We can only deploy to Heroku now when there are no delayed_jobs running. The reason being that upon deployment, Heroku sends a
SIGTERM which eventually results in a
SIGKILL on the job. This creates problems where I have to clean up the docker containers manually and may even leave our data in a dirty state (being cut off when POSTing the results).
We’re current raising an exception whenever we get a
SIGTERM but that is still bad.
Hopefully someone might have a better idea.