+++ This bug was initially created as a clone of Bug #1568772 +++ Description of problem: dockergc pods would get crashed if docker storage driver is devicemapper. Please see Bug 1511400. Installer/OpenShift health check should stop from moving the installation forward. Version-Release number of the following components: openshift-ansible-3.9.22-1.git.7.92620c6.el7.noarch.rpm How reproducible: always Steps to Reproduce: 1. trigger installation with devicemapper docker storage driver # cat inventory <--snip--> openshift_crio_use_rpm=true openshift_crio_docker_gc_node_selector={"role": "node"} <--snip--> Actual results: Installation succeeded, but dockergc pods got crashed: # oc get po NAME READY STATUS RESTARTS AGE docker-registry-1-xb8wj 1/1 Running 0 1h dockergc-rrxsd 0/1 CrashLoopBackOff 24 1h dockergc-xj6tv 0/1 CrashLoopBackOff 23 1h registry-console-1-nsw8q 1/1 Running 0 1h router-1-tp2vq 1/1 Running 0 1h # oc logs dockergc-rrxsd I0418 08:17:17.462417 1 dockergc.go:242] docker build garbage collection daemon I0418 08:17:17.462537 1 dockergc.go:246] MinimumGCAge: {1h0m0s}, ImageGCHighThresholdPercent: 80, ImageGCLowThresholdPercent: 60 Expected results: Additional info: Please attach logs from ansible-playbook with the -vvv flag
I don't think it possible to implement this in openshift-ansible. Docker GC is being setup as a daemonset, thus its cluster-wide. Docker driver config can be different per host, so its possible to have some hosts on overlay driver and some on devicemapper - docker gc would work fine in this case. Its also possible to deploy docker gc on a cluster with devicemapper nodes and scale up new hosts with overlayfs. The issue with restarting docker gc pods should be solved by proper node labeling and nodeselector. Scott, WDYT?
My understanding is that when deploying cri-o the dockergc is created with the selector specified in openshift_crio_docker_gc_node_selector. Need to verify that the install works in scenarios when cri-o is in use but runtime=cri-o is not applied to any nodes.
There appear to be no active cases related to this bug. As such we're closing this bug in order to focus on bugs that are still tied to active customer cases. Please re-open this bug if you feel it was closed in error or a new active case is attached.