Currently, at least on astronomer cloud, it seems that all deployments will force a server reboot, because dags and helpers are baked into the image.
In my experience with airflow so far, I have used a bind-mounted directory to store dags and helpers, and have found that I don’t need to alter the image very often.
Being able to do this on astronomer would be nice, because then you don’t have to worry about deployments taking down the server unless you need to change the image. This could be important if you have a lot of long-running tasks.
Am I correct that this is not possible on astronomer cloud? What about enterprise? Is it in the roadmap?
You are correct that on a deploy it will cause the webserver, scheduler and celery workers to reboot. This is true for cloud and enterprise. While there is nothing on the road map at the moment to change that, the KubernetesExecutor should offer some relief and it should be available in the the next release of Cloud.
With the KubernetesExecutor, there are no celery worker pods that need to reboot. just the webserver and scheduler. All tasks are running in a pod that only exist for the duration of that task. Upon deployment, current running task pods will be allowed to complete, but new tasks started after the deploy will utilized the newly deployed code. So this should solve the issue of long running tasks.