def single_task_duration(experiment_id ,service_name, single_task_duration): statsd.gauge(JQUEUER_SINGLE_TASK_DURATION, single_task_duration, tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, ] )
def experiment_actual_start_timestamp(experiment_id ,service_name, experiment_actual_start_timestamp): statsd.gauge(JQUEUER_EXPERIMENT_ACTUAL_START_TIMESTAMP, experiment_actual_start_timestamp, tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, ] )
def service_replicas_running(experiment_id ,service_name, service_replicas_running): statsd.gauge(JQUEUER_SERVICE_REPLICAS_RUNNING, service_replicas_running, tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, ] )
def service_replicas_max(experiment_id ,service_name, service_replicas_max): statsd.gauge(JQUEUER_SERVICE_REPLICAS_MAX, service_replicas_max, tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, ] )
def service_replicas_needed(experiment_id ,service_name, service_replicas_needed): statsd.gauge(JQUEUER_SERVICE_REPLICAS_NEEDED, service_replicas_needed, tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, ] )
def experiment_running_timestamp(experiment_id ,service_name, experiment_running_timestamp): statsd.gauge(JQUEUER_EXPERIMENT_RUNNING_TIMESTAMP, experiment_running_timestamp, tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, ] )
def experiment_deadline_timestamp(experiment_id ,service_name, experiment_deadline_timestamp): statsd.gauge(JQUEUER_EXPERIMENT_DEADLINE_TIMESTAMP, experiment_deadline_timestamp, tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, ] )
def add_job(experiment_id ,service_name, job_id): statsd.gauge(JQUEUER_JOB_ADDED_TIMESTAMP, time.time(), tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'job_id: %s' % job_id, ] ) statsd.gauge(JQUEUER_JOB_ADDED, time.time(), tags=[ 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'job_id: %s' % job_id, ] )
def task_failed(node_id, experiment_id ,service_name, qworker_id, job_id, task_id, fail_time): elapsed_time = time.time() - fail_time statsd.gauge(JQUEUER_TASK_FAILED_TIMESTAMP, time.time(), tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, # 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.histogram(JQUEUER_TASK_RUNNING_TIMESTAMP, time.time(), tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, # 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, ] ) statsd.gauge(JQUEUER_TASK_FAILED_DURATION, elapsed_time, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.gauge(JQUEUER_TASK_FAILED, 1, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.gauge(JQUEUER_TASK_RUNNING, 0, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] )
def terminate_task(node_id, experiment_id ,service_name, qworker_id, job_id, task_id, start_time): elapsed_time = time.time() - start_time statsd.gauge(JQUEUER_TASK_ACCOMPLISHED_TIMESTAMP, time.time(), tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, # 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.histogram(JQUEUER_TASK_RUNNING_TIMESTAMP, time.time(), tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, # 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, ] ) statsd.gauge(JQUEUER_TASK_ACCOMPLISHED_DURATION, elapsed_time, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, # 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.gauge(JQUEUER_TASK_ACCOMPLISHED, 1, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.gauge(JQUEUER_TASK_RUNNING, 0, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] )
def run_task(node_id, experiment_id ,service_name, qworker_id, job_id, task_id): statsd.gauge(JQUEUER_TASK_STARTED_TIMESTAMP, time.time(), tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, # 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.histogram(JQUEUER_TASK_RUNNING_TIMESTAMP, time.time(), tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, # 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.gauge(JQUEUER_TASK_RUNNING, 1, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] ) statsd.gauge(JQUEUER_TASK_STARTED, 1, tags=[ 'node_id:%s' % node_id, 'experiment_id:%s' % experiment_id, 'service_name:%s' % service_name, 'qworker_id: %s' % qworker_id, 'job_id: %s' % job_id, 'task_id: %s' % task_id, ] )