def master_main(options):
create_pycurl_thread(ciel.engine)
deferred_worker = DeferredWorkPlugin(ciel.engine)
deferred_worker.subscribe()
worker_pool = WorkerPool(ciel.engine, deferred_worker, None)
worker_pool.subscribe()
task_failure_investigator = TaskFailureInvestigator(
worker_pool, deferred_worker)
job_pool = JobPool(ciel.engine, options.journaldir, None,
task_failure_investigator, deferred_worker, worker_pool,
options.task_log_root)
job_pool.subscribe()
worker_pool.job_pool = job_pool
backup_sender = BackupSender(cherrypy.engine)
backup_sender.subscribe()
if options.hostname is not None:
local_hostname = options.hostname
else:
local_hostname = socket.getfqdn()
local_port = cherrypy.config.get('server.socket_port')
master_netloc = '%s:%d' % (local_hostname, local_port)
ciel.log('Local port is %d' % local_port, 'STARTUP', logging.INFO)
if options.blockstore is None:
static_content_root = tempfile.mkdtemp(
prefix=os.getenv('TEMP', default='/tmp/ciel-files-'))
ciel.log(
'Block store directory not specified: storing in %s' %
static_content_root, 'STARTUP', logging.WARN)
else:
static_content_root = options.blockstore
block_store_dir = os.path.join(static_content_root, "data")
try:
os.mkdir(block_store_dir)
except:
pass
block_store = BlockStore(local_hostname, local_port, block_store_dir)
block_store.build_pin_set()
block_store.check_local_blocks()
# TODO: Re-enable this and test thoroughly.
# if options.master is not None:
# monitor = MasterRecoveryMonitor(cherrypy.engine, 'http://%s/' % master_netloc, options.master, job_pool)
# monitor.subscribe()
# else:
monitor = None
recovery_manager = RecoveryManager(ciel.engine, job_pool, block_store,
deferred_worker)
recovery_manager.subscribe()
root = MasterRoot(worker_pool, block_store, job_pool, backup_sender,
monitor)
cherrypy.config.update({"server.thread_pool": 50})
cherrypy.config.update({"checker.on": False})
cherrypy_conf = dict()
app = cherrypy.tree.mount(root, "", cherrypy_conf)
lighty = LighttpdAdapter(ciel.engine, static_content_root, local_port)
lighty.subscribe()
# Zap CherryPy's original flavour server
cherrypy.server.unsubscribe()
server = cherrypy.process.servers.FlupFCGIServer(
application=app, bindAddress=lighty.socket_path)
adapter = cherrypy.process.servers.ServerAdapter(
cherrypy.engine, httpserver=server, bind_addr=lighty.socket_path)
# Insert a FastCGI server in its place
adapter.subscribe()
if hasattr(ciel.engine, "signal_handler"):
ciel.engine.signal_handler.subscribe()
if hasattr(ciel.engine, "console_control_handler"):
ciel.engine.console_control_handler.subscribe()
ciel.engine.start()
ciel.engine.block()
def master_main(options):
create_pycurl_thread(ciel.engine)
deferred_worker = DeferredWorkPlugin(ciel.engine)
deferred_worker.subscribe()
worker_pool = WorkerPool(ciel.engine, deferred_worker, None)
worker_pool.subscribe()
task_failure_investigator = TaskFailureInvestigator(worker_pool, deferred_worker)
job_pool = JobPool(ciel.engine, options.journaldir, None, task_failure_investigator, deferred_worker, worker_pool, options.task_log_root)
job_pool.subscribe()
worker_pool.job_pool = job_pool
mc_policy = PerformancePolicy('Performance')
mission_controller = MissionController(ciel.engine, job_pool, mc_policy)
mission_controller.subscribe()
backup_sender = BackupSender(cherrypy.engine)
backup_sender.subscribe()
if options.hostname is not None:
local_hostname = options.hostname
else:
local_hostname = socket.getfqdn()
local_port = cherrypy.config.get('server.socket_port')
master_netloc = '%s:%d' % (local_hostname, local_port)
ciel.log('Local port is %d' % local_port, 'STARTUP', logging.INFO)
if options.blockstore is None:
static_content_root = tempfile.mkdtemp(prefix=os.getenv('TEMP', default='/tmp/ciel-files-'))
ciel.log('Block store directory not specified: storing in %s' % static_content_root, 'STARTUP', logging.WARN)
else:
static_content_root = options.blockstore
block_store_dir = os.path.join(static_content_root, "data")
try:
os.mkdir(block_store_dir)
except:
pass
block_store = BlockStore(local_hostname, local_port, block_store_dir)
block_store.build_pin_set()
block_store.check_local_blocks()
# TODO: Re-enable this and test thoroughly.
#if options.master is not None:
# monitor = MasterRecoveryMonitor(cherrypy.engine, 'http://%s/' % master_netloc, options.master, job_pool)
# monitor.subscribe()
#else:
monitor = None
recovery_manager = RecoveryManager(ciel.engine, job_pool, block_store, deferred_worker)
recovery_manager.subscribe()
root = MasterRoot(worker_pool, block_store, job_pool, backup_sender, monitor)
cherrypy.config.update({"server.thread_pool" : 50})
cherrypy.config.update({"checker.on" : False})
cherrypy_conf = dict()
app = cherrypy.tree.mount(root, "", cherrypy_conf)
lighty = LighttpdAdapter(ciel.engine, static_content_root, local_port)
lighty.subscribe()
# Zap CherryPy's original flavour server
cherrypy.server.unsubscribe()
server = cherrypy.process.servers.FlupFCGIServer(application=app, bindAddress=lighty.socket_path)
adapter = cherrypy.process.servers.ServerAdapter(cherrypy.engine, httpserver=server, bind_addr=lighty.socket_path)
# Insert a FastCGI server in its place
adapter.subscribe()
if hasattr(ciel.engine, "signal_handler"):
ciel.engine.signal_handler.subscribe()
if hasattr(ciel.engine, "console_control_handler"):
ciel.engine.console_control_handler.subscribe()
ciel.engine.start()
ciel.engine.block()
def __init__(self, bus, port, options):
plugins.SimplePlugin.__init__(self, bus)
create_pycurl_thread(bus)
if options.aux_port is not None:
create_tcp_server(options.aux_port)
self.id = None
self.port = port
self.master_url = options.master
self.master_proxy = MasterProxy(self, bus, self.master_url)
self.master_proxy.subscribe()
# This will now be set by the pinger when it attempts to contact the master.
self.hostname = None
# if options.hostname is None:
# self.hostname = self.master_proxy.get_public_hostname()
# else:
# self.hostname = options.hostname
if options.blockstore is None:
self.static_content_root = tempfile.mkdtemp(prefix=os.getenv('TEMP', default='/tmp/ciel-files-'))
else:
self.static_content_root = options.blockstore
block_store_dir = os.path.join(self.static_content_root, "data")
try:
os.mkdir(block_store_dir)
except:
pass
# The hostname is None because we get this from the master.
self.block_store = BlockStore(None, self.port, block_store_dir, ignore_blocks=False)
self.block_store.build_pin_set()
self.block_store.check_local_blocks()
create_watcher_thread(bus, self.block_store)
self.upload_deferred_work = DeferredWorkPlugin(bus, 'upload_work')
self.upload_deferred_work.subscribe()
self.upload_manager = UploadManager(self.block_store, self.upload_deferred_work)
self.execution_features = ExecutionFeatures()
#self.task_executor = TaskExecutorPlugin(bus, self, self.master_proxy, self.execution_features, 1)
#self.task_executor.subscribe()
self.scheduling_classes = parse_scheduling_class_option(None, options.num_threads)
self.multiworker = MultiWorker(ciel.engine, self)
self.multiworker.subscribe()
self.process_pool = ProcessPool(bus, self, self.execution_features.process_cacheing_executors)
self.process_pool.subscribe()
self.runnable_executors = self.execution_features.runnable_executors.keys()
self.server_root = WorkerRoot(self)
self.pinger = Pinger(bus, self.master_proxy, None)
self.pinger.subscribe()
self.stopping = False
self.event_log = []
self.log_lock = Lock()
self.log_condition = Condition(self.log_lock)
self.cherrypy_conf = {}
cherrypy.config.update({"server.thread_pool" : 20})
cherrypy.config.update({"checker.on" : False})
self.subscribe()
class Worker(plugins.SimplePlugin):
def __init__(self, bus, port, options):
plugins.SimplePlugin.__init__(self, bus)
create_pycurl_thread(bus)
if options.aux_port is not None:
create_tcp_server(options.aux_port)
self.id = None
self.port = port
self.master_url = options.master
self.master_proxy = MasterProxy(self, bus, self.master_url)
self.master_proxy.subscribe()
# This will now be set by the pinger when it attempts to contact the master.
self.hostname = None
# if options.hostname is None:
# self.hostname = self.master_proxy.get_public_hostname()
# else:
# self.hostname = options.hostname
if options.blockstore is None:
self.static_content_root = tempfile.mkdtemp(prefix=os.getenv('TEMP', default='/tmp/ciel-files-'))
else:
self.static_content_root = options.blockstore
block_store_dir = os.path.join(self.static_content_root, "data")
try:
os.mkdir(block_store_dir)
except:
pass
# The hostname is None because we get this from the master.
self.block_store = BlockStore(None, self.port, block_store_dir, ignore_blocks=False)
self.block_store.build_pin_set()
self.block_store.check_local_blocks()
create_watcher_thread(bus, self.block_store)
self.upload_deferred_work = DeferredWorkPlugin(bus, 'upload_work')
self.upload_deferred_work.subscribe()
self.upload_manager = UploadManager(self.block_store, self.upload_deferred_work)
self.execution_features = ExecutionFeatures()
#self.task_executor = TaskExecutorPlugin(bus, self, self.master_proxy, self.execution_features, 1)
#self.task_executor.subscribe()
self.scheduling_classes = parse_scheduling_class_option(None, options.num_threads)
self.multiworker = MultiWorker(ciel.engine, self)
self.multiworker.subscribe()
self.process_pool = ProcessPool(bus, self, self.execution_features.process_cacheing_executors)
self.process_pool.subscribe()
self.runnable_executors = self.execution_features.runnable_executors.keys()
self.server_root = WorkerRoot(self)
self.pinger = Pinger(bus, self.master_proxy, None)
self.pinger.subscribe()
self.stopping = False
self.event_log = []
self.log_lock = Lock()
self.log_condition = Condition(self.log_lock)
self.cherrypy_conf = {}
cherrypy.config.update({"server.thread_pool" : 20})
cherrypy.config.update({"checker.on" : False})
self.subscribe()
def subscribe(self):
self.bus.subscribe('stop', self.stop, priority=10)
self.bus.subscribe("worker_event", self.add_log_entry)
def unsubscribe(self):
self.bus.unsubscribe('stop', self.stop)
self.bus.unsubscribe("worker_event", self.add_log_entry)
def netloc(self):
return '%s:%d' % (self.hostname, self.port)
def as_descriptor(self):
return {'netloc': self.netloc(), 'features': self.runnable_executors, 'has_blocks': not self.block_store.is_empty(), 'scheduling_classes': self.scheduling_classes}
def set_hostname(self, hostname):
self.hostname = hostname
self.block_store.set_hostname(hostname)
def set_master(self, master_details):
self.master_url = master_details['master']
self.master_proxy.change_master(self.master_url)
self.pinger.poke()
def start_running(self):
app = cherrypy.tree.mount(self.server_root, "", self.cherrypy_conf)
lighty = LighttpdAdapter(ciel.engine, self.static_content_root, self.port)
lighty.subscribe()
# Zap CherryPy's original flavour server
cherrypy.server.unsubscribe()
server = cherrypy.process.servers.FlupFCGIServer(application=app, bindAddress=lighty.socket_path)
adapter = cherrypy.process.servers.ServerAdapter(cherrypy.engine, httpserver=server, bind_addr=lighty.socket_path)
# Insert a FastCGI server in its place
adapter.subscribe()
ciel.engine.start()
if hasattr(ciel.engine, "signal_handler"):
ciel.engine.signal_handler.subscribe()
if hasattr(ciel.engine, "console_control_handler"):
ciel.engine.console_control_handler.subscribe()
ciel.engine.block()
def stop(self):
with self.log_lock:
self.stopping = True
self.log_condition.notify_all()
def submit_task(self, task_descriptor):
ciel.engine.publish("worker_event", "Start task " + repr(task_descriptor["task_id"]))
ciel.engine.publish('execute_task', task_descriptor)
def abort_task(self, task_id):
ciel.engine.publish("worker_event", "Abort task " + repr(task_id))
self.task_executor.abort_task(task_id)
def add_log_entry(self, log_string):
with self.log_lock:
self.event_log.append((datetime.now(), log_string))
self.log_condition.notify_all()
def get_log_entries(self, start_index, end_index):
with self.log_lock:
return self.event_log[start_index:end_index]
def await_log_entries_after(self, index):
with self.log_lock:
while len(self.event_log) <= int(index):
if self.stopping == True:
break
self.log_condition.wait()
if self.stopping:
raise Exception("Worker stopping")
def __init__(self, bus, port, options):
plugins.SimplePlugin.__init__(self, bus)
create_pycurl_thread(bus)
if options.aux_port is not None:
create_tcp_server(options.aux_port)
self.id = None
self.port = port
self.master_url = options.master
self.master_proxy = MasterProxy(self, bus, self.master_url)
self.master_proxy.subscribe()
# This will now be set by the pinger when it attempts to contact the master.
self.hostname = None
# if options.hostname is None:
# self.hostname = self.master_proxy.get_public_hostname()
# else:
# self.hostname = options.hostname
if options.blockstore is None:
self.static_content_root = tempfile.mkdtemp(
prefix=os.getenv('TEMP', default='/tmp/ciel-files-'))
else:
self.static_content_root = options.blockstore
block_store_dir = os.path.join(self.static_content_root, "data")
try:
os.mkdir(block_store_dir)
except:
pass
# The hostname is None because we get this from the master.
self.block_store = BlockStore(None,
self.port,
block_store_dir,
ignore_blocks=False)
self.block_store.build_pin_set()
self.block_store.check_local_blocks()
create_watcher_thread(bus, self.block_store)
self.upload_deferred_work = DeferredWorkPlugin(bus, 'upload_work')
self.upload_deferred_work.subscribe()
self.upload_manager = UploadManager(self.block_store,
self.upload_deferred_work)
self.execution_features = ExecutionFeatures()
#self.task_executor = TaskExecutorPlugin(bus, self, self.master_proxy, self.execution_features, 1)
#self.task_executor.subscribe()
self.scheduling_classes = parse_scheduling_class_option(
None, options.num_threads)
self.multiworker = MultiWorker(ciel.engine, self)
self.multiworker.subscribe()
self.process_pool = ProcessPool(
bus, self, self.execution_features.process_cacheing_executors)
self.process_pool.subscribe()
self.runnable_executors = self.execution_features.runnable_executors.keys(
)
self.server_root = WorkerRoot(self)
self.pinger = Pinger(bus, self.master_proxy, None)
self.pinger.subscribe()
self.stopping = False
self.event_log = []
self.log_lock = Lock()
self.log_condition = Condition(self.log_lock)
self.cherrypy_conf = {}
cherrypy.config.update({"server.thread_pool": 20})
cherrypy.config.update({"checker.on": False})
self.subscribe()
class Worker(plugins.SimplePlugin):
def __init__(self, bus, port, options):
plugins.SimplePlugin.__init__(self, bus)
create_pycurl_thread(bus)
if options.aux_port is not None:
create_tcp_server(options.aux_port)
self.id = None
self.port = port
self.master_url = options.master
self.master_proxy = MasterProxy(self, bus, self.master_url)
self.master_proxy.subscribe()
# This will now be set by the pinger when it attempts to contact the master.
self.hostname = None
# if options.hostname is None:
# self.hostname = self.master_proxy.get_public_hostname()
# else:
# self.hostname = options.hostname
if options.blockstore is None:
self.static_content_root = tempfile.mkdtemp(
prefix=os.getenv('TEMP', default='/tmp/ciel-files-'))
else:
self.static_content_root = options.blockstore
block_store_dir = os.path.join(self.static_content_root, "data")
try:
os.mkdir(block_store_dir)
except:
pass
# The hostname is None because we get this from the master.
self.block_store = BlockStore(None,
self.port,
block_store_dir,
ignore_blocks=False)
self.block_store.build_pin_set()
self.block_store.check_local_blocks()
create_watcher_thread(bus, self.block_store)
self.upload_deferred_work = DeferredWorkPlugin(bus, 'upload_work')
self.upload_deferred_work.subscribe()
self.upload_manager = UploadManager(self.block_store,
self.upload_deferred_work)
self.execution_features = ExecutionFeatures()
#self.task_executor = TaskExecutorPlugin(bus, self, self.master_proxy, self.execution_features, 1)
#self.task_executor.subscribe()
self.scheduling_classes = parse_scheduling_class_option(
None, options.num_threads)
self.multiworker = MultiWorker(ciel.engine, self)
self.multiworker.subscribe()
self.process_pool = ProcessPool(
bus, self, self.execution_features.process_cacheing_executors)
self.process_pool.subscribe()
self.runnable_executors = self.execution_features.runnable_executors.keys(
)
self.server_root = WorkerRoot(self)
self.pinger = Pinger(bus, self.master_proxy, None)
self.pinger.subscribe()
self.stopping = False
self.event_log = []
self.log_lock = Lock()
self.log_condition = Condition(self.log_lock)
self.cherrypy_conf = {}
cherrypy.config.update({"server.thread_pool": 20})
cherrypy.config.update({"checker.on": False})
self.subscribe()
def subscribe(self):
self.bus.subscribe('stop', self.stop, priority=10)
self.bus.subscribe("worker_event", self.add_log_entry)
def unsubscribe(self):
self.bus.unsubscribe('stop', self.stop)
self.bus.unsubscribe("worker_event", self.add_log_entry)
def netloc(self):
return '%s:%d' % (self.hostname, self.port)
def as_descriptor(self):
return {
'netloc': self.netloc(),
'features': self.runnable_executors,
'has_blocks': not self.block_store.is_empty(),
'scheduling_classes': self.scheduling_classes
}
def set_hostname(self, hostname):
self.hostname = hostname
self.block_store.set_hostname(hostname)
def set_master(self, master_details):
self.master_url = master_details['master']
self.master_proxy.change_master(self.master_url)
self.pinger.poke()
def start_running(self):
app = cherrypy.tree.mount(self.server_root, "", self.cherrypy_conf)
lighty = LighttpdAdapter(ciel.engine, self.static_content_root,
self.port)
lighty.subscribe()
# Zap CherryPy's original flavour server
cherrypy.server.unsubscribe()
server = cherrypy.process.servers.FlupFCGIServer(
application=app, bindAddress=lighty.socket_path)
adapter = cherrypy.process.servers.ServerAdapter(
cherrypy.engine, httpserver=server, bind_addr=lighty.socket_path)
# Insert a FastCGI server in its place
adapter.subscribe()
ciel.engine.start()
if hasattr(ciel.engine, "signal_handler"):
ciel.engine.signal_handler.subscribe()
if hasattr(ciel.engine, "console_control_handler"):
ciel.engine.console_control_handler.subscribe()
ciel.engine.block()
def stop(self):
with self.log_lock:
self.stopping = True
self.log_condition.notify_all()
def submit_task(self, task_descriptor):
ciel.engine.publish("worker_event",
"Start task " + repr(task_descriptor["task_id"]))
ciel.engine.publish('execute_task', task_descriptor)
def abort_task(self, task_id):
ciel.engine.publish("worker_event", "Abort task " + repr(task_id))
self.task_executor.abort_task(task_id)
def add_log_entry(self, log_string):
with self.log_lock:
self.event_log.append((datetime.now(), log_string))
self.log_condition.notify_all()
def get_log_entries(self, start_index, end_index):
with self.log_lock:
return self.event_log[start_index:end_index]
def await_log_entries_after(self, index):
with self.log_lock:
while len(self.event_log) <= int(index):
if self.stopping == True:
break
self.log_condition.wait()
if self.stopping:
raise Exception("Worker stopping")
