I have a question regarding the Kubernetes Executor (running the jobs as k8s pods) using a helm chart and configmaps.

I am running on the 1.10.7 (tried older versions as well) and there seems to be a different between the pod being scheduled from the web server interface (manual trigger) and the one that's done by the scheduler.

In the case of the manual trigger the airflow_local_settings.py is passed on to the container/pod and the pod_mutation_hook is executed correctly, whereas the scheduler starts the pods and ignores the contents of the airflow_local_settings.py file.

Do you have any suggestions or have you encountered this behaviour before? Thanks!