All for it. I think part of the k8s configuration being in the "core" is because of its relation to K8s executor. We should make sure that Dask, Celery, Kubernetes executors should come with their own "extras" (i.e. need to install a provider) and have an "own" configuration. I will require some core behaviour change as we "hard-code' some of those executor checks in core code.
I think, this is very much coupled with the discussion here: https://lists.apache.org/thread/cycwt15bqr76vdol87477v808f1c2d1x where the conclusion is that we should finally agree on the Executor responsibilities and interface. Maybe I am wrong but I think the separation should be solved together, there is a good reason why some k8s settings are in core. J. On Sat, Apr 2, 2022 at 1:00 AM Daniel Standish <[email protected]> wrote: > > Small clarification, KPO doesn't currently take into account the default > namespace specified in kube settings. AFAIK it considers the following: > > in_cluster > cluster_context > config_file > enable_tcp_keepalive > verify_ssl > > And so my proposal would be that, after a deprecation period, we only derive > these settings from the hook + KPO params. During deprecation period, we > would emit a warning when the effective setting (i.e. the setting derived > after applying all config sources in order of precedence) would be different > after core airflow K8s config is excluded from KPO config. >
