QE reproduced this issue,but it should be a normal issue, add more info: Nginx* pods were created manually through "oc create -f xx.yaml" without dc or rc. So after "Drain Node", it should disappear. At the same time, those managed by dc/rc would be restored. So it is expected. But found another issue in above env, docker-registry and router can not be re-deployed, qe will try to reproduce this new find.
Another issue about docker-registry and router can not be re-deployed which resulted that other app pods going error state. Checked above cluster, it should be a normal issue too. Because a pre-release docker registry need specified in /etc/sysconfig/docker, but not default one registry.access.redhat.com. So when upgrade 3.5 to 3.6 with quick installer, user should specify this option in the installer.cfg.yml. case-quick installer openshift_docker_additional_registries: x.x.com So, according to QE's reproduce and test, it should not be a bug.
Thanks, that seems correct. Setting environment variables when running atomic-openshift-utils should enable those pre-release registries. OO_INSTALL_INSECURE_REGISTRIES=registry.ops.openshift.com OO_INSTALL_ADDITIONAL_REGISTRIES=registry.ops.openshift.com