prometheus exporter for Django Q

django_q_exporter.py

from datetime import timedelta

from django.conf import settings
from django.http import HttpResponse, HttpResponseNotFound
from django.utils import timezone
from django_q import models
from django_q.monitor import Stat
from django_redis import get_redis_connection
from prometheus_client import Gauge, generate_latest

# Env moved from settings to here
ENABLE_EXPORTER = True
DJANGO_Q_REDIS_DB_NAME = "1"
DJANGO_Q_EXPORTER_INTERVAL = 30 # seconds

CONTENT_TYPE_LATEST = str("text/plain; version=0.0.4; charset=utf-8")

django_q_state_metric = Gauge(
    "django_q_state",
    "State of Django Q Cluster: Starting: 0, Idle: 1, Working: 2, Stopping: 3, Stopped: 4",
    ["env", "cluster_id", "state"],
)

django_q_uptime_metric = Gauge(
    "django_q_uptime_seconds", "Django Q Uptime Seconds", ["env", "cluster_id", "state"]
)
django_q_task_q_size_metric = Gauge(
    "django_q_task_q_size", "Django Q Task Q Size", ["env", "cluster_id", "state"]
)
django_q_task_successes_metric = Gauge(
    "django_q_task_successes",
    "Successes tasks in Django Q",
    [
        "env",
    ],
)
django_q_task_successes_relative_time_metric = Gauge(
    "django_q_task_successes_relative_time_metric",
    "Successes task in Django Q in relative time",
    ["env", "interval"],
)
django_q_task_failures_metric = Gauge(
    "django_q_task_failures",
    "Failures tasks in Django Q",
    [
        "env",
    ],
)
django_q_task_failures_relative_time_metric = Gauge(
    "django_q_task_failures_relative_time_metric",
    "Failures task in Django Q in relative time",
    ["env", "interval"],
)
django_q_task_schedule_metric = Gauge(
    "django_q_task_schedule",
    "Schedule tasks in Django Q",
    [
        "env",
    ],
)
django_q_clusters_metric = Gauge(
    "django_q_clusters",
    "Number of Clusters in Django Q",
    [
        "env",
    ],
)
django_q_workers_metric = Gauge(
    "django_q_workers",
    "Number of Workers in Django Q",
    [
        "env",
    ],
)
django_q_task_per_houre_metric = Gauge(
    "django_q_tasks_per_hour",
    "Tasks Per last Hour in Django Q",
    [
        "env",
    ],
)
django_q_task_per_day_metric = Gauge(
    "django_q_tasks_per_day",
    "Tasks Per last 24 Hours in Django Q",
    [
        "env",
    ],
)
django_q_redis_connections_metric = Gauge(
    "django_q_redis_connections",
    " Django Q Redis Connections",
    [
        "env",
    ],
)

django_q_states = {"Starting": 0, "Idle": 1, "Working": 2, "Stopping": 3, "Stopped": 4}


def get_django_q_redis_connections(django_q_redis_db_name: str = "1"):
    redis_connections = 0
    redis_con = get_redis_connection("django_q")
    # filter django_q redis connections
    for c in redis_con.client_list():
        if c["db"] == django_q_redis_db_name:
            redis_connections += 1
    redis_con.close()
    return redis_connections


def get_metrics(*args, **kwargs):

    if ENABLE_EXPORTER:
        workers = 0

        # cluster(s) metrics
        for stat in Stat.get_all():
            workers += len(stat.workers)
            django_q_state_metric.labels(
                f"{settings.ENV}", str(stat.cluster_id), stat.status
            ).set(django_q_states.get(stat.status))
            django_q_uptime_metric.labels(
                f"{settings.ENV}", str(stat.cluster_id), stat.status
            ).set(stat.uptime())
            django_q_task_q_size_metric.labels(
                f"{settings.ENV}", str(stat.cluster_id), stat.status
            ).set(float(stat.task_q_size))

        # general metrics
        django_q_task_successes_metric.labels(
            f"{settings.ENV}",
        ).set(models.Success.objects.count())
        django_q_task_successes_relative_time_metric.labels(
            f"{settings.ENV}", f"{DJANGO_Q_EXPORTER_INTERVAL}"
        ).set(
            models.Success.objects.filter(
                stopped__gte=timezone.now() - timedelta(seconds=30)
            ).count()
        )
        django_q_task_failures_metric.labels(
            f"{settings.ENV}",
        ).set(models.Failure.objects.count())
        django_q_task_failures_relative_time_metric.labels(
            f"{settings.ENV}", f"{DJANGO_Q_EXPORTER_INTERVAL}"
        ).set(
            models.Failure.objects.filter(
                stopped__gte=timezone.now() - timedelta(seconds=30)
            ).count()
        )
        django_q_task_schedule_metric.labels(
            f"{settings.ENV}",
        ).set(models.Schedule.objects.count())
        django_q_clusters_metric.labels(
            f"{settings.ENV}",
        ).set(len(Stat.get_all()))
        django_q_workers_metric.labels(
            f"{settings.ENV}",
        ).set(workers)
        django_q_task_per_houre_metric.labels(
            f"{settings.ENV}",
        ).set(
            models.Success.objects.filter(
                stopped__gte=timezone.now() - timedelta(hours=1)
            ).count()
        )
        django_q_task_per_day_metric.labels(
            f"{settings.ENV}",
        ).set(
            models.Success.objects.filter(
                stopped__gte=timezone.now() - timedelta(hours=24)
            ).count()
        )
        django_q_redis_connections_metric.labels(
            f"{settings.ENV}",
        ).set(
            get_django_q_redis_connections(
                django_q_redis_db_name=DJANGO_Q_REDIS_DB_NAME
            )
        )

        return HttpResponse(generate_latest(), content_type=CONTENT_TYPE_LATEST)

    else:
        return HttpResponseNotFound(
            """
		<!doctype html>
		<html lang="en">
		<head>
		  <title>Not Found</title>
		</head>
		<body>
		  <h1>Not Found</h1><p>The requested resource was not found on this server.</p>
		</body>
		</html>
		"""
        )