Scheduler "Parent instance is not bound to a Session" Error

[ahipp13]

Under which category would you file this issue?

Airflow Core

Apache Airflow version

3.2.2

What happened and how to reproduce it?

I am running Airflow 3.2.2 on the 1.18.0 Helm Chart using the Kubernetes executor. Airflow 3.2.1 worked just fine, but when I upgraded to Airflow 3.2.2 my schedulers started throwing the below error. I run 2 schedulers at once, so I think what is happening is somehow there is a race condition where both of them are trying to touch something the other has touched which is causing the error. At first this error only happens to one scheduler, and then it affects both of them and brings both of them down.

2026-06-11T17:13:14.598244Z [error ] Exception when executing SchedulerJob._run_scheduler_loop [airflow.jobs.scheduler_job_runner.SchedulerJobRunner] loc=scheduler_job_runner.py:1551
Traceback (most recent call last):
File "/home/airflow/.local/lib/python3.12/site-packages/airflow/jobs/scheduler_job_runner.py", line 1546, in _execute
self._run_scheduler_loop()
File "/home/airflow/.local/lib/python3.12/site-packages/airflow/jobs/scheduler_job_runner.py", line 1612, in _run_scheduler_loop
self.adopt_or_reset_orphaned_tasks()
File "/home/airflow/.local/lib/python3.12/site-packages/airflow/utils/session.py", line 100, in wrapper
return func(*args, session=session, **kwargs) # type: ignore[arg-type]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/airflow/jobs/scheduler_job_runner.py", line 2814, in adopt_or_reset_orphaned_tasks
for attempt in run_with_db_retries(logger=self.log):
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/tenacity/init.py", line 438, in iter
do = self.iter(retry_state=retry_state)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/tenacity/init.py", line 371, in iter
result = action(retry_state)
^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/tenacity/init.py", line 393, in
self._add_action_func(lambda rs: rs.outcome.result())
^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/concurrent/futures/_base.py", line 449, in result
return self.__get_result()
^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/concurrent/futures/_base.py", line 401, in __get_result
raise self.exception
File "/home/airflow/.local/lib/python3.12/site-packages/airflow/jobs/scheduler_job_runner.py", line 2864, in adopt_or_reset_orphaned_tasks
reset_tis_message.append(repr(ti))
^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/airflow/models/taskinstance.py", line 1134, in repr
return prefix + f"[{self.state}] ti_id={self.id}>"
^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/sqlalchemy/orm/attributes.py", line 569, in get
return self.impl.get(state, dict) # type: ignore[no-any-return]
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/sqlalchemy/orm/attributes.py", line 1096, in get
value = self._fire_loader_callables(state, key, passive)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/sqlalchemy/orm/attributes.py", line 1129, in fire_loader_callables
return callable(state, passive)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/sqlalchemy/orm/strategies.py", line 578, in call
return strategy._load_for_state(state, passive)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/airflow/.local/lib/python3.12/site-packages/sqlalchemy/orm/strategies.py", line 536, in _load_for_state
raise orm_exc.DetachedInstanceError(
sqlalchemy.orm.exc.DetachedInstanceError: Parent instance <TaskInstance at 0x7fb1b521e390> is not bound to a Session; deferred load operation of attribute 'state' cannot proceed (Background on this error at: https://sqlalche.me/e/20/bhk3)
2026-06-11T17:13:14.601149Z [info ] Shutting down Kubernetes executor [airflow.providers.cncf.kubernetes.executors.kubernetes_executor.KubernetesExecutor] loc=kubernetes_executor.py:867

What you think should happen instead?

Schedulers should run as normal together

Operating System

(airflow)cat /etc/os-release ID=chainguard NAME="Chainguard" PRETTY_NAME="Chainguard" VERSION_ID="20230214" HOME_URL="https://chainguard.dev/"

Deployment

Official Apache Airflow Helm Chart

Apache Airflow Provider(s)

No response

Versions of Apache Airflow Providers

No response

Official Helm Chart version

1.18.0

Kubernetes Version

PS C:\Users\ahippe1> k version Client Version: v1.34.5-dispatcher Kustomize Version: v5.7.1 Server Version: v1.34.3-gke.400

Helm Chart configuration

My values file is huge, so if there are specific pieces you want to look at please let me know and I will paste them in

Docker Image customizations

No response

Anything else?

When I run the command:

UPDATE task_instance
SET state = 'failed'
WHERE state IN ('running', 'queued')

in the Airflow database, it shuts down all running DAGs, but then it fixes the problem and it does not happen...

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[ahipp13]

Anybody have any idea why this is happening?

[DRACULA1729]

Took a look at this one and I think I found it.

The KubernetesExecutor grew a helper called _alive_other_scheduler_job_ids() (added in #66400, shipped in a recent cncf-kubernetes provider release). It opens its own with create_session() as session: block, but by the time it runs the scheduler is already inside the adopt_or_reset_orphaned_tasks transaction.

The catch: settings.Session is a thread-local scoped_session, so that nested create_session() hands back the same session the scheduler is already using. When the helper's with block exits it calls session.close(), which detaches every TaskInstance the scheduler just loaded. The next repr(ti) then triggers a deferred load of ti.state on a detached instance, and that's your DetachedInstanceError.

Fits both things you saw: it only bites under the KubernetesExecutor, and the UPDATE ... SET state='failed' makes it go away because there are no orphaned tasks left to reset, so the loop calling repr(ti) never runs.

Fix is to give that helper its own independent (non-scoped) session so closing it can't pull the rug out from under the caller. PR coming shortly.

[DRACULA1729]

Quick correction to my last comment: I went to write the fix and it turns out it's already been done. PR #67850 ("Fix scheduler crashloop from KubernetesExecutor completed-pod adoption") makes exactly the change I described, switching that helper to an independent non-scoped session.

It shipped in apache-airflow-providers-cncf-kubernetes 10.18.0 (released June 8). The bad code only ever existed in provider 10.17.1 (May 19); 10.17.0 and earlier don't have it. So you're almost certainly on 10.17.1, and bumping the cncf-kubernetes provider to 10.18.0 or newer should clear it up. This is purely on the provider side, so your Airflow core version (3.2.2) doesn't matter here.

No new PR needed from me then. I think this can be closed as already-fixed by #67850. Sorry for the false start!

[ahipp13]

@DRACULA1729 that would make sense, as I install my packages with the constraints file and 10.17.1 is still pinned to the constraints file for 3.2.2... Thank you for the response!!! Helps me out a ton

[ahipp13]

I am almost wondering in situations like this if I should even be using the constraints file when I install my Python packages