def process( self, message ) :
""" Batch job (i.e. user script) threw an exception, try it again
TODO: need to check against some "max contiguous errors",
i.e. if this fails 3x in a row then terminate the request
:param message: SQS Message instance
"""
try:
spot_request_msg = SpotRequestMsg( raw_json=message.get_body() )
spot_request_uuid = spot_request_msg.spot_request_uuid
spot_master_uuid = spot_request_msg.spot_master_uuid
logger.info( fmt_request_uuid_msg_hdr( spot_request_uuid ) + 'process() for spot_master_uuid: ' + spot_master_uuid )
spot_request_item = get_spot_request_item( self.spot_request_table_name, spot_request_msg.spot_request_uuid, region_name=self.region_name, profile_name=self.profile_name )
ts_cmd_complete = spot_request_msg.name_value_pairs[ SpotRequestMsg.PAIR_NAME_BATCH_PROCESS_COMPLETE_TIMESTAMP]
cmd_exception_message = spot_request_msg.name_value_pairs[ SpotRequestMsg.PAIR_NAME_INSTANCE_BATCH_PROCESS_START_EXCEPTION_MESSAGE]
cmd_exception_traceback = spot_request_msg.name_value_pairs[ SpotRequestMsg.PAIR_NAME_INSTANCE_BATCH_PROCESS_START_EXCEPTION_TRACEBACK]
key_value_pairs = {
TableSpotRequest.is_open:0,
TableSpotRequest.spot_request_state_code:SpotRequestStateCode.instance_complete_exception,
TableSpotRequest.ts_cmd_complete:ts_cmd_complete,
TableSpotRequest.cmd_exception_message:cmd_exception_message,
TableSpotRequest.cmd_exception_traceback:cmd_exception_traceback,
}
spot_request_row_partial_save( self.spot_request_table_name, spot_request_item, key_value_pairs, region_name=self.region_name, profile_name=self.profile_name )
self.spot_request_sqs_message_durable.delete_message(message)
except StandardError as e:
logger.error( fmt_request_uuid_msg_hdr( spot_request_uuid ) + 'Exiting SpotRequestDispatcher due to exception' )
logger.error( fmt_request_uuid_msg_hdr( spot_request_uuid ) + str(e) )
logger.error( fmt_request_uuid_msg_hdr( spot_request_uuid ) + traceback.format_exc() )
def process( self, message ) :
""" This should never happen, so don't terminate the instance, leave it up so it can get SSH'ed into
to determine the cause of failure
:param message: SQS Message instance
"""
try:
spot_request_msg = SpotRequestMsg( raw_json=message.get_body() )
instance_termination_exception = spot_request_msg.name_value_pairs[ SpotRequestMsg.PAIR_NAME_INSTANCE_TERMINATION_TIME_EXCEPTION ]
spot_request_item = get_spot_request_item( self.spot_request_table_name, spot_request_msg.spot_request_uuid, region_name=self.region_name, profile_name=self.profile_name )
logger.error( fmt_request_uuid_msg_hdr( spot_request_msg.spot_request_uuid ) + 'process_pending_termination_exception for spot_request_uuid, instance_termination_exception=' + instance_termination_exception )
ts_now = int( time.time() )
spot_request_row_partial_save( self.spot_request_table_name, spot_request_item,
{
TableSpotRequest.is_open:0,
TableSpotRequest.ts_end:ts_now,
TableSpotRequest.spot_request_state_code:SpotRequestStateCode.instance_force_termination_exception,
TableSpotRequest.instance_termination_exception:instance_termination_exception
},
region_name=self.region_name, profile_name=self.profile_name )
self.spot_request_sqs_message_durable.delete_message(message)
except StandardError as e:
logger.error( fmt_request_uuid_msg_hdr( spot_request_msg.spot_request_uuid ) + 'Exiting SpotRequestDispatcher due to exception' )
logger.error( fmt_request_uuid_msg_hdr( spot_request_msg.spot_request_uuid ) + str(e) )
logger.error( fmt_request_uuid_msg_hdr( spot_request_msg.spot_request_uuid ) + traceback.format_exc() )
def process( self, message ) :
""" Process the message
:param message: SQS Message instance
"""
try:
spot_request_msg = SpotRequestMsg( raw_json=message.get_body() )
spot_request_uuid = spot_request_msg.spot_request_uuid
spot_master_uuid = spot_request_msg.spot_master_uuid
spot_request_id = spot_request_msg.spot_request_id
logger.info( fmt_request_uuid_msg_hdr( spot_request_uuid ) + 'process_check_status' )
# Get spot request row from DynamoDB and process based on state
spot_request_item = get_spot_request_item( self.spot_request_table_name, spot_request_uuid, region_name=self.region_name, profile_name=self.profile_name )
logger.info( fmt_request_uuid_msg_hdr( spot_request_uuid ) + 'spot request state=' + spot_request_item[TableSpotRequest.spot_request_state_code])
next_status_msg_delay_secs = 60
is_send_request_msg_check_status = True
spot_request_state_code = spot_request_item[TableSpotRequest.spot_request_state_code]
# Update the LastStateCheck timestamp
spot_request_row_partial_save( self.spot_request_table_name, spot_request_item, {
TableSpotRequest.ts_last_state_check:int( time.time() ),
},
region_name=self.region_name, profile_name=self.profile_name )
if SpotRequestStateCode.spot_request_in_progress == spot_request_state_code:
self.handle_state_request_spot_request_in_progress( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
elif SpotRequestStateCode.instance_starting == spot_request_state_code:
self.handle_state_request_instance_starting( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
elif SpotRequestStateCode.instance_running == spot_request_state_code:
self.handle_state_request_instance_running( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
elif SpotRequestStateCode.instance_complete == spot_request_state_code:
self.handle_state_request_instance_complete( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
is_send_request_msg_check_status = False
elif SpotRequestStateCode.instance_state_unknown == spot_request_state_code:
self.handle_state_request_instance_state_unknown( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
elif SpotRequestStateCode.constraint_encountered == spot_request_state_code:
self.handle_state_request_constraint_encountered( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
elif SpotRequestStateCode.instance_force_termination_pending == spot_request_state_code:
self.handle_state_instance_force_termination_pending( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
elif SpotRequestStateCode.instance_force_terminated == spot_request_state_code:
self.handle_state_request_instance_force_terminated( spot_request_msg, spot_request_item, spot_request_uuid, spot_master_uuid )
is_send_request_msg_check_status = False
if is_send_request_msg_check_status:
spot_request_msg_check_status = SpotRequestMsg( spot_request_uuid=spot_request_uuid,
spot_master_uuid=spot_master_uuid,
spot_request_msg_type=SpotRequestMsg.TYPE_CHECK_STATUS,
spot_request_id=spot_request_id )
message_attributes = create_microsvc_message_attributes( awsspotbatch.common.const.MICROSVC_REQUEST_CLASSNAME_SpotRequestMessageCheckStatus )
self.spot_request_sqs_message_durable.send_message( spot_request_msg_check_status.to_json(),
delay_seconds=next_status_msg_delay_secs,
message_attributes=message_attributes )
self.spot_request_sqs_message_durable.delete_message(message)
except StandardError as e:
logger.error( fmt_request_uuid_msg_hdr( spot_request_uuid ) + 'Exiting SpotRequestDispatcher due to exception' )
logger.error( fmt_request_uuid_msg_hdr( spot_request_uuid ) + str(e) )
logger.error( fmt_request_uuid_msg_hdr( spot_request_uuid ) + traceback.format_exc() )
def process(self, message):
"""
Spot Request has completed, write completion info to SpotRequestItem in DynamoDB,
let master know this request has completed so the master can determine if the job has completed
:param message: SQS Message instance
"""
try:
spot_request_msg = SpotRequestMsg(raw_json=message.get_body())
spot_request_item = get_spot_request_item(
self.spot_request_table_name,
spot_request_msg.spot_request_uuid,
region_name=self.region_name,
profile_name=self.profile_name,
)
ts_cmd_complete = spot_request_msg.name_value_pairs[
SpotRequestMsg.PAIR_NAME_BATCH_PROCESS_COMPLETE_TIMESTAMP
]
cmd_returncode = spot_request_msg.name_value_pairs[SpotRequestMsg.PAIR_NAME_BATCH_PROCESS_RETURNCODE]
cmd_std_out = spot_request_msg.name_value_pairs[SpotRequestMsg.PAIR_NAME_BATCH_PROCESS_STD_OUT]
cmd_std_err = spot_request_msg.name_value_pairs[SpotRequestMsg.PAIR_NAME_BATCH_PROCESS_STD_ERR]
key_value_pairs = {
TableSpotRequest.is_open: 0,
TableSpotRequest.spot_request_state_code: SpotRequestStateCode.instance_complete,
TableSpotRequest.ts_cmd_complete: ts_cmd_complete,
TableSpotRequest.cmd_returncode: cmd_returncode,
}
if cmd_std_out != None and len(cmd_std_out) > 0:
key_value_pairs[TableSpotRequest.cmd_std_out] = cmd_std_out
if cmd_std_err != None and len(cmd_std_err) > 0:
key_value_pairs[TableSpotRequest.cmd_std_err] = cmd_std_err
spot_request_row_partial_save(
self.spot_request_table_name,
spot_request_item,
key_value_pairs,
region_name=self.region_name,
profile_name=self.profile_name,
)
# let the Master increment the completion count to determine if the job is complete
master_msg_incr_instance_success = SpotMasterMsg(
spot_master_uuid=spot_request_msg.spot_master_uuid,
spot_master_msg_type=SpotMasterMsg.TYPE_INCR_INSTANCE_SUCCESS_CNT,
)
message_attributes = create_microsvc_message_attributes(
awsspotbatch.common.const.MICROSVC_MASTER_CLASSNAME_SpotMasterMessageIncrSuccessCnt
)
spot_master_sqs_message_durable = SqsMessageDurable(
self.spot_master_queue_name, self.region_name, profile_name=self.profile_name
)
spot_master_sqs_message_durable.send_message(
master_msg_incr_instance_success.to_json(), message_attributes=message_attributes
)
self.spot_request_sqs_message_durable.delete_message(message)
except StandardError as e:
logger.error(fmt_request_item_msg_hdr(spot_request_item) + "Exiting SpotRequestDispatcher due to exception")
logger.error(fmt_request_item_msg_hdr(spot_request_item) + str(e))
logger.error(fmt_request_item_msg_hdr(spot_request_item) + traceback.format_exc())
def process( self, message ) :
""" AWS is going to terminate the request - update the status in SpotRequestItem. A SpotRequestCheckStatus
message is processed at a regular interval and will detect the status change and process accordingly
:param message: SQS Message instance
"""
try:
spot_request_msg = SpotRequestMsg( raw_json=message.get_body() )
spot_request_item = get_spot_request_item( self.spot_request_table_name, spot_request_msg.spot_request_uuid, region_name=self.region_name, profile_name=self.profile_name )
logger.info( fmt_request_item_msg_hdr( spot_request_item ) + 'process_pending_termination_detected' )
ts_pending_termination_detected = spot_request_msg.name_value_pairs[ SpotRequestMsg.PAIR_NAME_INSTANCE_TS_PENDING_TERMINATION_DETECTED]
spot_request_row_partial_save( self.spot_request_table_name, spot_request_item, {
TableSpotRequest.spot_request_state_code:SpotRequestStateCode.instance_force_termination_pending,
TableSpotRequest.ts_pending_termination_detected:ts_pending_termination_detected
},
region_name=self.region_name, profile_name=self.profile_name )
self.spot_request_sqs_message_durable.delete_message(message)
except StandardError as e:
logger.error( fmt_request_item_msg_hdr( spot_request_item ) + 'Exiting SpotRequestDispatcher due to exception' )
logger.error( fmt_request_item_msg_hdr( spot_request_item ) + str(e) )
logger.error( fmt_request_item_msg_hdr( spot_request_item ) + traceback.format_exc() )
def process(self, message):
""" Heartbeat daemon successfully started, update timestamp in SpotRequestItem
TODO: need to add check to determine the heartbeat daemon has started within a reasonable
time period after the instance has started. If the heartbeat daemon can't start, the
users' script will never run.
:param message: SQS Message instance
"""
try:
spot_request_msg = SpotRequestMsg(raw_json=message.get_body())
spot_request_item = get_spot_request_item(
self.spot_request_table_name,
spot_request_msg.spot_request_uuid,
region_name=self.region_name,
profile_name=self.profile_name,
)
logger.info(fmt_request_item_msg_hdr(spot_request_item) + "process_heartbeat_daemon_started")
ts_heartbeat_daemon_started = spot_request_msg.name_value_pairs[
SpotRequestMsg.PAIR_NAME_INSTANCE_HEARTBEAT_DAEMON_STARTED_TIMESTAMP
]
spot_request_row_partial_save(
self.spot_request_table_name,
spot_request_item,
{
TableSpotRequest.ts_heartbeat_daemon_started: ts_heartbeat_daemon_started,
TableSpotRequest.ts_heartbeat: ts_heartbeat_daemon_started,
},
region_name=self.region_name,
profile_name=self.profile_name,
)
self.spot_request_sqs_message_durable.delete_message(message)
except StandardError as e:
logger.error(fmt_request_item_msg_hdr(spot_request_item) + "Exiting SpotRequestDispatcher due to exception")
logger.error(fmt_request_item_msg_hdr(spot_request_item) + str(e))
logger.error(fmt_request_item_msg_hdr(spot_request_item) + traceback.format_exc())
def process( self, message ) :
""" Write heartbeat timestamp to SpotRequestItem
:param message: SQS Message instance
"""
try:
before = time.time()
spot_request_msg = SpotRequestMsg( raw_json=message.get_body() )
spot_request_item = get_spot_request_item( self.spot_request_table_name, spot_request_msg.spot_request_uuid, region_name=self.region_name, profile_name=self.profile_name )
logger.info( fmt_request_item_msg_hdr( spot_request_item ) + 'process_instance_heartbeat' )
if spot_request_item != None:
ts_heartbeat = spot_request_msg.name_value_pairs[ SpotRequestMsg.PAIR_NAME_INSTANCE_HEARTBEAT_TIMESTAMP]
spot_request_row_partial_save( self.spot_request_table_name, spot_request_item,
{TableSpotRequest.ts_heartbeat:ts_heartbeat },
region_name=self.region_name, profile_name=self.profile_name )
else: logger.warning('Heartbeat not saved, spot_request_uuid not found: ' + spot_request_msg.spot_request_uuid )
self.spot_request_sqs_message_durable.delete_message(message)
# logging.info('>>> Elapsed message process for heartbeat: ' + str(time.time() - before) )
except StandardError as e:
logger.error( fmt_request_item_msg_hdr( spot_request_item ) + 'Exiting SpotRequestDispatcher due to exception' )
logger.error( fmt_request_item_msg_hdr( spot_request_item ) + str(e) )
logger.error( fmt_request_item_msg_hdr( spot_request_item ) + traceback.format_exc() )
