How to clear an older TaskInstance and run it with the latest dag version?

[Kjir]

We recently updated our Airlflow from 2.x to the latest 3.x. I now need to re-run some older instances that were created with the 2.x version of Airflow, but the task is never enqueued with this log message:

2026-07-07T15:39:54.730911Z [warning  [] TaskInstance <TaskInstance: [...]> does not have a dag_version_id set, cannot be enqueued. This would get unstuck and dag_version_id updated. [airflow.jobs.scheduler_job_runner.SchedulerJobRunner] loc=scheduler_job_runner.py:1033

How can I re-run this task instance and use the latest version of the dag for it? I do not see any option for that in the clear modal.

[apoorva-01]

Short version: those task instances have no dag_version_id because 2.x had no DAG versioning and the upgrade doesn't backfill it. The 3.x scheduler refuses to enqueue any TI that's missing one, which is exactly the warning you're seeing at scheduler_job_runner.py.

It's meant to self-heal: when the run goes back through verify_integrity, the scheduler stamps the latest DagVersion onto the run's task instances and they enqueue. For that to actually fire, two things have to hold:

1. The DAG has to be parsed and active in 3.x right now, so there's a current row in the dag_version table to point at. If the DAG got dropped from your bundle or isn't parsing, get_latest_version returns nothing and there's nothing to backfill from. That's the usual reason one of these stays wedged forever instead of recovering on its own.
2. Clear the whole DAG run, not just the single task. Clearing one TI doesn't re-verify the run, so the null dag_version_id never gets filled in. Clearing the run puts it back through the scheduler, which re-verifies and backfills the version onto the tasks.

One version note: the "skip and backfill instead of crashing the scheduler" handling landed in 3.1.8 (#61813). If you're below that, upgrade first, otherwise these null-version TIs can wedge the scheduler.

[Kjir]

Clearing the DAG run not only does not work, unfortunately, but also doesn't propose an option to clear future DAG runs. Basically there is not easy way to clear hundreds of DAG runs.

My Airflow version is currently 3.2.2, but I see 3.3.0 was just released and should have an improvement in this regard. I'll try to update and see whether that works.

[Kjir]

While I wasn't able to make this work with Airflow 3.2.2, Airflow 3.3.0 does provide the option of using the latest version on clear, and that fixes the issue.

[Kjir]

I spoke too soon: some DAGs I seem to be able to do that, but other dags do not present the checkbox to run with the latest version. It's unclear why.

[Kjir]

Defining rerun_with_latest_version=True at the DAG level after the upgrade to 3.3.0 seems to work, even though the checkbox doesn't always consistently show up in the UI.

At the very least this obstacle seems to have been surpassed.