class RAtoOTDBPropagator():
def __init__(self,
radb_busname=RADB_BUSNAME,
radb_servicename=RADB_SERVICENAME,
radb_broker=None,
otdb_busname=DEFAULT_OTDB_SERVICE_BUSNAME,
otdb_servicename=DEFAULT_OTDB_SERVICENAME,
mom_busname=DEFAULT_MOMQUERY_BUSNAME,
mom_servicename=DEFAULT_MOMQUERY_SERVICENAME,
otdb_broker=None,
mom_broker=None,
broker=None):
"""
RAtoOTDBPropagator updates tasks in the OTDB after the ResourceAssigner is done with them.
:param radb_busname: busname on which the radb service listens (default: lofar.ra.command)
:param radb_servicename: servicename of the radb service (default: RADBService)
:param radb_broker: valid Qpid broker host (default: None, which means localhost)
:param otdb_busname: busname on which the OTDB service listens (default: lofar.otdb.command)
:param otdb_servicename: servicename of the OTDB service (default: OTDBService)
:param otdb_broker: valid Qpid broker host (default: None, which means localhost)
:param broker: if specified, overrules radb_broker and otdb_broker. Valid Qpid broker host (default: None, which means localhost)
"""
if broker:
radb_broker = broker
otdb_broker = broker
mom_broker = broker
self.radbrpc = RADBRPC(
busname=radb_busname,
servicename=radb_servicename,
broker=radb_broker
) ## , ForwardExceptions=True hardcoded in RPCWrapper right now
self.otdbrpc = OTDBRPC(
busname=otdb_busname,
servicename=otdb_servicename,
broker=otdb_broker
) ## , ForwardExceptions=True hardcoded in RPCWrapper right now
self.momrpc = MoMQueryRPC(busname=mom_busname,
servicename=mom_servicename,
broker=mom_broker)
self.translator = RAtoOTDBTranslator()
def __enter__(self):
"""Internal use only. (handles scope 'with')"""
self.open()
return self
def __exit__(self, exc_type, exc_val, exc_tb):
"""Internal use only. (handles scope 'with')"""
self.close()
def open(self):
"""Open rpc connections to radb service and resource estimator service"""
self.radbrpc.open()
self.otdbrpc.open()
self.momrpc.open()
def close(self):
"""Close rpc connections to radb service and resource estimator service"""
self.radbrpc.close()
self.otdbrpc.close()
self.momrpc.close()
def doTaskConflict(self, otdb_id):
logger.info('doTaskConflict: otdb_id=%s' % (otdb_id, ))
if not otdb_id:
logger.warning('doTaskConflict no valid otdb_id: otdb_id=%s' %
(otdb_id, ))
return
try:
self.otdbrpc.taskSetStatus(otdb_id, 'conflict')
except Exception as e:
logger.error(e)
def doTaskScheduled(self, ra_id, otdb_id, mom_id):
try:
logger.info('doTaskScheduled: ra_id=%s otdb_id=%s mom_id=%s' %
(ra_id, otdb_id, mom_id))
if not otdb_id:
logger.warning('doTaskScheduled no valid otdb_id: otdb_id=%s' %
(otdb_id, ))
return
ra_info = self.getRAinfo(ra_id)
logger.info('RA info for ra_id=%s otdb_id=%s: %s' %
(ra_id, otdb_id, ra_info))
# check if this is a CEP4 task, or an old CEP2 task
# at this moment the most simple check is to see if RA claimed (CEP4) storage
# TODO: do proper check on cluster/storage/etc
if not ra_info['storage']:
logger.info(
"No (CEP4) storage claimed for ra_id=%s otdb_id=%s, skipping otdb specification update."
% (ra_id, otdb_id))
return
#get mom project name
try:
project = self.momrpc.getProjectDetails(mom_id)
logger.info(project)
project_name = "_".join(
project[str(mom_id)]['project_name'].split())
except (RPCException, KeyError) as e:
logger.error(
'Could not get project name from MoM for mom_id %s: %s' %
(mom_id, str(e)))
logger.info('Using \'unknown\' as project name.')
project_name = 'unknown'
otdb_info = self.translator.CreateParset(otdb_id, ra_info,
project_name)
logger.debug("Parset info for OTDB: %s" % otdb_info)
self.setOTDBinfo(otdb_id, otdb_info, 'scheduled')
except Exception as e:
logger.error(e)
self.doTaskConflict(otdb_id)
def getRAinfo(self, ra_id):
info = {}
info["storage"] = {}
task = self.radbrpc.getTask(ra_id)
claims = self.radbrpc.getResourceClaims(task_ids=ra_id,
extended=True,
include_properties=True)
for claim in claims:
logger.debug("Processing claim: %s" % claim)
if claim['resource_type_name'] == 'storage':
info['storage'] = claim
info["starttime"] = task["starttime"]
info["endtime"] = task["endtime"]
info["status"] = task["status"]
return info
def setOTDBinfo(self, otdb_id, otdb_info, otdb_status):
try:
logger.info('Setting specticication for otdb_id %s: %s' %
(otdb_id, otdb_info))
self.otdbrpc.taskSetSpecification(otdb_id, otdb_info)
self.otdbrpc.taskPrepareForScheduling(
otdb_id, otdb_info["LOFAR.ObsSW.Observation.startTime"],
otdb_info["LOFAR.ObsSW.Observation.stopTime"])
logger.info('Setting status (%s) for otdb_id %s' %
(otdb_status, otdb_id))
self.otdbrpc.taskSetStatus(otdb_id, otdb_status)
except Exception as e:
logger.error(e)
self.doTaskConflict(otdb_id)
class OTDBtoRATaskStatusPropagator(OTDBBusListener):
def __init__(self,
otdb_notification_busname=DEFAULT_OTDB_NOTIFICATION_BUSNAME,
otdb_notification_subject=DEFAULT_OTDB_NOTIFICATION_SUBJECT,
radb_busname=DEFAULT_RADB_BUSNAME,
radb_servicename=DEFAULT_RADB_SERVICENAME,
broker=None, **kwargs):
super(OTDBtoRATaskStatusPropagator, self).__init__(busname=otdb_notification_busname,
subject=otdb_notification_subject,
broker=broker,
**kwargs)
self.radb = RARPC(busname=radb_busname, servicename=radb_servicename, broker=broker)
def start_listening(self, **kwargs):
self.radb.open()
super(OTDBtoRATaskStatusPropagator, self).start_listening(**kwargs)
def stop_listening(self, **kwargs):
self.radb.close()
super(OTDBtoRATaskStatusPropagator, self).stop_listening(**kwargs)
def _update_radb_task_status(self, otdb_id, task_status):
logger.info("updating radb-task with otdb_id %s to status %s" % (otdb_id, task_status))
result = self.radb.updateTaskStatusForOtdbId(otdb_id=otdb_id, status=task_status)
if not result or 'updated' not in result or not result['updated']:
logger.warning("could not update task with otdb_id %s to status %s" % (otdb_id, task_status))
def onObservationPrepared(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'prepared')
def onObservationApproved(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'approved')
def onObservationOnHold(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'on_hold')
def onObservationConflict(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'conflict')
def onObservationPrescheduled(self, treeId, modificationTime):
logger.info("not propagating prescheduled status for otdb_id %s to radb because the resource assigner takes care of this" % (treeId))
def onObservationScheduled(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'scheduled')
def onObservationQueued(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'queued')
def onObservationStarted(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'active')
def onObservationCompleting(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'completing')
def onObservationFinished(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'finished')
def onObservationAborted(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'aborted')
class OTDBtoRATaskStatusPropagator(OTDBBusListener):
def __init__(self,
otdb_notification_busname=DEFAULT_OTDB_NOTIFICATION_BUSNAME,
otdb_notification_subject=DEFAULT_OTDB_NOTIFICATION_SUBJECT,
radb_busname=DEFAULT_RADB_BUSNAME,
radb_servicename=DEFAULT_RADB_SERVICENAME,
broker=None,
**kwargs):
super(OTDBtoRATaskStatusPropagator,
self).__init__(busname=otdb_notification_busname,
subject=otdb_notification_subject,
broker=broker,
**kwargs)
self.radb = RARPC(busname=radb_busname,
servicename=radb_servicename,
broker=broker)
def start_listening(self, **kwargs):
self.radb.open()
super(OTDBtoRATaskStatusPropagator, self).start_listening(**kwargs)
def stop_listening(self, **kwargs):
self.radb.close()
super(OTDBtoRATaskStatusPropagator, self).stop_listening(**kwargs)
def _update_radb_task_status(self, otdb_id, task_status):
logger.info("updating radb-task with otdb_id %s to status %s" %
(otdb_id, task_status))
result = self.radb.updateTaskStatusForOtdbId(otdb_id=otdb_id,
status=task_status)
if not result or 'updated' not in result or not result['updated']:
logger.warning(
"could not update task with otdb_id %s to status %s" %
(otdb_id, task_status))
def onObservationPrepared(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'prepared')
def onObservationApproved(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'approved')
def onObservationOnHold(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'on_hold')
def onObservationConflict(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'conflict')
def onObservationPrescheduled(self, treeId, modificationTime):
logger.info(
"not propagating prescheduled status for otdb_id %s to radb because the resource assigner takes care of this"
% (treeId))
def onObservationScheduled(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'scheduled')
def onObservationQueued(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'queued')
def onObservationStarted(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'active')
def onObservationCompleting(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'completing')
def onObservationFinished(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'finished')
def onObservationAborted(self, treeId, modificationTime):
self._update_radb_task_status(treeId, 'aborted')
class ResourceAssigner():
def __init__(self,
radb_busname=RADB_BUSNAME,
radb_servicename=RADB_SERVICENAME,
re_busname=RE_BUSNAME,
re_servicename=RE_SERVICENAME,
ssdb_busname=DEFAULT_SSDB_BUSNAME,
ssdb_servicename=DEFAULT_SSDB_SERVICENAME,
otdb_busname=DEFAULT_OTDB_SERVICE_BUSNAME,
otdb_servicename=DEFAULT_OTDB_SERVICENAME,
broker=None):
"""
ResourceAssigner inserts/updates tasks in the radb and assigns resources to it based on incoming parset.
:param radb_busname: busname on which the radb service listens (default: lofar.ra.command)
:param radb_servicename: servicename of the radb service (default: RADBService)
:param re_busname: busname on which the resource estimator service listens (default: lofar.ra.command)
:param re_servicename: servicename of the resource estimator service (default: ResourceEstimation)
:param ssdb_busname: busname on which the ssdb service listens (default: lofar.system)
:param ssdb_servicename: servicename of the radb service (default: SSDBService)
:param broker: Valid Qpid broker host (default: None, which means localhost)
"""
self.radbrpc = RARPC(servicename=radb_servicename, busname=radb_busname, broker=broker)
self.rerpc = RPC(re_servicename, busname=re_busname, broker=broker, ForwardExceptions=True)
self.ssdbrpc = SSDBRPC(servicename=ssdb_servicename, busname=ssdb_busname, broker=broker)
self.otdbrpc = OTDBRPC(busname=otdb_busname, servicename=otdb_servicename, broker=broker) ## , ForwardExceptions=True hardcoded in RPCWrapper right now
def __enter__(self):
"""Internal use only. (handles scope 'with')"""
self.open()
return self
def __exit__(self, exc_type, exc_val, exc_tb):
"""Internal use only. (handles scope 'with')"""
self.close()
def open(self):
"""Open rpc connections to radb service and resource estimator service"""
self.radbrpc.open()
self.rerpc.open()
self.otdbrpc.open()
self.ssdbrpc.open()
def close(self):
"""Close rpc connections to radb service and resource estimator service"""
self.radbrpc.close()
self.rerpc.close()
self.otdbrpc.close()
self.ssdbrpc.close()
def doAssignment(self, specification_tree):
logger.info('doAssignment: specification_tree=%s' % (specification_tree))
otdb_id = specification_tree['otdb_id']
taskType = specification_tree.get('task_type', '').lower()
status = specification_tree.get('state', '').lower()
if status not in ['approved', 'prescheduled']: # cep2 accepts both, cep4 only prescheduled, see below
logger.info('skipping specification for otdb_id=%s because status=%s', (otdb_id, status))
#parse main parset...
mainParset = parameterset(specification_tree['specification'])
momId = mainParset.getInt('Observation.momID', -1)
try:
startTime = datetime.strptime(mainParset.getString('Observation.startTime'), '%Y-%m-%d %H:%M:%S')
endTime = datetime.strptime(mainParset.getString('Observation.stopTime'), '%Y-%m-%d %H:%M:%S')
except ValueError:
logger.warning('cannot parse for start/end time from specification for otdb_id=%s', (otdb_id, ))
# insert new task and specification in the radb
# any existing specification and task with same otdb_id will be deleted automatically
logger.info('doAssignment: insertSpecification momId=%s, otdb_id=%s, status=%s, taskType=%s, startTime=%s, endTime=%s' %
(momId, otdb_id, status, taskType, startTime, endTime))
result = self.radbrpc.insertSpecificationAndTask(momId, otdb_id, status, taskType, startTime, endTime, str(mainParset))
if not result['inserted']:
logger.error('could not insert specification and task')
return
specificationId = result['specification_id']
taskId = result['task_id']
logger.info('doAssignment: inserted specification (id=%s) and task (id=%s)' % (specificationId,taskId))
# do not assign resources to task for other clusters than cep4
if not self.checkClusterIsCEP4(mainParset):
return
if status != 'prescheduled':
logger.info('skipping resource assignment for CEP4 task otdb_id=%s because status=%s' % (otdb_id, status))
return
needed = self.getNeededResouces(specification_tree)
logger.info('doAssignment: getNeededResouces=%s' % (needed,))
if not str(otdb_id) in needed:
logger.error("no otdb_id %s found in estimator results %s" % (otdb_id, needed))
return
if not taskType in needed[str(otdb_id)]:
logger.error("no task type %s found in estimator results %s" % (taskType, needed[str(otdb_id)]))
return
# make sure the availability in the radb is up to date
# TODO: this should be updated regularly
try:
self.updateAvailableResources('cep4')
except Exception as e:
logger.warning("Exception while updating available resources: %s" % str(e))
# claim the resources for this task
# during the claim inserts the claims are automatically validated
# and if not enough resources are available, then they are put to conflict status
# also, if any claim is in conflict state, then the task is put to conflict status as well
main_needed = needed[str(otdb_id)]
task = self.radbrpc.getTask(taskId)
claimed, claim_ids = self.claimResources(main_needed, task)
if claimed:
conflictingClaims = self.radbrpc.getResourceClaims(task_ids=taskId, status='conflict')
if conflictingClaims:
logger.warning('doAssignment: %s conflicting claims detected. Task cannot be scheduled. %s' %
(len(conflictingClaims), conflictingClaims))
else:
logger.info('doAssignment: all claims for task %s were succesfully claimed. Setting task status to scheduled' % (taskId,))
self.radbrpc.updateTaskAndResourceClaims(taskId, task_status='scheduled', claim_status='allocated')
self.processPredecessors(specification_tree)
def processPredecessors(self, specification_tree):
try:
predecessor_trees = specification_tree['predecessors']
if predecessor_trees:
otdb_id = specification_tree['otdb_id']
task = self.radbrpc.getTask(otdb_id=otdb_id)
for predecessor_tree in predecessor_trees:
pred_otdb_id = predecessor_tree['otdb_id']
predecessor_task = self.radbrpc.getTask(otdb_id=pred_otdb_id)
if predecessor_task:
self.radbrpc.insertTaskPredecessor(task['id'], predecessor_task['id'])
self.processPredecessors(predecessor_tree)
except Exception as e:
logger.error(e)
def checkClusterIsCEP4(self, parset):
# check storageClusterName for enabled DataProducts
# if any storageClusterName is not CEP4, we do not accept this parset
keys = ['Output_Correlated',
'Output_IncoherentStokes',
'Output_CoherentStokes',
'Output_InstrumentModel',
'Output_SkyImage',
'Output_Pulsar']
for key in keys:
if parset.getBool('Observation.DataProducts.%s.enabled' % key, False):
if parset.getString('Observation.DataProducts.%s.storageClusterName' % key, '') != 'CEP4':
logger.warn("storageClusterName not CEP4, rejecting specification.")
return False
logger.info("all enabled storageClusterName's are CEP4, accepting specification.")
return True
def getNeededResouces(self, specification_tree):
replymessage, status = self.rerpc({"specification_tree":specification_tree}, timeout=10)
logger.info('getNeededResouces: %s' % replymessage)
return replymessage
def updateAvailableResources(self, cluster):
# find out which resources are available
# and what is their capacity
# For now, only look at CEP4 storage
# Later, also look at stations up/down for short term scheduling
#get all active groupnames, find id for cluster group
groupnames = self.ssdbrpc.getactivegroupnames()
cluster_group_id = next(k for k,v in groupnames.items() if v == cluster)
# for CEP4 cluster, do hard codes lookup of first and only node
node_info = self.ssdbrpc.gethostsforgid(cluster_group_id)['nodes'][0]
storage_resources = self.radbrpc.getResources(resource_types='storage', include_availability=True)
cep4_storage_resource = next(x for x in storage_resources if 'cep4' in x['name'])
active = node_info['statename'] == 'Active'
total_capacity = node_info['totalspace']
available_capacity = total_capacity - node_info['usedspace']
logger.info("Updating resource availability of %s (id=%s) to active=%s available_capacity=%s total_capacity=%s" %
(cep4_storage_resource['name'], cep4_storage_resource['id'], active, available_capacity, total_capacity))
self.radbrpc.updateResourceAvailability(cep4_storage_resource['id'],
active=active,
available_capacity=available_capacity,
total_capacity=total_capacity)
def claimResources(self, needed_resources, task):
logger.info('claimResources: task %s needed_resources=%s' % (task, needed_resources))
# get the needed resources for the task type
needed_resources_for_task_type = needed_resources[task['type']]
# get db lists
rc_property_types = {rcpt['name']:rcpt['id'] for rcpt in self.radbrpc.getResourceClaimPropertyTypes()}
resource_types = {rt['name']:rt['id'] for rt in self.radbrpc.getResourceTypes()}
resources = self.radbrpc.getResources()
# loop over needed_resources -> resource_type -> claim (and props)
# flatten the tree dict to a list of claims (with props)
claims = []
for resource_type_name, needed_claim_for_resource_type in needed_resources_for_task_type.items():
if resource_type_name in resource_types:
logger.info('claimResources: processing resource_type: %s' % resource_type_name)
db_resource_type_id = resource_types[resource_type_name]
db_resources_for_type = [r for r in resources if r['type_id'] == db_resource_type_id]
# needed_claim_for_resource_type is a dict containing exactly one kvp of which the value is an int
# that value is the value for the claim
needed_claim_value = next((v for k,v in needed_claim_for_resource_type.items() if isinstance(v, int)))
# FIXME: right now we just pick the first resource from the 'cep4' resources.
# estimator will deliver this info in the future
db_cep4_resources_for_type = [r for r in db_resources_for_type if 'cep4' in r['name'].lower()]
if db_cep4_resources_for_type:
claim = {'resource_id':db_cep4_resources_for_type[0]['id'],
'starttime':task['starttime'],
'endtime':task['endtime'],
'status':'claimed',
'claim_size':needed_claim_value}
#FIXME: find proper way to extend storage time with a month
if 'storage' in db_cep4_resources_for_type[0]['name']:
claim['endtime'] += timedelta(days=31)
# if the needed_claim_for_resource_type dict contains more kvp's,
# then the subdict contains groups of properties for the claim
if len(needed_claim_for_resource_type) > 1:
claim['properties'] = []
needed_prop_groups = next((v for k,v in needed_claim_for_resource_type.items() if isinstance(v, collections.Iterable)))
def processProperties(propertiesDict, sap_nr=None):
for prop_type_name, prop_value in propertiesDict.items():
if prop_type_name in rc_property_types:
rc_property_type_id = rc_property_types[prop_type_name]
property = {'type':rc_property_type_id, 'value':prop_value}
if sap_nr is not None:
property['sap_nr'] = sap_nr
claim['properties'].append(property)
else:
logger.error('claimResources: unknown prop_type:%s' % prop_type_name)
for group_name, needed_prop_group in needed_prop_groups.items():
if group_name == 'saps':
for sap_dict in needed_prop_group:
processProperties(sap_dict['properties'], sap_dict['sap_nr'])
else:
processProperties(needed_prop_group)
logger.info('claimResources: created claim:%s' % claim)
claims.append(claim)
else:
logger.error('claimResources: unknown resource_type:%s' % resource_type_name)
logger.info('claimResources: inserting %d claims in the radb' % len(claims))
claim_ids = self.radbrpc.insertResourceClaims(task['id'], claims, 1, 'anonymous', -1)['ids']
logger.info('claimResources: %d claims were inserted in the radb' % len(claim_ids))
return len(claim_ids) == len(claims), claim_ids
class RAtoOTDBPropagator():
def __init__(self,
radb_busname=RADB_BUSNAME,
radb_servicename=RADB_SERVICENAME,
radb_broker=None,
otdb_busname=DEFAULT_OTDB_SERVICE_BUSNAME,
otdb_servicename=DEFAULT_OTDB_SERVICENAME,
mom_busname=DEFAULT_MOMQUERY_BUSNAME,
mom_servicename=DEFAULT_MOMQUERY_SERVICENAME,
otdb_broker=None,
mom_broker=None,
broker=None):
"""
RAtoOTDBPropagator updates tasks in the OTDB after the ResourceAssigner is done with them.
:param radb_busname: busname on which the radb service listens (default: lofar.ra.command)
:param radb_servicename: servicename of the radb service (default: RADBService)
:param radb_broker: valid Qpid broker host (default: None, which means localhost)
:param otdb_busname: busname on which the OTDB service listens (default: lofar.otdb.command)
:param otdb_servicename: servicename of the OTDB service (default: OTDBService)
:param otdb_broker: valid Qpid broker host (default: None, which means localhost)
:param broker: if specified, overrules radb_broker and otdb_broker. Valid Qpid broker host (default: None, which means localhost)
"""
if broker:
radb_broker = broker
otdb_broker = broker
mom_broker = broker
self.radbrpc = RADBRPC(busname=radb_busname, servicename=radb_servicename, broker=radb_broker) ## , ForwardExceptions=True hardcoded in RPCWrapper right now
self.otdbrpc = OTDBRPC(busname=otdb_busname, servicename=otdb_servicename, broker=otdb_broker) ## , ForwardExceptions=True hardcoded in RPCWrapper right now
self.momrpc = MoMQueryRPC(busname=mom_busname, servicename=mom_servicename, broker=mom_broker)
self.translator = RAtoOTDBTranslator()
def __enter__(self):
"""Internal use only. (handles scope 'with')"""
self.open()
return self
def __exit__(self, exc_type, exc_val, exc_tb):
"""Internal use only. (handles scope 'with')"""
self.close()
def open(self):
"""Open rpc connections to radb service and resource estimator service"""
self.radbrpc.open()
self.otdbrpc.open()
self.momrpc.open()
def close(self):
"""Close rpc connections to radb service and resource estimator service"""
self.radbrpc.close()
self.otdbrpc.close()
self.momrpc.close()
def doTaskConflict(self, otdb_id):
logger.info('doTaskConflict: otdb_id=%s' % (otdb_id,))
if not otdb_id:
logger.warning('doTaskConflict no valid otdb_id: otdb_id=%s' % (otdb_id,))
return
try:
self.otdbrpc.taskSetStatus(otdb_id, 'conflict')
except Exception as e:
logger.error(e)
def doTaskScheduled(self, ra_id, otdb_id, mom_id):
try:
logger.info('doTaskScheduled: ra_id=%s otdb_id=%s mom_id=%s' % (ra_id, otdb_id, mom_id))
if not otdb_id:
logger.warning('doTaskScheduled no valid otdb_id: otdb_id=%s' % (otdb_id,))
return
ra_info = self.getRAinfo(ra_id)
logger.info('RA info for ra_id=%s otdb_id=%s: %s' % (ra_id, otdb_id, ra_info))
# check if this is a CEP4 task, or an old CEP2 task
# at this moment the most simple check is to see if RA claimed (CEP4) storage
# TODO: do proper check on cluster/storage/etc
if not ra_info['storage']:
logger.info("No (CEP4) storage claimed for ra_id=%s otdb_id=%s, skipping otdb specification update." % (ra_id, otdb_id))
return
#get mom project name
try:
project = self.momrpc.getProjectDetails(mom_id)
logger.info(project)
project_name = "_".join(project[str(mom_id)]['project_name'].split())
except (RPCException, KeyError) as e:
logger.error('Could not get project name from MoM for mom_id %s: %s' % (mom_id, str(e)))
logger.info('Using \'unknown\' as project name.')
project_name = 'unknown'
otdb_info = self.translator.CreateParset(otdb_id, ra_info, project_name)
logger.debug("Parset info for OTDB: %s" %otdb_info)
self.setOTDBinfo(otdb_id, otdb_info, 'scheduled')
except Exception as e:
logger.error(e)
self.doTaskConflict(otdb_id)
def getRAinfo(self, ra_id):
info = {}
info["storage"] = {}
task = self.radbrpc.getTask(ra_id)
claims = self.radbrpc.getResourceClaims(task_ids=ra_id, extended=True, include_properties=True)
for claim in claims:
logger.debug("Processing claim: %s" % claim)
if claim['resource_type_name'] == 'storage':
info['storage'] = claim
info["starttime"] = task["starttime"]
info["endtime"] = task["endtime"]
info["status"] = task["status"]
return info
def setOTDBinfo(self, otdb_id, otdb_info, otdb_status):
try:
logger.info('Setting specticication for otdb_id %s: %s' % (otdb_id, otdb_info))
self.otdbrpc.taskSetSpecification(otdb_id, otdb_info)
self.otdbrpc.taskPrepareForScheduling(otdb_id, otdb_info["LOFAR.ObsSW.Observation.startTime"], otdb_info["LOFAR.ObsSW.Observation.stopTime"])
logger.info('Setting status (%s) for otdb_id %s' % (otdb_status, otdb_id))
self.otdbrpc.taskSetStatus(otdb_id, otdb_status)
except Exception as e:
logger.error(e)
self.doTaskConflict(otdb_id)
class ResourceAssigner():
def __init__(self,
radb_busname=RADB_BUSNAME,
radb_servicename=RADB_SERVICENAME,
re_busname=RE_BUSNAME,
re_servicename=RE_SERVICENAME,
ssdb_busname=DEFAULT_SSDB_BUSNAME,
ssdb_servicename=DEFAULT_SSDB_SERVICENAME,
otdb_busname=DEFAULT_OTDB_SERVICE_BUSNAME,
otdb_servicename=DEFAULT_OTDB_SERVICENAME,
broker=None):
"""
ResourceAssigner inserts/updates tasks in the radb and assigns resources to it based on incoming parset.
:param radb_busname: busname on which the radb service listens (default: lofar.ra.command)
:param radb_servicename: servicename of the radb service (default: RADBService)
:param re_busname: busname on which the resource estimator service listens (default: lofar.ra.command)
:param re_servicename: servicename of the resource estimator service (default: ResourceEstimation)
:param ssdb_busname: busname on which the ssdb service listens (default: lofar.system)
:param ssdb_servicename: servicename of the radb service (default: SSDBService)
:param broker: Valid Qpid broker host (default: None, which means localhost)
"""
self.radbrpc = RARPC(servicename=radb_servicename,
busname=radb_busname,
broker=broker)
self.rerpc = RPC(re_servicename,
busname=re_busname,
broker=broker,
ForwardExceptions=True)
self.ssdbrpc = SSDBRPC(servicename=ssdb_servicename,
busname=ssdb_busname,
broker=broker)
self.otdbrpc = OTDBRPC(
busname=otdb_busname, servicename=otdb_servicename, broker=broker
) ## , ForwardExceptions=True hardcoded in RPCWrapper right now
def __enter__(self):
"""Internal use only. (handles scope 'with')"""
self.open()
return self
def __exit__(self, exc_type, exc_val, exc_tb):
"""Internal use only. (handles scope 'with')"""
self.close()
def open(self):
"""Open rpc connections to radb service and resource estimator service"""
self.radbrpc.open()
self.rerpc.open()
self.otdbrpc.open()
self.ssdbrpc.open()
def close(self):
"""Close rpc connections to radb service and resource estimator service"""
self.radbrpc.close()
self.rerpc.close()
self.otdbrpc.close()
self.ssdbrpc.close()
def doAssignment(self, specification_tree):
logger.info('doAssignment: specification_tree=%s' %
(specification_tree))
otdb_id = specification_tree['otdb_id']
taskType = specification_tree.get('task_type', '').lower()
status = specification_tree.get('state', '').lower()
if status not in [
'approved', 'prescheduled'
]: # cep2 accepts both, cep4 only prescheduled, see below
logger.info(
'skipping specification for otdb_id=%s because status=%s',
(otdb_id, status))
#parse main parset...
mainParset = parameterset(specification_tree['specification'])
momId = mainParset.getInt('Observation.momID', -1)
try:
startTime = datetime.strptime(
mainParset.getString('Observation.startTime'),
'%Y-%m-%d %H:%M:%S')
endTime = datetime.strptime(
mainParset.getString('Observation.stopTime'),
'%Y-%m-%d %H:%M:%S')
except ValueError:
logger.warning(
'cannot parse for start/end time from specification for otdb_id=%s',
(otdb_id, ))
# insert new task and specification in the radb
# any existing specification and task with same otdb_id will be deleted automatically
logger.info(
'doAssignment: insertSpecification momId=%s, otdb_id=%s, status=%s, taskType=%s, startTime=%s, endTime=%s'
% (momId, otdb_id, status, taskType, startTime, endTime))
result = self.radbrpc.insertSpecificationAndTask(
momId, otdb_id, status, taskType, startTime, endTime,
str(mainParset))
if not result['inserted']:
logger.error('could not insert specification and task')
return
specificationId = result['specification_id']
taskId = result['task_id']
logger.info(
'doAssignment: inserted specification (id=%s) and task (id=%s)' %
(specificationId, taskId))
# do not assign resources to task for other clusters than cep4
if not self.checkClusterIsCEP4(mainParset):
return
if status != 'prescheduled':
logger.info(
'skipping resource assignment for CEP4 task otdb_id=%s because status=%s'
% (otdb_id, status))
return
needed = self.getNeededResouces(specification_tree)
logger.info('doAssignment: getNeededResouces=%s' % (needed, ))
if not str(otdb_id) in needed:
logger.error("no otdb_id %s found in estimator results %s" %
(otdb_id, needed))
return
if not taskType in needed[str(otdb_id)]:
logger.error("no task type %s found in estimator results %s" %
(taskType, needed[str(otdb_id)]))
return
# make sure the availability in the radb is up to date
# TODO: this should be updated regularly
try:
self.updateAvailableResources('cep4')
except Exception as e:
logger.warning("Exception while updating available resources: %s" %
str(e))
# claim the resources for this task
# during the claim inserts the claims are automatically validated
# and if not enough resources are available, then they are put to conflict status
# also, if any claim is in conflict state, then the task is put to conflict status as well
main_needed = needed[str(otdb_id)]
task = self.radbrpc.getTask(taskId)
claimed, claim_ids = self.claimResources(main_needed, task)
if claimed:
conflictingClaims = self.radbrpc.getResourceClaims(
task_ids=taskId, status='conflict')
if conflictingClaims:
logger.warning(
'doAssignment: %s conflicting claims detected. Task cannot be scheduled. %s'
% (len(conflictingClaims), conflictingClaims))
else:
logger.info(
'doAssignment: all claims for task %s were succesfully claimed. Setting task status to scheduled'
% (taskId, ))
self.radbrpc.updateTaskAndResourceClaims(
taskId, task_status='scheduled', claim_status='allocated')
self.processPredecessors(specification_tree)
def processPredecessors(self, specification_tree):
try:
predecessor_trees = specification_tree['predecessors']
if predecessor_trees:
otdb_id = specification_tree['otdb_id']
task = self.radbrpc.getTask(otdb_id=otdb_id)
for predecessor_tree in predecessor_trees:
pred_otdb_id = predecessor_tree['otdb_id']
predecessor_task = self.radbrpc.getTask(
otdb_id=pred_otdb_id)
if predecessor_task:
self.radbrpc.insertTaskPredecessor(
task['id'], predecessor_task['id'])
self.processPredecessors(predecessor_tree)
except Exception as e:
logger.error(e)
def checkClusterIsCEP4(self, parset):
# check storageClusterName for enabled DataProducts
# if any storageClusterName is not CEP4, we do not accept this parset
keys = [
'Output_Correlated', 'Output_IncoherentStokes',
'Output_CoherentStokes', 'Output_InstrumentModel',
'Output_SkyImage', 'Output_Pulsar'
]
for key in keys:
if parset.getBool('Observation.DataProducts.%s.enabled' % key,
False):
if parset.getString(
'Observation.DataProducts.%s.storageClusterName' % key,
'') != 'CEP4':
logger.warn(
"storageClusterName not CEP4, rejecting specification."
)
return False
logger.info(
"all enabled storageClusterName's are CEP4, accepting specification."
)
return True
def getNeededResouces(self, specification_tree):
replymessage, status = self.rerpc(
{"specification_tree": specification_tree}, timeout=10)
logger.info('getNeededResouces: %s' % replymessage)
return replymessage
def updateAvailableResources(self, cluster):
# find out which resources are available
# and what is their capacity
# For now, only look at CEP4 storage
# Later, also look at stations up/down for short term scheduling
#get all active groupnames, find id for cluster group
groupnames = self.ssdbrpc.getactivegroupnames()
cluster_group_id = next(k for k, v in groupnames.items()
if v == cluster)
# for CEP4 cluster, do hard codes lookup of first and only node
node_info = self.ssdbrpc.gethostsforgid(cluster_group_id)['nodes'][0]
storage_resources = self.radbrpc.getResources(
resource_types='storage', include_availability=True)
cep4_storage_resource = next(x for x in storage_resources
if 'cep4' in x['name'])
active = node_info['statename'] == 'Active'
total_capacity = node_info['totalspace']
available_capacity = total_capacity - node_info['usedspace']
logger.info(
"Updating resource availability of %s (id=%s) to active=%s available_capacity=%s total_capacity=%s"
% (cep4_storage_resource['name'], cep4_storage_resource['id'],
active, available_capacity, total_capacity))
self.radbrpc.updateResourceAvailability(
cep4_storage_resource['id'],
active=active,
available_capacity=available_capacity,
total_capacity=total_capacity)
def claimResources(self, needed_resources, task):
logger.info('claimResources: task %s needed_resources=%s' %
(task, needed_resources))
# get the needed resources for the task type
needed_resources_for_task_type = needed_resources[task['type']]
# get db lists
rc_property_types = {
rcpt['name']: rcpt['id']
for rcpt in self.radbrpc.getResourceClaimPropertyTypes()
}
resource_types = {
rt['name']: rt['id']
for rt in self.radbrpc.getResourceTypes()
}
resources = self.radbrpc.getResources()
# loop over needed_resources -> resource_type -> claim (and props)
# flatten the tree dict to a list of claims (with props)
claims = []
for resource_type_name, needed_claim_for_resource_type in needed_resources_for_task_type.items(
):
if resource_type_name in resource_types:
logger.info('claimResources: processing resource_type: %s' %
resource_type_name)
db_resource_type_id = resource_types[resource_type_name]
db_resources_for_type = [
r for r in resources if r['type_id'] == db_resource_type_id
]
# needed_claim_for_resource_type is a dict containing exactly one kvp of which the value is an int
# that value is the value for the claim
needed_claim_value = next(
(v for k, v in needed_claim_for_resource_type.items()
if isinstance(v, int)))
# FIXME: right now we just pick the first resource from the 'cep4' resources.
# estimator will deliver this info in the future
db_cep4_resources_for_type = [
r for r in db_resources_for_type
if 'cep4' in r['name'].lower()
]
if db_cep4_resources_for_type:
claim = {
'resource_id': db_cep4_resources_for_type[0]['id'],
'starttime': task['starttime'],
'endtime': task['endtime'],
'status': 'claimed',
'claim_size': needed_claim_value
}
#FIXME: find proper way to extend storage time with a month
if 'storage' in db_cep4_resources_for_type[0]['name']:
claim['endtime'] += timedelta(days=31)
# if the needed_claim_for_resource_type dict contains more kvp's,
# then the subdict contains groups of properties for the claim
if len(needed_claim_for_resource_type) > 1:
claim['properties'] = []
needed_prop_groups = next((
v
for k, v in needed_claim_for_resource_type.items()
if isinstance(v, collections.Iterable)))
def processProperties(propertiesDict, sap_nr=None):
for prop_type_name, prop_value in propertiesDict.items(
):
if prop_type_name in rc_property_types:
rc_property_type_id = rc_property_types[
prop_type_name]
property = {
'type': rc_property_type_id,
'value': prop_value
}
if sap_nr is not None:
property['sap_nr'] = sap_nr
claim['properties'].append(property)
else:
logger.error(
'claimResources: unknown prop_type:%s'
% prop_type_name)
for group_name, needed_prop_group in needed_prop_groups.items(
):
if group_name == 'saps':
for sap_dict in needed_prop_group:
processProperties(sap_dict['properties'],
sap_dict['sap_nr'])
else:
processProperties(needed_prop_group)
logger.info('claimResources: created claim:%s' % claim)
claims.append(claim)
else:
logger.error('claimResources: unknown resource_type:%s' %
resource_type_name)
logger.info('claimResources: inserting %d claims in the radb' %
len(claims))
claim_ids = self.radbrpc.insertResourceClaims(task['id'], claims, 1,
'anonymous', -1)['ids']
logger.info('claimResources: %d claims were inserted in the radb' %
len(claim_ids))
return len(claim_ids) == len(claims), claim_ids