Closed matthieu-foucault closed 5 years ago
Somehow --wait
does wait and streams all the logs, but the error is printed et the end anyway, so I'm not sure that is the right solution...
As per the discussion in https://chat.pathfinder.gov.bc.ca/channel/devops-how-to?msg=99XhYDxRs5uM8hXfR, either using a parallel
build policy (which I see you're using in cas-postgres) or creating a build config per branch and using the SerialLatestOnly
policy are two workarounds for this
Yes, I leaned towards setting parallel
to overcome this, as Circle will attempt a build parallelism of 4 by default. Another option if this proves problematic is to build on our OKD 3.11 cluster on CI and then push to the pathfinder OCP 3.11 cluster
When using the parallel
strategy, isn't there a risk of having the wrong version being deployed? Assuming that two commits are pushed in quick succession, and the second build is quicker than the first one, then the older commit would end up being deployed last wouldn't it?
This is the nice thing about splitting off the CI
process from the CD
process. Nothing gets deployed by the Continuous Integration flow so parallel builds on CI
are safe.
The Continuous Deployment system is responsible for monitoring the commit activity on a single git branch and deploying that branch to a single namespace, so either
Got it! I believe this PR can be closed then.
If another build is running, or the new pod takes a long time to start for whatever reason,
oc start-build --follow
will timeout, and output the following:Adding the
--wait
option fixes that issue. This was tested ad-hoc with a new build that had to wait several minutes before starting.