def _on_periodic_tasks_stop(self, fut):
try:
fut.result()
except Exception as exc:
LOG.critical(_LC('Periodic tasks worker has failed: %s'), exc)
else:
LOG.info(_LI('Successfully shut down periodic tasks'))
def _on_periodic_tasks_stop(self, fut):
try:
fut.result()
except Exception as exc:
LOG.critical(_LC('Periodic tasks worker has failed: %s'), exc)
else:
LOG.info(_LI('Successfully shut down periodic tasks'))
def init_host(self, admin_context=None):
"""Initialize the conductor host.
:param admin_context: the admin context to pass to periodic tasks.
:raises: RuntimeError when conductor is already running.
:raises: NoDriversLoaded when no drivers are enabled on the conductor.
:raises: DriverNotFound if a driver is enabled that does not exist.
:raises: DriverLoadError if an enabled driver cannot be loaded.
"""
if self._started:
raise RuntimeError(_('Attempt to start an already running '
'conductor manager'))
self.dbapi = dbapi.get_instance()
self._keepalive_evt = threading.Event()
"""Event for the keepalive thread."""
# TODO(dtantsur): make the threshold configurable?
rejection_func = rejection.reject_when_reached(
CONF.conductor.workers_pool_size)
self._executor = futurist.GreenThreadPoolExecutor(
max_workers=CONF.conductor.workers_pool_size,
check_and_reject=rejection_func)
"""Executor for performing tasks async."""
self.ring_manager = hash.HashRingManager()
"""Consistent hash ring which maps drivers to conductors."""
# NOTE(deva): this call may raise DriverLoadError or DriverNotFound
drivers = driver_factory.drivers()
if not drivers:
msg = _LE("Conductor %s cannot be started because no drivers "
"were loaded. This could be because no drivers were "
"specified in 'enabled_drivers' config option.")
LOG.error(msg, self.host)
raise exception.NoDriversLoaded(conductor=self.host)
# NOTE(jroll) this is passed to the dbapi, which requires a list, not
# a generator (which keys() returns in py3)
driver_names = list(drivers)
# Collect driver-specific periodic tasks.
# Conductor periodic tasks accept context argument, driver periodic
# tasks accept this manager and context. We have to ensure that the
# same driver interface class is not traversed twice, otherwise
# we'll have several instances of the same task.
LOG.debug('Collecting periodic tasks')
self._periodic_task_callables = []
periodic_task_classes = set()
self._collect_periodic_tasks(self, (admin_context,))
for driver_obj in drivers.values():
self._collect_periodic_tasks(driver_obj, (self, admin_context))
for iface_name in driver_obj.all_interfaces:
iface = getattr(driver_obj, iface_name, None)
if iface and iface.__class__ not in periodic_task_classes:
self._collect_periodic_tasks(iface, (self, admin_context))
periodic_task_classes.add(iface.__class__)
if (len(self._periodic_task_callables) >
CONF.conductor.workers_pool_size):
LOG.warning(_LW('This conductor has %(tasks)d periodic tasks '
'enabled, but only %(workers)d task workers '
'allowed by [conductor]workers_pool_size option'),
{'tasks': len(self._periodic_task_callables),
'workers': CONF.conductor.workers_pool_size})
self._periodic_tasks = periodics.PeriodicWorker(
self._periodic_task_callables,
executor_factory=periodics.ExistingExecutor(self._executor))
# clear all locks held by this conductor before registering
self.dbapi.clear_node_reservations_for_conductor(self.host)
try:
# Register this conductor with the cluster
cdr = self.dbapi.register_conductor({'hostname': self.host,
'drivers': driver_names})
except exception.ConductorAlreadyRegistered:
# This conductor was already registered and did not shut down
# properly, so log a warning and update the record.
LOG.warning(
_LW("A conductor with hostname %(hostname)s "
"was previously registered. Updating registration"),
{'hostname': self.host})
cdr = self.dbapi.register_conductor({'hostname': self.host,
'drivers': driver_names},
update_existing=True)
self.conductor = cdr
# Start periodic tasks
self._periodic_tasks_worker = self._executor.submit(
self._periodic_tasks.start, allow_empty=True)
self._periodic_tasks_worker.add_done_callback(
self._on_periodic_tasks_stop)
# NOTE(lucasagomes): If the conductor server dies abruptly
# mid deployment (OMM Killer, power outage, etc...) we
# can not resume the deployment even if the conductor
# comes back online. Cleaning the reservation of the nodes
# (dbapi.clear_node_reservations_for_conductor) is not enough to
# unstick it, so let's gracefully fail the deployment so the node
# can go through the steps (deleting & cleaning) to make itself
# available again.
filters = {'reserved': False,
'provision_state': states.DEPLOYING}
last_error = (_("The deployment can't be resumed by conductor "
"%s. Moving to fail state.") % self.host)
self._fail_if_in_state(ironic_context.get_admin_context(), filters,
states.DEPLOYING, 'provision_updated_at',
last_error=last_error)
# Spawn a dedicated greenthread for the keepalive
try:
self._spawn_worker(self._conductor_service_record_keepalive)
LOG.info(_LI('Successfully started conductor with hostname '
'%(hostname)s.'),
{'hostname': self.host})
except exception.NoFreeConductorWorker:
with excutils.save_and_reraise_exception():
LOG.critical(_LC('Failed to start keepalive'))
self.del_host()
self._started = True
def init_host(self):
self.dbapi = dbapi.get_instance()
self._keepalive_evt = threading.Event()
"""Event for the keepalive thread."""
self._worker_pool = greenpool.GreenPool(
size=CONF.conductor.workers_pool_size)
"""GreenPool of background workers for performing tasks async."""
self.ring_manager = hash.HashRingManager()
"""Consistent hash ring which maps drivers to conductors."""
# NOTE(deva): instantiating DriverFactory may raise DriverLoadError
# or DriverNotFound
self._driver_factory = driver_factory.DriverFactory()
"""Driver factory loads all enabled drivers."""
self.drivers = self._driver_factory.names
"""List of driver names which this conductor supports."""
if not self.drivers:
msg = _LE("Conductor %s cannot be started because no drivers "
"were loaded. This could be because no drivers were "
"specified in 'enabled_drivers' config option.")
LOG.error(msg, self.host)
raise exception.NoDriversLoaded(conductor=self.host)
# Collect driver-specific periodic tasks
for driver_obj in driver_factory.drivers().values():
self._collect_periodic_tasks(driver_obj)
for iface_name in (driver_obj.core_interfaces +
driver_obj.standard_interfaces + ['vendor']):
iface = getattr(driver_obj, iface_name, None)
if iface:
self._collect_periodic_tasks(iface)
# clear all locks held by this conductor before registering
self.dbapi.clear_node_reservations_for_conductor(self.host)
try:
# Register this conductor with the cluster
cdr = self.dbapi.register_conductor({
'hostname': self.host,
'drivers': self.drivers
})
except exception.ConductorAlreadyRegistered:
# This conductor was already registered and did not shut down
# properly, so log a warning and update the record.
LOG.warning(
_LW("A conductor with hostname %(hostname)s "
"was previously registered. Updating registration"),
{'hostname': self.host})
cdr = self.dbapi.register_conductor(
{
'hostname': self.host,
'drivers': self.drivers
},
update_existing=True)
self.conductor = cdr
# NOTE(lucasagomes): If the conductor server dies abruptly
# mid deployment (OMM Killer, power outage, etc...) we
# can not resume the deployment even if the conductor
# comes back online. Cleaning the reservation of the nodes
# (dbapi.clear_node_reservations_for_conductor) is not enough to
# unstick it, so let's gracefully fail the deployment so the node
# can go through the steps (deleting & cleaning) to make itself
# available again.
filters = {'reserved': False, 'provision_state': states.DEPLOYING}
last_error = (_("The deployment can't be resumed by conductor "
"%s. Moving to fail state.") % self.host)
self._fail_if_in_state(ironic_context.get_admin_context(),
filters,
states.DEPLOYING,
'provision_updated_at',
last_error=last_error)
# Spawn a dedicated greenthread for the keepalive
try:
self._spawn_worker(self._conductor_service_record_keepalive)
LOG.info(
_LI('Successfully started conductor with hostname '
'%(hostname)s.'), {'hostname': self.host})
except exception.NoFreeConductorWorker:
with excutils.save_and_reraise_exception():
LOG.critical(_LC('Failed to start keepalive'))
self.del_host()
def init_host(self, admin_context=None):
"""Initialize the conductor host.
:param admin_context: the admin context to pass to periodic tasks.
:raises: RuntimeError when conductor is already running.
:raises: NoDriversLoaded when no drivers are enabled on the conductor.
:raises: DriverNotFound if a driver is enabled that does not exist.
:raises: DriverLoadError if an enabled driver cannot be loaded.
:raises: DriverNameConflict if a classic driver and a dynamic driver
are both enabled and have the same name.
"""
if self._started:
raise RuntimeError(
_('Attempt to start an already running '
'conductor manager'))
self.dbapi = dbapi.get_instance()
self._keepalive_evt = threading.Event()
"""Event for the keepalive thread."""
# TODO(dtantsur): make the threshold configurable?
rejection_func = rejection.reject_when_reached(
CONF.conductor.workers_pool_size)
self._executor = futurist.GreenThreadPoolExecutor(
max_workers=CONF.conductor.workers_pool_size,
check_and_reject=rejection_func)
"""Executor for performing tasks async."""
self.ring_manager = hash_ring.HashRingManager()
"""Consistent hash ring which maps drivers to conductors."""
_check_enabled_interfaces()
# NOTE(deva): these calls may raise DriverLoadError or DriverNotFound
# NOTE(vdrok): Instantiate network and storage interface factory on
# startup so that all the interfaces are loaded at the very
# beginning, and failures prevent the conductor from starting.
drivers = driver_factory.drivers()
hardware_types = driver_factory.hardware_types()
driver_factory.NetworkInterfaceFactory()
driver_factory.StorageInterfaceFactory()
# NOTE(jroll) this is passed to the dbapi, which requires a list, not
# a generator (which keys() returns in py3)
driver_names = list(drivers)
hardware_type_names = list(hardware_types)
# check that at least one driver is loaded, whether classic or dynamic
if not driver_names and not hardware_type_names:
msg = _LE("Conductor %s cannot be started because no drivers "
"were loaded. This could be because no classic drivers "
"were specified in the 'enabled_drivers' config option "
"and no dynamic drivers were specified in the "
"'enabled_hardware_types' config option.")
LOG.error(msg, self.host)
raise exception.NoDriversLoaded(conductor=self.host)
# check for name clashes between classic and dynamic drivers
name_clashes = set(driver_names).intersection(hardware_type_names)
if name_clashes:
name_clashes = ', '.join(name_clashes)
msg = _LE("Conductor %(host)s cannot be started because there is "
"one or more name conflicts between classic drivers and "
"dynamic drivers (%(names)s). Check any external driver "
"plugins and the 'enabled_drivers' and "
"'enabled_hardware_types' config options.")
LOG.error(msg, {'host': self.host, 'names': name_clashes})
raise exception.DriverNameConflict(names=name_clashes)
# Collect driver-specific periodic tasks.
# Conductor periodic tasks accept context argument, driver periodic
# tasks accept this manager and context. We have to ensure that the
# same driver interface class is not traversed twice, otherwise
# we'll have several instances of the same task.
LOG.debug('Collecting periodic tasks')
self._periodic_task_callables = []
periodic_task_classes = set()
self._collect_periodic_tasks(self, (admin_context, ))
for driver_obj in drivers.values():
for iface_name in driver_obj.all_interfaces:
iface = getattr(driver_obj, iface_name, None)
if iface and iface.__class__ not in periodic_task_classes:
self._collect_periodic_tasks(iface, (self, admin_context))
periodic_task_classes.add(iface.__class__)
if (len(self._periodic_task_callables) >
CONF.conductor.workers_pool_size):
LOG.warning(
_LW('This conductor has %(tasks)d periodic tasks '
'enabled, but only %(workers)d task workers '
'allowed by [conductor]workers_pool_size option'), {
'tasks': len(self._periodic_task_callables),
'workers': CONF.conductor.workers_pool_size
})
self._periodic_tasks = periodics.PeriodicWorker(
self._periodic_task_callables,
executor_factory=periodics.ExistingExecutor(self._executor))
# clear all target_power_state with locks by this conductor
self.dbapi.clear_node_target_power_state(self.host)
# clear all locks held by this conductor before registering
self.dbapi.clear_node_reservations_for_conductor(self.host)
try:
# Register this conductor with the cluster
self.conductor = objects.Conductor.register(
admin_context, self.host, driver_names)
except exception.ConductorAlreadyRegistered:
# This conductor was already registered and did not shut down
# properly, so log a warning and update the record.
LOG.warning(
_LW("A conductor with hostname %(hostname)s "
"was previously registered. Updating registration"),
{'hostname': self.host})
self.conductor = objects.Conductor.register(admin_context,
self.host,
driver_names,
update_existing=True)
# register hardware types and interfaces supported by this conductor
# and validate them against other conductors
try:
self._register_and_validate_hardware_interfaces(hardware_types)
except (exception.DriverLoadError, exception.DriverNotFound,
exception.ConductorHardwareInterfacesAlreadyRegistered,
exception.InterfaceNotFoundInEntrypoint,
exception.NoValidDefaultForInterface) as e:
with excutils.save_and_reraise_exception():
LOG.error(_LE('Failed to register hardware types. %s'), e)
self.del_host()
# Start periodic tasks
self._periodic_tasks_worker = self._executor.submit(
self._periodic_tasks.start, allow_empty=True)
self._periodic_tasks_worker.add_done_callback(
self._on_periodic_tasks_stop)
# NOTE(lucasagomes): If the conductor server dies abruptly
# mid deployment (OMM Killer, power outage, etc...) we
# can not resume the deployment even if the conductor
# comes back online. Cleaning the reservation of the nodes
# (dbapi.clear_node_reservations_for_conductor) is not enough to
# unstick it, so let's gracefully fail the deployment so the node
# can go through the steps (deleting & cleaning) to make itself
# available again.
filters = {'reserved': False, 'provision_state': states.DEPLOYING}
last_error = (_("The deployment can't be resumed by conductor "
"%s. Moving to fail state.") % self.host)
self._fail_if_in_state(ironic_context.get_admin_context(),
filters,
states.DEPLOYING,
'provision_updated_at',
last_error=last_error)
# Start consoles if it set enabled in a greenthread.
try:
self._spawn_worker(self._start_consoles,
ironic_context.get_admin_context())
except exception.NoFreeConductorWorker:
LOG.warning(_LW('Failed to start worker for restarting consoles.'))
# Spawn a dedicated greenthread for the keepalive
try:
self._spawn_worker(self._conductor_service_record_keepalive)
LOG.info(
_LI('Successfully started conductor with hostname '
'%(hostname)s.'), {'hostname': self.host})
except exception.NoFreeConductorWorker:
with excutils.save_and_reraise_exception():
LOG.critical(_LC('Failed to start keepalive'))
self.del_host()
self._started = True
def init_host(self):
self.dbapi = dbapi.get_instance()
self._keepalive_evt = threading.Event()
"""Event for the keepalive thread."""
self._worker_pool = greenpool.GreenPool(
size=CONF.conductor.workers_pool_size)
"""GreenPool of background workers for performing tasks async."""
self.ring_manager = hash.HashRingManager()
"""Consistent hash ring which maps drivers to conductors."""
# NOTE(deva): instantiating DriverFactory may raise DriverLoadError
# or DriverNotFound
self._driver_factory = driver_factory.DriverFactory()
"""Driver factory loads all enabled drivers."""
self.drivers = self._driver_factory.names
"""List of driver names which this conductor supports."""
if not self.drivers:
msg = _LE("Conductor %s cannot be started because no drivers "
"were loaded. This could be because no drivers were "
"specified in 'enabled_drivers' config option.")
LOG.error(msg, self.host)
raise exception.NoDriversLoaded(conductor=self.host)
# Collect driver-specific periodic tasks
for driver_obj in driver_factory.drivers().values():
self._collect_periodic_tasks(driver_obj)
for iface_name in (driver_obj.core_interfaces +
driver_obj.standard_interfaces +
['vendor']):
iface = getattr(driver_obj, iface_name, None)
if iface:
self._collect_periodic_tasks(iface)
# clear all locks held by this conductor before registering
self.dbapi.clear_node_reservations_for_conductor(self.host)
try:
# Register this conductor with the cluster
cdr = self.dbapi.register_conductor({'hostname': self.host,
'drivers': self.drivers})
except exception.ConductorAlreadyRegistered:
# This conductor was already registered and did not shut down
# properly, so log a warning and update the record.
LOG.warning(
_LW("A conductor with hostname %(hostname)s "
"was previously registered. Updating registration"),
{'hostname': self.host})
cdr = self.dbapi.register_conductor({'hostname': self.host,
'drivers': self.drivers},
update_existing=True)
self.conductor = cdr
# NOTE(lucasagomes): If the conductor server dies abruptly
# mid deployment (OMM Killer, power outage, etc...) we
# can not resume the deployment even if the conductor
# comes back online. Cleaning the reservation of the nodes
# (dbapi.clear_node_reservations_for_conductor) is not enough to
# unstick it, so let's gracefully fail the deployment so the node
# can go through the steps (deleting & cleaning) to make itself
# available again.
filters = {'reserved': False,
'provision_state': states.DEPLOYING}
last_error = (_("The deployment can't be resumed by conductor "
"%s. Moving to fail state.") % self.host)
self._fail_if_in_state(ironic_context.get_admin_context(), filters,
states.DEPLOYING, 'provision_updated_at',
last_error=last_error)
# Spawn a dedicated greenthread for the keepalive
try:
self._spawn_worker(self._conductor_service_record_keepalive)
LOG.info(_LI('Successfully started conductor with hostname '
'%(hostname)s.'),
{'hostname': self.host})
except exception.NoFreeConductorWorker:
with excutils.save_and_reraise_exception():
LOG.critical(_LC('Failed to start keepalive'))
self.del_host()
