Why is my Airflow Task stuck in "Queued" state? (5 Quick Fixes)

 

You’ve triggered your DAG, the UI shows the task as grey (Queued), but nothing happens for minutes—or hours. This is a classic Airflow bottleneck. Here is how to diagnose and fix it.

1. Check the "Concurrency" Limits

Airflow has several "safety brakes" to prevent your server from crashing. If you hit these limits, tasks will stay queued until others finish.

• parallelism: The max number of task instances that can run across your entire Airflow environment.

• dag_concurrency: The max number of tasks that can run for a single DAG.

• max_active_runs_per_dag: If you have too many "Backfills" running, new tasks won't start.

The Fix: Check your airflow.cfg or your DAG definition. Increase max_active_tasks if your hardware can handle it.

2. Is the Scheduler actually alive?

Sometimes the Airflow UI looks fine, but the Scheduler process has died or hung.

• Check the UI: Look at the top of the Airflow page. If there is a red banner saying "The scheduler does not appear to be running," that’s your answer.

• The Fix: Restart the scheduler service:

• systemctl restart airflow-scheduler

  # OR if using Docker:

  docker restart airflow-scheduler

3. "No Slots Available" in Pools

Airflow uses Pools to manage resources (like limiting how many tasks can hit a specific database at once). If your task belongs to a pool with 5 slots and 5 tasks are already running, your 6th task will stay Queued forever.

The Fix: Go to Admin -> Pools in the UI. Check if the "Default Pool" or your custom pool is full. Increase the slots if necessary.

4. Celery Worker Issues (For Production Setups)

If you are using the CeleryExecutor, the task is queued in Redis or RabbitMQ, but the Worker might not be picking it up.

• The Check: Run airflow celery inspect_short to see if workers are online.

• The Fix: Ensure your workers are pointed to the same Metadata DB and Broker as your Scheduler.

5. Resource Starvation (OOM)

If your worker node is out of RAM or CPU, it might accept the task but fail to initialize it, leading to a loop where the task stays queued.