Docs › Airflow DAG monitoring with RunVouch

Airflow DAG monitoring with RunVouch

Airflow DAG monitoring for what a green DAG run cannot prove: Airflow is very good at telling you a task failed. A task that succeeds with an empty DataFrame, a DAG that is paused, or a scheduler that stopped scheduling all look like silence.

How it runs on Apache Airflow

Wrap the command of a BashOperator, or use the Python client inside a PythonOperator / @task. One RunVouch agent per DAG (or per critical task) is the usual shape.

Store the key

An Airflow Variable or Connection backed by your secrets backend; pass it into the task environment (env={"RUNVOUCH_KEY": ...}) rather than hardcoding.

Wrap the job

from airflow.decorators import dag, task
from airflow.operators.bash import BashOperator
from airflow.models import Variable
import pendulum, runvouch

@dag(schedule="0 2 * * *", start_date=pendulum.datetime(2026, 1, 1, tz="UTC"), catchup=False)
def nightly_report():
    build = BashOperator(task_id="build",
        env={"RUNVOUCH_KEY": Variable.get("runvouch_key")},
        bash_command="rv run nightly-report --evidence-file /out/report.html -- python /opt/report.py")

    @task
    def summarise():
        with runvouch.vouch("nightly-summary", evidence=lambda: {"rows": n > 0}) as run:
            n = summarise_rows(run)   # run.tool(...) per LLM call for cost and loop detection

    build >> summarise()

nightly_report()

rv fails open: if RunVouch is unreachable the job still runs and you get one warning line.

Register the cadence and caps

rv agent nightly-report --cadence 24h --grace 30m --max-runtime 1h --evidence --cap-run-cost 2

Register the agent once, from anywhere with the key. Cadence is what turns a schedule that stopped into an alert; --evidence makes a run without evidence a failure; the caps pause the agent when it overspends.

What goes silent on Apache Airflow

• A paused DAG (or one that failed to import after a bad deploy) schedules nothing; the DAG import error banner is not an alert. MISSED is.
• sla_miss_callback only fires when the scheduler is healthy enough to notice.
• A task with retries=3 around an LLM call is a retry storm with a budget; the per-run cost cap ends it.

What Apache Airflow does not tell you

Airflow's on_failure_callback, e-mail on failure and SLA misses all need a task instance that ran or was scheduled. A DAG that failed to import after a deploy shows a red banner in the UI and creates no task instances; a paused DAG creates none either; a scheduler that is down creates none for any DAG. The scheduler's own health check is an HTTP endpoint you have to poll yourself.

What RunVouch detects

Set up in two minutes

1. Get a free key (3 agents, no card) and store it where this page says.
2. wrap the job: copy the snippet above into the scheduled job.
3. Register the cadence once: rv agent nightly-report --cadence 24h --grace 30m --evidence, or let the first run create the agent and set the cadence on the dashboard.
4. Send one test alert: curl -X POST https://api.runvouch.com/v1/settings/test-alert -H "X-API-Key: $RUNVOUCH_KEY". The next missed, failed or empty run reaches the same channels.

Also running orchestrators?

Prefect. Dagster. Celery beat. APScheduler (in-process). The full list of runtimes is on integrations, and the same agent can report from several of them: one job, one cadence, wherever it happens to run.

Need a key? Get a free key · Stuck? contact