We follow the strategy to re-build Docker with each deployment, because sometimes DAG changes can include new or updated dependencies. This Docker-first approach makes local development nice too (run same Docker images on local machine during debug as production). This isnt painful for us because we’ve also automated the dockerizarion, deployment, and graceful restart of the Airflow cluster with our CLI and API, and because we’re using a Docker repository, roll back is possible.
-Ry Sent from mobile On Aug 21, 2018 at 8:25 AM, Dev Aleksander <d...@sumowski.pl> wrote: Hi all, in the place I'm currently at we're building and redeploying a new set of containers with the latest code every time we want to update a DAG. That doesn't feel like the fastest way. Anyone can share their approach? Thanks, Aleksander
