Backfill marks tasks as complete but not DAG runs

[Johnny Brown]

I have a DAG that was turned off for a while as it was broken. Once I got around to deploying the fixed version, I ran a backfill (using mark-success, as the underlying need was getting solved by other means), which resulted in a tree view showing all the tasks as complete, but without creating dag runs for the backfilled dates.

I tried using backfill, using trigger_dag, and waiting for the scheduler, but it seems nothing will mark these dag runs as success. When I click one of the white dag_run squares, I see the "mushroom cloud" page, with a message saying 

DataError: (psycopg2.DataError) invalid input syntax for integer: "undefined"
LINE 3: WHERE dag_run.id = 'undefined'

I've also tried restarting the webserver and scheduler. Any idea what to do here? Airflow version is 1.6.2

Thanks,

Johnny

[Jeremiah Lowin]

Hi Johnny,

Right now, DagRuns are only used by the Scheduler, not backfills. So what you're seeing is correct, though we're aware it's confusing and have plans to address it. Hard to say why the DagRuns in question are stalled -- you may want to try simply deleting them and letting the Scheduler pick up from the latest point (the scheduler will pick up from the latest finished task, including tasks run via backfill). If you're feeling brave, there are a number of fixes in the current GitHub master that address dagruns and/or backfills that loop forever. I don't think they will be part of the forthcoming 1.7.1 release but certainly 1.8 (if not a future 1.7.x).

The issue with the mushroom cloud is definitely a bug and please file an issue for it!