Commit 22282375 authored by Carlos Natalino's avatar Carlos Natalino
Browse files

Improving Prometheus metrics declaration.

parent 598ccfa6

src/opticalattackdetector/service/OpticalAttackDetectorServiceServicerImpl.py

+8 −28
Original line number Diff line number Diff line
@@ -19,6 +19,8 @@ import random 
import grpc

import numpy as np

import redis

from prometheus_client import Histogram



from common.Constants import ServiceNameEnum

from common.method_wrappers.Decorator import (MetricsPool, MetricTypeEnum,

                                              safe_and_metered_rpc_method)
@@ -36,7 +38,6 @@ from dbscanserving.client.DbscanServingClient import DbscanServingClient 
from monitoring.client.MonitoringClient import MonitoringClient

from opticalattackmitigator.client.OpticalAttackMitigatorClient import \

    OpticalAttackMitigatorClient

from prometheus_client import Histogram



LOGGER = logging.getLogger(__name__)
@@ -51,29 +52,11 @@ METRICS_POOL_DETAILS = MetricsPool( 
    },

)





metric_labels = dict(operation="detect")

histogram_duration: Histogram = METRICS_POOL_DETAILS.get_or_create(

METRIC_LABELS = dict(operation="detect")

HISTOGRAM_DURATION: Histogram = METRICS_POOL_DETAILS.get_or_create(

    "details", MetricTypeEnum.HISTOGRAM_DURATION

)





DETECTION_RESPONSE_TIME = Histogram(

    "tfs_opticalattackdetector_inference_response_time",

    "Time taken by the inference component to reply",

)

MONITORING_RESPONSE_TIME = Histogram(

    "tfs_opticalattackdetector_monitoring_response_time",

    "Time taken by the monitoring component to reply",

)

MITIGATION_RESPONSE_TIME = Histogram(

    "tfs_opticalattackdetector_mitigation_response_time",

    "Time taken by the attack mitigator to reply",

)

CACHE_RESPONSE_TIME = Histogram(

    "tfs_opticalattackdetector_cache_response_time", "Time taken by the cache to reply"

)



monitoring_client: MonitoringClient = MonitoringClient()

dbscanserving_client: DbscanServingClient = DbscanServingClient()

attack_mitigator_client: OpticalAttackMitigatorClient = OpticalAttackMitigatorClient()
@@ -197,8 +180,7 @@ class OpticalAttackDetectorServiceServicerImpl(OpticalAttackDetectorServiceServi 
                    detection_sample.append(random.uniform(5000.0, 6000.0))



        # adding the sample to the cache and recovering the cache

        with histogram_duration.labels(step="cachefetch", **metric_labels).time():

            # with CACHE_RESPONSE_TIME.time():

        with HISTOGRAM_DURATION.labels(step="cachefetch", **METRIC_LABELS).time():

            r.rpush(

                "opm_{}".format(s_uuid.replace("-", "_")),

                pickle.dumps(tuple(detection_sample)),
@@ -215,8 +197,7 @@ class OpticalAttackDetectorServiceServicerImpl(OpticalAttackDetectorServiceServi 
                detection_sample.features.append(feature)

            detection_request.samples.append(detection_sample)



        with histogram_duration.labels(step="uldetection", **metric_labels).time():

            # with DETECTION_RESPONSE_TIME.time():

        with HISTOGRAM_DURATION.labels(step="uldetection", **METRIC_LABELS).time():

            response: dbscan.DetectionResponse = dbscanserving_client.Detect(

                detection_request

            )
@@ -236,15 +217,14 @@ class OpticalAttackDetectorServiceServicerImpl(OpticalAttackDetectorServiceServi 
        else:

            kpi.kpi_value.int32Val = 0



        with histogram_duration.labels(step="includekpi", **metric_labels).time():

            # with MONITORING_RESPONSE_TIME.time():

        with HISTOGRAM_DURATION.labels(step="includekpi", **METRIC_LABELS).time():

            monitoring_client.IncludeKpi(kpi)



        # if -1 in response.cluster_indices:  # attack detected

        if kpi.kpi_value.int32Val == -1:

            attack = AttackDescription()

            attack.cs_id.uuid = request.service_id.service_uuid.uuid

            with histogram_duration.labels(step="mitigation", **metric_labels).time():

            with HISTOGRAM_DURATION.labels(step="mitigation", **METRIC_LABELS).time():

                # with MITIGATION_RESPONSE_TIME.time():

                response: AttackResponse = attack_mitigator_client.NotifyAttack(attack)