def will_analyze_alert(self, alert):
'''Whether or not to analyze an alert
'''
# Will not analyze duplicate alert
alertId = alert.get_rec_id()
if alert.dup_alert_recid is not None:
if alert.dup_alert_recid > 0:
registry.get_logger().debug('Duplicate alert rec id %d is not analyzed', alert.get_rec_id())
return False
else:
# If the dup_alert_recid field is None, then it has not been set, so determine if
# there are duplicates as before
if self.alertMgr.is_duplicate(alertId):
registry.get_logger().debug('Duplicate alert id %d is not analyzed.', alert.get_rec_id())
return False
# Will not analyze alert without hardware location (C: Compute; I: I/O)
loc_type = alert.event_loc.get_id()
if loc_type != 'C' and loc_type != 'I':
registry.get_logger().debug('Alert id %d with location type %s is not analyzed.', alert.get_rec_id(), loc_type)
return False
# Will analyze all other alerts
return True
def analyze_alert(self, alert):
'''Analyze an alert
'''
alert_recId = alert.get_rec_id()
alert_id = alert.get_incident_id()
loc_type = alert.event_loc.get_id()
location = alert.event_loc.get_location()
#alert_msgId = alert.get_incident_id()
registry.get_logger().info('Analyzing alert id %d loc_type: %s: %s', alert_recId, loc_type, location)
# There should only be one condition event associated with the alert.
events = alert.condition_events
if len(events) == 0:
registry.get_logger().error('No event associated with the alert recid %d', alert_recId)
registry.get_service(SERVICE_ALERT_DELIVERY_Q).put(alert)
return
event = events.pop()
if (alert_id == 'BQL01'):
# No need to analyze BQL01 alerts, just pass it to the delivery queue
registry.get_logger().info('Nothing to analyze for alert id %s ', alert_id)
registry.get_service(SERVICE_ALERT_DELIVERY_Q).put(alert)
return
# Get the location
loc = Location(loc_type, location)
locName = self.get_loc_name(loc)
# No need to analyze alert with rack location
alert_time = str(alert.get_time_occurred())
if locName == 'rack':
registry.get_logger().info('Nothing to analyze for alert recid %d with rack location', alert_recId)
registry.get_service(SERVICE_ALERT_DELIVERY_Q).put(alert)
return
# Find out if there are other alerts with the same block id (for ENDJOB01 and THRES01)
dup_qry = ''
if (alert_id == 'ENDJOB01' or alert_id == 'THRES01'):
if event.raw_data['block'] is None:
event_block = None
else:
event_block = event.raw_data['block'].strip()
if event_block is None or event_block == BGQ_EVENT_NULL_BLOCK:
# Found no prior alert with the same block id, pass current alert to the delivery queue
registry.get_logger().info('No block id for alert id %d, no common alert generated for block: %s', alert_recId, event_block)
registry.get_service(SERVICE_ALERT_DELIVERY_Q).put(alert)
return
same_block = False
# Get db connection needed for query
dbi = registry.get_service(SERVICE_DB_INTERFACE)
dbConn = dbi.get_connection()
cursor = dbConn.cursor()
if (alert_id == 'ENDJOB01'):
# For ENDJOB01, look for alert id HWERR01 or COMMON01 with the same block id
same_block = self.has_matching_blockId(event_block, alert_time, cursor)
else:
# For THRES0101, look for alert id HWERR01 or COMMON01 or ENDJOB01 with the same block id
same_block = self.has_matching_blockId(event_block, alert_time, cursor)
if same_block:
# Found prior alert with the same block id, close current alert
registry.get_logger().info('Closing current alert recid %d due to prior alert with the same block id', alert_recId)
registry.get_service(SERVICE_ALERT_MGR).close(alert_recId)
else:
# Found no prior alert with the same block id, pass current alert to the delivery queue
registry.get_logger().info('No common block id found for alert id %d within the last %s', alert_recId, self.window_time)
registry.get_service(SERVICE_ALERT_DELIVERY_Q).put(alert)
return
elif (alert_id == 'BQL01'):
# No need to analyze BQL01 alerts, just pass it to the delivery queue
registry.get_logger().info('Nothing to analyze for alert id %s.', alert_id)
registry.get_service(SERVICE_ALERT_DELIVERY_Q).put(alert)
return
# The following will handle the rest of the alert ids (HWERR01 or COMMON01).
# Find out if there is common mode alert already exist for the same location or higher hierarchy
loc_parent, loc_parent_list = self.get_loc_parent(loc)
loc_qry = '('
idx = 0
for pLoc in loc_parent_list:
if idx != 0:
loc_qry += " or "
loc_qry += " \"event_loc\" like '" + pLoc + "'"
idx += 1
dup_qry2 = self.dup_query + loc_qry + ")"
loc_qry += " or \"event_loc\" like '" + location + "')"
dup_qry = self.dup_query + loc_qry
dbi = registry.get_service(SERVICE_DB_INTERFACE)
dbConn = dbi.get_connection()
cursor = dbConn.cursor()
dup = self.has_duplicate(alert_time, dup_qry, cursor)
if dup:
# Found prior alert with the same block id, close current alert
registry.get_logger().info('Closing current alert recid %d due to prior alert with same common location', alert_recId)
registry.get_service(SERVICE_ALERT_MGR).close(alert_recId)
return
# Look for a common hardware problem if there are multiple alerts for different location
# on the same hardware.
sendAlert = self.has_common_location(loc, alert_time, self.query, cursor)
if sendAlert:
# Send commmon alert
self.send_common_alert(loc, alert_recId, event, alert_time, dup_qry2, cursor)
else:
# Pass current alert to the delivery queue
registry.get_logger().info('No common location for %s found for alert id: %d within the last %s ', location, alert_recId, self.window_time)
registry.get_service(SERVICE_ALERT_DELIVERY_Q).put(alert)
return