HA: new container naming scheme "cluster-common-tag" make deployment fail
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
tripleo |
Fix Released
|
High
|
Damien Ciabrini |
Bug Description
Since [1,2], HA containers are now configured to use a new image name scheme that acts as an intermediate tag which enables changing container image name during a minor update without service disruption.
The feature has passed CI but was only exercised for docker, since currently we don't support podman in pacemaker on CentOS 7.0.
Unfortunately, podman has a slightly different semantics than docker when it comes to tagging images with arbitrary tag name, i.e. when doing the following high-level command:
# podman tag undercloud-
podman prepends 'localhost/' in front of the new tag:
# podman images | grep cluster
localhost/
This in itself is not a big deal because one can still run podman commands without passing "localhost/" for the image name:
# podman run --rm --net=host -it cluster-
bin boot dev etc home lib lib64 lost+found media mnt openstack opt proc root run sbin srv sys tmp usr var
The problem is that in pacemaker, the podman resource agent uses regular expressions to check whether image tag 'cluster-
[1] Id369154d147cd5
[2] I7a63e8e2d9457c
Fix proposed to branch: master /review. opendev. org/701388
Review: https:/