Airflow triggerer

Airflow’s triggerer runs deferrable operators. If you use sensors in deferred mode - and in a modern deployment you almost certainly do - this process is what resumes them.

Key: airflow-triggerer · applied automatically to services emitting airflow.triggers. or airflow.triggerer metrics.

A dead triggerer is invisible everywhere else

A deferred task is not running and not failing. It is waiting - which looks identical to waiting correctly. So when the triggerer stops, nothing on the scheduler reports anything: no task failures, no failed runs, no alert. A fleet of tasks simply never resumes and never complains.

That is the whole case for this type. See Apache Airflow for the scheduler, the DAG processor for parsing, and the Airflow page for setup, which is the same for all three.

Starter checks

Check	Condition	Severity	Why it matters
Triggerer not running	airflow.triggerer_heartbeat stops arriving	Critical	Every deferred task stays deferred, silently.
Trigger blocking the main thread	airflow.triggers.blocked_main_thread rising	Warning	One trigger doing synchronous work starves every other deferred task on the process.
Triggerer at capacity	airflow.triggerer.capacity_left < 1	Warning	New deferred tasks queue behind the ones running.

The heartbeat fires on the absence of its metric, since a dead process emits nothing left to threshold, and is guarded to only fire for a metric that has reported before.

Tuning notes

• Blocked main thread is a code smell with a metric. It means a custom trigger is doing blocking I/O in an async context. The firing tells you it happened; finding which trigger means reading the triggerer’s logs.
• Capacity at < 1 fires at saturation, not before. Raise it to a fraction of default_capacity to hear about it while there is still room.
• Run more than one triggerer if capacity is a recurring firing rather than an incident: they partition work between them.