So, worst case today we should hold onto the cluster for an extra hour, or possibly reuse it if another cluster is waiting. That's not awful.
It's also exactly what we do for every other variant. Why should this be different? (Note that I think the original reason to file this ticket was to prevent reusing the logkeeper cluster, and it turns out that was fixed last year.)
The mark idle shouldn't be called if /data/infrastructure_provisioning/terraform/provisioned.${cluster} doesn't exist. It looks like we should handle that in infrastructure_teardown.py, but don't. Would it work if we called infrastructure_teardown in the /data/infrastructure_provisioning directory, instead of ${work}?
Good point. infrastructure_teardown.py should clean up regardless of where it is called (when bootstrap.production=True) but apparently doesn't. Until now it hasn't been needed. In any case, I don't see that as in scope for this ticket.
We could just close as won't fix also.
Ok, I will.
|