Repository navigation
We need to always delete the kind cluster if we create kind cluster with Prow jobs #14
Description
Currently when we run test flows with kind, we run kubetest2 kind --up --down which will also delete the kind cluster after the test flow finishes in the normal way. But if the Prow job is interrupted in the middle, the cluster will not be properly deleted thus will be causing memory leakage, as described in kubernetes-sigs/kind#303 (comment)
We should add a trap command to make sure the cluster can still be cleaned up even when the Prow job is interrupted.
Activity
Apply proposal (knative#14)
BenTheElder commentedon Jan 28, 2021
you also most likely need to set the prowjob timeout / grace period to something other than the 15s? default. bootstrap.py used to do 15m by default, so prow.k8s.io uses that as the default (it's configurable for a prow instance), otherwise your scripts don't get much time to cleanup.
unfortunately nested kubernetes brings quite a suite of problems, you probably want to look through #303 for anything else you've missed (e.g. inotify).