Racy usage of docker image by cassandra tests
porridge opened this issue · 2 comments
When working on mesosphere/kudo-cassandra-operator#56 I noticed there is a race condition between building+pushing a docker image and actually using it in tests, because we seem to be using the static tag which is being pushed to simultaneously from multiple different PRs.
FTR, spark operator build dynamically generates the image tag based on checksum of its inputs. That is somewhat fragile (filenames are ignored, as well as any non-hermetic dependency fetched from the internet) but we could consider doing something similar, just based on the ID of the built image.
The main issue will be plumbing the name from the image build step into the test step.
Yeah, so after some investigation I found that the green checkmark on e1619ec in mesosphere/kudo-cassandra-operator#56 was a lie: the updated image never had a chance to work so this issue is not theoretical, it's really biting us.