router':
worker = RouterWorkerProcess(self, id, details.caller, keeplog=options.get('traceback', None))
elif wtype == 'container':
worker = ContainerWorkerProcess(self, id, details.caller, keeplog=options.get('traceback', None))
elif wtype == 'websocket-testee':
worker = WebSocketTesteeWorkerProcess(self, id, details.caller, keeplog=options.get('traceback', None))
else:
raise Exception("logic error")
self._workers[id] = worker
# create a (custom) process endpoint.
#
if platform.isWindows():
childFDs = None # Use the default Twisted ones
else:
# The communication between controller and container workers is
# using WAMP running over 2 pipes.
# For controller->container traffic this runs over FD 0 (`stdin`)
# and for the container->controller traffic, this runs over FD 3.
#
# Note: We use FD 3, not FD 1 (`stdout`) or FD 2 (`stderr`) for
# container->controller traffic, so that components running in the
# container which happen to write to `stdout` or `stderr` do not
# interfere with the container-controller communication.
childFDs = {0: "w", 1: "r", 2: "r", 3: "r"}
ep = WorkerProcessEndpoint(
self._node._reactor, exe, args, env=worker_env, worker=worker,
childFDs=childFDs)
# ready handling
#
def on_ready_success(id):
self.log.info("{worker} with ID '{id}' and PID {pid} started",
worker=worker_logname, id=worker.id, pid=worker.pid)
self._node._reactor.addSystemEventTrigger(
'before', 'shutdown',
self._cleanup_worker, self._node._reactor, worker,
)
worker.status = 'started'
worker.started = datetime.utcnow()
started_info = {
'id': worker.id,
'status': worker.status,
'started': utcstr(worker.started),
'who': worker.who
}
# FIXME: make start of stats printer dependent on log level ..
worker.log_stats(5.)
self.publish(started_topic, started_info, options=PublishOptions(exclude=[details.caller]))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'Failed to start native worker: {}'.format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg, worker.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
worker.log_stats(0)
del self._workers[worker.id]
return worker.id
def on_exit_error(err):
worker.log_stats(0)
del self._workers[worker.id]
return worker.id
def check_for_shutdown(worker_id):
shutdown = True
if not self._workers:
shutdown = True
self.log.info("Node worker {} ended ({} workers left)".format(worker_id, len(self._workers)))
if shutdown:
if not self._shutdown_requested:
self.log.info("Node shutting down ..")
self._shutdown_requested = True
self.shutdown()
else:
# shutdown already initiated
pass
d_on_exit = worker.exit.addCallbacks(on_exit_success, on_exit_error)
d_on_exit.addBoth(check_for_shutdown)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_native_worker_client_factory(self._node._router_session_factory, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
'id': id,
'status': worker.status,
'created': utcstr(worker.created),
'who': worker.who
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic, starting_info, options=PublishOptions(exclude=[details.caller]))
# now actually fork the worker ..
#
self.log.info("Starting {worker} with ID '{id}'...",
worker=worker_logname, id=id)
self.log.debug("{worker} '{id}' command line is '{cmdline}'",
worker=worker_logname, id=id, cmdline=' '.join(args))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will errback ..
pid = proto.transport.pid
self.log.debug("Native worker process connected with PID {pid}",
pid=pid)
# note the PID of the worker
worker.pid = pid
# proto is an instance of NativeWorkerClientProtocol
worker.proto = proto
worker.status = 'connected'
worker.connected = datetime.utcnow()
def on_connect_error(err):
# not sure when this errback is triggered at all ..
self.log.error("Interal error: connection to forked native worker failed ({err})", err=err)
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def start_guest(self, id, config, details=None):
"""
Start a new guest process on this node.
:param config: The guest process configuration.
:type config: obj
:returns: int -- The PID of the new process.
"""
# prohibit starting a worker twice
#
if id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(id)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.worker_already_running', emsg)
try:
checkconfig.check_guest(config)
except Exception as e:
raise ApplicationError(u'crossbar.error.invalid_configuration', 'invalid guest worker configuration: {}'.format(e))
options = config.get('options', {})
# guest process working directory
#
workdir = self._node._cbdir
if 'workdir' in options:
workdir = os.path.join(workdir, options['workdir'])
workdir = os.path.abspath(workdir)
# guest process executable and command line arguments
#
# first try to configure the fully qualified path for the guest
# executable by joining workdir and configured exectuable ..
exe = os.path.abspath(os.path.join(workdir, config['executable']))
if check_executable(exe):
self.log.info("Using guest worker executable '{exe}' (executable path taken from configuration)",
exe=exe)
else:
# try to detect the fully qualified path for the guest
# executable by doing a "which" on the configured executable name
exe = shutil.which(config['executable'])
if exe is not None and check_executable(exe):
self.log.info("Using guest worker executable '{exe}' (executable path detected from environment)",
exe=exe)
else:
emsg = "Could not start worker: could not find and executable for '{}'".format(config['executable'])
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration', emsg)
# guest process command line arguments
#
args = [exe]
args.extend(config.get('arguments', []))
# guest process environment
#
worker_env = create_process_env(options)
# log name of worker
#
worker_logname = 'Guest'
# topic URIs used (later)
#
starting_topic = 'crossbar.node.{}.on_guest_starting'.format(self._node_id)
started_topic = 'crossbar.node.{}.on_guest_started'.format(self._node_id)
# add worker tracking instance to the worker map ..
#
worker = GuestWorkerProcess(self, id, details.caller, keeplog=options.get('traceback', None))
self._workers[id] = worker
# create a (custom) process endpoint
#
ep = WorkerProcessEndpoint(self._node._reactor, exe, args, path=workdir, env=worker_env, worker=worker)
# ready handling
#
def on_ready_success(proto):
worker.pid = proto.transport.pid
worker.status = 'started'
worker.started = datetime.utcnow()
self.log.info("{worker} with ID '{id}' and PID {pid} started",
worker=worker_logname, id=worker.id, pid=worker.pid)
self._node._reactor.addSystemEventTrigger(
'before', 'shutdown',
self._cleanup_worker, self._node._reactor, worker,
)
# directory watcher
#
if 'watch' in options:
if HAS_FS_WATCHER:
# assemble list of watched directories
watched_dirs = []
for d in options['watch'].get('directories', []):
watched_dirs.append(os.path.abspath(os.path.join(self._node._cbdir, d)))
worker.watch_timeout = options['watch'].get('timeout', 1)
# create a filesystem watcher
worker.watcher = FilesystemWatcher(workdir, watched_dirs=watched_dirs)
# make sure to stop the watch upon Twisted being shut down
def on_shutdown():
worker.watcher.stop()
self._node._reactor.addSystemEventTrigger('before', 'shutdown', on_shutdown)
# this handler will get fired by the watcher upon detecting an FS event
def on_filesystem_change(fs_event):
worker.watcher.stop()
proto.signal('TERM')
if options['watch'].get('action', None) == 'restart':
self.log.info("Filesystem watcher detected change {fs_event} - restarting guest in {watch_timeout} seconds ..", fs_event=fs_event, watch_timeout=worker.watch_timeout)
# Add a timeout large enough (perhaps add a config option later)
self._node._reactor.callLater(worker.watch_timeout, self.start_guest, id, config, details)
# Shut the worker down, after the restart event is scheduled
# FIXME: all workers should have a stop() method ..
# -> 'GuestWorkerProcess' object has no attribute 'stop'
# worker.stop()
else:
self.log.info("Filesystem watcher detected change {fs_event} - no action taken!", fs_event=fs_event)
# now start watching ..
worker.watcher.start(on_filesystem_change)
else:
self.log.warn("Cannot watch directories for changes - feature not available")
# assemble guest worker startup information
#
started_info = {
u'id': worker.id,
u'status': worker.status,
u'started': utcstr(worker.started),
u'who': worker.who,
}
self.publish(started_topic, started_info, options=PublishOptions(exclude=details.caller))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'Failed to start guest worker: {}'.format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg, ep.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
self.log.info("Guest {id} exited with success", id=worker.id)
del self._workers[worker.id]
def on_exit_error(err):
self.log.error("Guest {id} exited with error {err.value}",
id=worker.id, err=err)
del self._workers[worker.id]
worker.exit.addCallbacks(on_exit_success, on_exit_error)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_guest_worker_client_factory(config, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
u'id': id,
u'status': worker.status,
u'created': utcstr(worker.created),
u'who': worker.who,
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic, starting_info, options=PublishOptions(exclude=details.caller))
# now actually fork the worker ..
#
self.log.info("Starting {worker} with ID '{id}'...",
worker=worker_logname, id=id)
self.log.debug("{worker} '{id}' using command line '{cli}'...",
worker=worker_logname, id=id, cli=' '.join(args))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will
# errback - probably only if the forking of a new process fails
# at OS level due to out of memory conditions or such.
pid = proto.transport.pid
self.log.debug("Guest worker process connected with PID {pid}",
pid=pid)
worker.pid = pid
# proto is an instance of GuestWorkerClientProtocol
worker.proto = proto
worker.status = 'connected'
worker.connected = datetime.utcnow()
def on_connect_error(err):
# not sure when this errback is triggered at all .. see above.
self.log.failure(
"Internal error: connection to forked guest worker failed ({log_failure.value})",
)
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def _start_native_worker(self, worker_type, worker_id, worker_options=None, details=None):
# prohibit starting a worker twice
#
if worker_id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(worker_id)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.worker_already_running', emsg)
# check worker options
#
options = worker_options or {}
try:
if worker_type in self._node._native_workers:
if self._node._native_workers[worker_type]['checkconfig_options']:
self._node._native_workers[worker_type]['checkconfig_options'](options)
else:
raise Exception('No checkconfig_options for worker type "{worker_type}" implemented!'.format(worker_type=worker_type))
else:
raise Exception('invalid worker type "{}"'.format(worker_type))
except Exception as e:
emsg = "Could not start native worker: invalid configuration ({})".format(e)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration', emsg)
# the fully qualified worker class as a string
worker_class = qual(self._node._native_workers[worker_type]['worker_class'])
# allow override Python executable from options
#
if 'python' in options:
exe = options['python']
# the executable must be an absolute path, e.g. /home/oberstet/pypy-2.2.1-linux64/bin/pypy
#
if not os.path.isabs(exe):
emsg = "Invalid worker configuration: python executable '{}' must be an absolute path".format(exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration', emsg)
# of course the path must exist and actually be executable
#
if not (os.path.isfile(exe) and os.access(exe, os.X_OK)):
emsg = "Invalid worker configuration: python executable '{}' does not exist or isn't an executable".format(exe)
self.log.error(emsg)
raise ApplicationError(u'crossbar.error.invalid_configuration', emsg)
else:
exe = sys.executable
# allow override default Python module search paths from options
#
if 'pythonpath' in options:
pythonpaths_to_add = [os.path.abspath(os.path.join(self._node._cbdir, p)) for p in options.get('pythonpath', [])]
else:
pythonpaths_to_add = []
# assemble command line for forking the worker
#
# all native workers (routers and containers for now) start
# from the same script in crossbar/worker/process.py -- we're
# invoking via "-m" so that .pyc files, __pycache__ etc work
# properly.
#
args = [exe, "-u", "-m", "crossbar.worker.process"]
args.extend(["--cbdir", self._node._cbdir])
args.extend(["--node", str(self._node._node_id)])
args.extend(["--worker", str(worker_id)])
args.extend(["--realm", self._realm])
args.extend(["--klass", worker_class])
args.extend(["--loglevel", get_global_log_level()])
if "shutdown" in options:
args.extend(["--shutdown", options["shutdown"]])
# Node-level callback to inject worker arguments
#
self._node._extend_worker_args(args, options)
# allow override worker process title from options
#
if options.get('title', None):
args.extend(['--title', options['title']])
# forward explicit reactor selection
#
if 'reactor' in options and sys.platform in options['reactor']:
args.extend(['--reactor', options['reactor'][sys.platform]])
# FIXME
# elif self._node.options.reactor:
# args.extend(['--reactor', self._node.options.reactor])
# create worker process environment
#
worker_env = create_process_env(options)
# We need to use the same PYTHONPATH we were started with, so we can
# find the Crossbar we're working with -- it may not be the same as the
# one on the default path
worker_env["PYTHONPATH"] = os.pathsep.join(pythonpaths_to_add + sys.path)
# log name of worker
#
worker_logname = self._node._native_workers[worker_type]['logname']
# each worker is run under its own dedicated WAMP auth role
#
worker_auth_role = u'crossbar.worker.{}'.format(worker_id)
# topic URIs used (later)
#
starting_topic = self._node._native_workers[worker_type]['topics']['starting']
started_topic = self._node._native_workers[worker_type]['topics']['started']
# add worker tracking instance to the worker map ..
#
WORKER = self._node._native_workers[worker_type]['class']
worker = WORKER(self, worker_id, details.caller, keeplog=options.get('traceback', None))
self._workers[worker_id] = worker
# create a (custom) process endpoint.
#
if platform.isWindows():
childFDs = None # Use the default Twisted ones
else:
# The communication between controller and container workers is
# using WAMP running over 2 pipes.
# For controller->container traffic this runs over FD 0 (`stdin`)
# and for the container->controller traffic, this runs over FD 3.
#
# Note: We use FD 3, not FD 1 (`stdout`) or FD 2 (`stderr`) for
# container->controller traffic, so that components running in the
# container which happen to write to `stdout` or `stderr` do not
# interfere with the container-controller communication.
childFDs = {0: "w", 1: "r", 2: "r", 3: "r"}
ep = WorkerProcessEndpoint(
self._node._reactor, exe, args, env=worker_env, worker=worker,
childFDs=childFDs)
# ready handling
#
def on_ready_success(worker_id):
self.log.info('{worker_type} worker "{worker_id}" process {pid} started',
worker_type=worker_logname, worker_id=worker.id, pid=worker.pid)
self._node._reactor.addSystemEventTrigger(
'before', 'shutdown',
self._cleanup_worker, self._node._reactor, worker,
)
worker.on_worker_started()
started_info = {
u'id': worker.id,
u'status': worker.status,
u'started': utcstr(worker.started),
u'who': worker.who,
}
# FIXME: make start of stats printer dependent on log level ..
if False:
worker.log_stats(5.)
self.publish(started_topic, started_info, options=PublishOptions(exclude=details.caller))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = 'Failed to start native worker: {}'.format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg, worker.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(_):
self.log.info("Node worker {worker.id} ended successfully", worker=worker)
# clear worker log
worker.log_stats(0)
# remove the dedicated node router authrole we dynamically
# added for the worker
self._node._drop_worker_role(worker_auth_role)
# remove our metadata tracking for the worker
del self._workers[worker.id]
# indicate that the worker excited successfully
return True
def on_exit_error(err):
self.log.info("Node worker {worker.id} ended with error ({err})", worker=worker, err=err)
# clear worker log
worker.log_stats(0)
# remove the dedicated node router authrole we dynamically
# added for the worker
self._node._drop_worker_role(worker_auth_role)
# remove our metadata tracking for the worker
del self._workers[worker.id]
# indicate that the worker excited with error
return False
def check_for_shutdown(was_successful):
self.log.info('Checking for node shutdown: worker_exit_success={worker_exit_success}, shutdown_requested={shutdown_requested}, node_shutdown_triggers={node_shutdown_triggers}', worker_exit_success=was_successful, shutdown_requested=self._shutdown_requested, node_shutdown_triggers=self._node._node_shutdown_triggers)
shutdown = self._shutdown_requested
# automatically shutdown node whenever a worker ended (successfully, or with error)
#
if checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT in self._node._node_shutdown_triggers:
self.log.info("Node worker ended, and trigger '{trigger}' active", trigger=checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT)
shutdown = True
# automatically shutdown node when worker ended with error
#
if not was_successful and checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT_WITH_ERROR in self._node._node_shutdown_triggers:
self.log.info("Node worker ended with error, and trigger '{trigger}' active", trigger=checkconfig.NODE_SHUTDOWN_ON_WORKER_EXIT_WITH_ERROR)
shutdown = True
# automatically shutdown node when no more workers are left
#
if len(self._workers) == 0 and checkconfig.NODE_SHUTDOWN_ON_LAST_WORKER_EXIT in self._node._node_shutdown_triggers:
self.log.info("No more node workers running, and trigger '{trigger}' active", trigger=checkconfig.NODE_SHUTDOWN_ON_LAST_WORKER_EXIT)
shutdown = True
# initiate shutdown (but only if we are not already shutting down)
#
if shutdown:
self.shutdown()
else:
self.log.info('Node will continue to run!')
d_on_exit = worker.exit.addCallbacks(on_exit_success, on_exit_error)
d_on_exit.addBoth(check_for_shutdown)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_native_worker_client_factory(self._node._router_session_factory, worker_auth_role, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[worker_id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {
u'id': worker_id,
u'status': worker.status,
u'created': utcstr(worker.created),
u'who': worker.who,
}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic, starting_info, options=PublishOptions(exclude=details.caller))
# now actually fork the worker ..
#
self.log.info('{worker_logname} worker "{worker_id}" starting ..',
worker_logname=worker_logname, worker_id=worker_id)
self.log.debug('{worker_logname} "{worker_id}" command line is "{cmdline}"',
worker_logname=worker_logname, worker_id=worker_id, cmdline=' '.join(args))
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will errback ..
self.log.debug('Native worker "{worker_id}" connected',
worker_id=worker_id)
worker.on_worker_connected(proto)
# dynamically add a dedicated authrole to the router
# for the worker we've just started
self._node._add_worker_role(worker_auth_role, options)
def on_connect_error(err):
# not sure when this errback is triggered at all ..
self.log.error("Interal error: connection to forked native worker failed ({err})", err=err)
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready
def start_guest(self, id, config, details=None):
"""
Start a new guest process on this node.
:param config: The guest process configuration.
:type config: obj
:returns: int -- The PID of the new process.
"""
# prohibit starting a worker twice
#
if id in self._workers:
emsg = "Could not start worker: a worker with ID '{}' is already running (or starting)".format(id)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.worker_already_running", emsg)
try:
checkconfig.check_guest(config)
except Exception as e:
raise ApplicationError(
u"crossbar.error.invalid_configuration", "invalid guest worker configuration: {}".format(e)
)
options = config.get("options", {})
# guest process working directory
#
workdir = self._node._cbdir
if "workdir" in options:
workdir = os.path.join(workdir, options["workdir"])
workdir = os.path.abspath(workdir)
# guest process executable and command line arguments
#
# first try to configure the fully qualified path for the guest
# executable by joining workdir and configured exectuable ..
exe = os.path.abspath(os.path.join(workdir, config["executable"]))
if check_executable(exe):
self.log.info("Using guest worker executable '{exe}' (executable path taken from configuration)", exe=exe)
else:
# try to detect the fully qualified path for the guest
# executable by doing a "which" on the configured executable name
exe = shutil.which(config["executable"])
if exe is not None and check_executable(exe):
self.log.info(
"Using guest worker executable '{exe}' (executable path detected from environment)", exe=exe
)
else:
emsg = "Could not start worker: could not find and executable for '{}'".format(config["executable"])
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.invalid_configuration", emsg)
# guest process command line arguments
#
args = [exe]
args.extend(config.get("arguments", []))
# guest process environment
#
worker_env = create_process_env(options)
# log name of worker
#
worker_logname = "Guest"
# topic URIs used (later)
#
starting_topic = "crossbar.node.{}.on_guest_starting".format(self._node_id)
started_topic = "crossbar.node.{}.on_guest_started".format(self._node_id)
# add worker tracking instance to the worker map ..
#
worker = GuestWorkerProcess(self, id, details.caller, keeplog=options.get("traceback", None))
self._workers[id] = worker
# create a (custom) process endpoint
#
ep = WorkerProcessEndpoint(self._node._reactor, exe, args, path=workdir, env=worker_env, worker=worker)
# ready handling
#
def on_ready_success(proto):
worker.pid = proto.transport.pid
worker.status = "started"
worker.started = datetime.utcnow()
self.log.info(
"{worker} with ID '{id}' and PID {pid} started", worker=worker_logname, id=worker.id, pid=worker.pid
)
self._node._reactor.addSystemEventTrigger(
"before", "shutdown", self._cleanup_worker, self._node._reactor, worker
)
# directory watcher
#
if "watch" in options:
if HAS_FSNOTIFY:
# assemble list of watched directories
watched_dirs = []
for d in options["watch"].get("directories", []):
watched_dirs.append(os.path.abspath(os.path.join(self._node._cbdir, d)))
worker.watch_timeout = options["watch"].get("timeout", 1)
# create a directory watcher
worker.watcher = DirWatcher(dirs=watched_dirs, notify_once=True)
# make sure to stop the background thread running inside the
# watcher upon Twisted being shut down
def on_shutdown():
worker.watcher.stop()
self._node._reactor.addSystemEventTrigger("before", "shutdown", on_shutdown)
# this handler will get fired by the watcher upon detecting an FS event
def on_fsevent(evt):
worker.watcher.stop()
proto.signal("TERM")
if options["watch"].get("action", None) == "restart":
self.log.info("Restarting guest ..")
# Add a timeout large enough (perhaps add a config option later)
self._node._reactor.callLater(worker.watch_timeout, self.start_guest, id, config, details)
# Shut the worker down, after the restart event is scheduled
worker.stop()
# now run the watcher on a background thread
deferToThread(worker.watcher.loop, on_fsevent)
else:
self.log.warn("Warning: cannot watch directory for changes - feature DirWatcher unavailable")
# assemble guest worker startup information
#
started_info = {
u"id": worker.id,
u"status": worker.status,
u"started": utcstr(worker.started),
u"who": worker.who,
}
self.publish(started_topic, started_info, options=PublishOptions(exclude=details.caller))
return started_info
def on_ready_error(err):
del self._workers[worker.id]
emsg = "Failed to start guest worker: {}".format(err.value)
self.log.error(emsg)
raise ApplicationError(u"crossbar.error.cannot_start", emsg, ep.getlog())
worker.ready.addCallbacks(on_ready_success, on_ready_error)
def on_exit_success(res):
self.log.info("Guest {id} exited with success", id=worker.id)
del self._workers[worker.id]
def on_exit_error(err):
self.log.error("Guest {id} exited with error {err.value}", id=worker.id, err=err)
del self._workers[worker.id]
worker.exit.addCallbacks(on_exit_success, on_exit_error)
# create a transport factory for talking WAMP to the native worker
#
transport_factory = create_guest_worker_client_factory(config, worker.ready, worker.exit)
transport_factory.noisy = False
self._workers[id].factory = transport_factory
# now (immediately before actually forking) signal the starting of the worker
#
starting_info = {u"id": id, u"status": worker.status, u"created": utcstr(worker.created), u"who": worker.who}
# the caller gets a progressive result ..
if details.progress:
details.progress(starting_info)
# .. while all others get an event
self.publish(starting_topic, starting_info, options=PublishOptions(exclude=details.caller))
# now actually fork the worker ..
#
self.log.info("Starting {worker} with ID '{id}'...", worker=worker_logname, id=id)
self.log.debug(
"{worker} '{id}' using command line '{cli}'...", worker=worker_logname, id=id, cli=" ".join(args)
)
d = ep.connect(transport_factory)
def on_connect_success(proto):
# this seems to be called immediately when the child process
# has been forked. even if it then immediately fails because
# e.g. the executable doesn't even exist. in other words,
# I'm not sure under what conditions the deferred will
# errback - probably only if the forking of a new process fails
# at OS level due to out of memory conditions or such.
pid = proto.transport.pid
self.log.debug("Guest worker process connected with PID {pid}", pid=pid)
worker.pid = pid
# proto is an instance of GuestWorkerClientProtocol
worker.proto = proto
worker.status = "connected"
worker.connected = datetime.utcnow()
def on_connect_error(err):
# not sure when this errback is triggered at all .. see above.
self.log.error("Internal error: connection to forked guest worker failed ({})".format(err))
# in any case, forward the error ..
worker.ready.errback(err)
d.addCallbacks(on_connect_success, on_connect_error)
return worker.ready