Airflow metrics with Prometheus - Commands & Configuration

Introduction

When running many workflows, it is hard to know how well Airflow is performing or if tasks are failing often. Prometheus helps by collecting detailed metrics from Airflow so you can watch its health and performance in real time.

When you want to monitor how many tasks are running or failing in Airflow over time

When you need to alert your team if Airflow scheduler or workers stop working

When you want to see how long tasks take to run and identify slow workflows

When you want to collect metrics from Airflow and visualize them in Grafana dashboards

When you want to track Airflow resource usage and task success rates automatically

Config File - airflow.cfg

airflow.cfg

[metrics]
# Enable StatsD metrics exporter
statsd_on = True
statsd_host = localhost
statsd_port = 8125
statsd_prefix = airflow

[metrics.prometheus]
enabled = True
port = 9090

[webserver]
# Expose Prometheus metrics endpoint
expose_prometheus = True

This configuration enables Airflow to send metrics to Prometheus.

statsd_on turns on metrics collection.

statsd_host and statsd_port tell Airflow where the StatsD server listens.

expose_prometheus in the webserver section allows Prometheus to scrape metrics directly from Airflow's webserver.

Commands

Starts the Airflow webserver which now exposes Prometheus metrics on port 8080.

Terminal

airflow webserver --port 8080

Expected Output

[2024-06-01 12:00:00,000] {webserver.py:123} INFO - Starting Airflow webserver on port 8080

[2024-06-01 12:00:01,000] {prometheus.py:45} INFO - Prometheus metrics endpoint enabled at /metrics

→

--port - Specifies the port where the Airflow webserver listens

Starts the Airflow scheduler which runs tasks and updates metrics that Prometheus will collect.

Terminal

airflow scheduler

Expected Output

[2024-06-01 12:00:05,000] {scheduler.py:78} INFO - Starting the scheduler

[2024-06-01 12:00:06,000] {scheduler.py:90} INFO - Scheduler heartbeat

Fetches the raw Prometheus metrics from Airflow's metrics endpoint to verify metrics are exposed.

Terminal

curl http://localhost:8080/metrics

Expected Output

# HELP airflow_task_duration_seconds Duration of Airflow tasks in seconds

# TYPE airflow_task_duration_seconds histogram

airflow_task_duration_seconds_bucket{task_id="example_task",le="5"} 3

airflow_task_duration_seconds_bucket{task_id="example_task",le="10"} 5

# HELP airflow_scheduler_heartbeat_seconds Time between scheduler heartbeats

# TYPE airflow_scheduler_heartbeat_seconds gauge

airflow_scheduler_heartbeat_seconds 1.0

Starts Prometheus server with configuration to scrape Airflow metrics from the webserver endpoint.

Terminal

prometheus --config.file=prometheus.yml

Expected Output

level=info ts=2024-06-01T12:00:10.000Z caller=main.go:123 msg="Starting Prometheus" version="2.45.0"

level=info ts=2024-06-01T12:00:10.001Z caller=main.go:456 msg="Loading configuration file" filename=prometheus.yml

level=info ts=2024-06-01T12:00:10.002Z caller=main.go:789 msg="Server is ready to receive web requests."

→

--config.file - Specifies the Prometheus configuration file to use

Key Concept

If you remember nothing else from this pattern, remember: enabling Prometheus metrics in Airflow lets you monitor task and scheduler health in real time.

Common Mistakes

Not enabling the Prometheus metrics endpoint in airflow.cfg

Prometheus cannot scrape metrics if Airflow does not expose them, so no data is collected.

Set 'expose_prometheus = True' under the [webserver] section in airflow.cfg.

Starting Prometheus without configuring it to scrape Airflow metrics

Prometheus will run but will not collect any Airflow metrics, making monitoring useless.

Add Airflow's metrics endpoint to prometheus.yml scrape_configs.

Running Airflow webserver on a different port than Prometheus expects

Prometheus will fail to connect and collect metrics, causing missing data.

Ensure the webserver port matches the one configured in Prometheus scrape target.

Summary

Configure airflow.cfg to enable Prometheus metrics and expose the metrics endpoint.

Start Airflow webserver and scheduler to generate and expose metrics.

Verify metrics are available by fetching the /metrics endpoint.

Run Prometheus with a config that scrapes Airflow metrics for monitoring.