Instead of having the 'web' pod watch started 'runner' pods to note their exit, update the database, and delete them, this makes the 'runner' pods do it themselves.
This means less processing in the 'web' pod and easier horizontal scaling, but is more dangerous: if something goes wrong in the 'runner' pod no one is deleting it.
Maybe there should still be a regular process in the 'web' pod going over terminated pods, noting problems, and reaping them.
Originally opened 2020-03-30 13:44 EDT by @remram44
Fixes #46
Instead of having the 'web' pod watch started 'runner' pods to note their exit, update the database, and delete them, this makes the 'runner' pods do it themselves.
This means less processing in the 'web' pod and easier horizontal scaling, but is more dangerous: if something goes wrong in the 'runner' pod no one is deleting it.
Maybe there should still be a regular process in the 'web' pod going over terminated pods, noting problems, and reaping them.