KubernetesExecutor shadowing timeout_seconds and _request_seconds

[insomnes]

Apache Airflow Provider(s)

cncf-kubernetes

Versions of Apache Airflow Providers

apache-airflow-providers-cncf-kubernetes==10.0.1

Apache Airflow version

2.10.4

Operating System

Debian Bookworm

Deployment

Other

Deployment details

It's deployment independent and related to code directly

What happened

kube client query options of timeout_seconds and _request_seconds provided via kube_client_request_args config option are re-written via hardcoded values in executor code of KubernetesJobWatcher.

What you think should happen instead

Values set via AIRFLOW__KUBERNETES_EXECUTOR__KUBE_CLIENT_REQUEST_ARGS for timeout_seconds and _request_seconds should be respected, hardcoded values should be provided only in case of missing keys in the provided JSON.

Simple if for checking if key is already present would be enough

How to reproduce

1. Set the kube_client_request_args option with JSON: {"timeout_seconds": 1800, "_request_timeout": 60}
2. Run airflow scheduler with kubernetes executor configured
3. In logs you can see that without pod events the KubernetesJobWatcher will still die every 30 seconds instead of 60.

Anything else

The problem was introduced with:
610747d#diff-d884d637ab746b1301ce80b30ba2c1908299ba6f13edcad6535c837fb4d30938R151

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

[insomnes]

I am happy to provide the PR fixing it.

I see this behavior as a 100% bug. But I want to be sure it's not intended in any way.

[RNHTTR]

Is the idea to update these lines to default to 30/3600 but allow them to be overwritten via kube_client_request_args? If so, I think that makes sense.

[insomnes]

Yes. I was thinking about checking the presence of these keys:

        if "_request_timeout" not in kwargs:
            client_timeout = 30
            kwargs["_request_timeout"] = client_timeout

[RNHTTR]

Cool -- assigned you.

[Mizokuiam]

Thank you for raising this issue! I'll look into it and try to help if I can.