Description of problem: When performing a minor update on an OSP10 overcloud that uses the ceph-external Heat template where the nodes only have the Ceph tools repo enabled (for client library updates), the update fails due to an attempt to reinstall ceph-osd. The ceph-osd package is removed as part of the yum update script, but it is reinstalled due to the presence of ceph-osd in ceph::params::packages in ceph-external.yaml. Removing ceph-mon and ceph-osd from that list in ceph-external.yaml mitigates the issue. How reproducible: 1. Deploy an overcloud with OSP10 using external Ceph 2. Enable the rhel-7-server-rhceph-2-tools-rpms on all controller and compute nodes. 3. Do not enable any other Ceph repos. (Customers using external Ceph will not have subscriptions to these repos for OSP nodes) 4. Attempt a minor update of the overcloud. Actual results: A SoftwareDeployment fails during the update. Viewing the output of the failed deployment shows either a version conflict between Ceph packages or a failed attempt to install ceph-osd depending on whether ceph-mon and ceph-osd were stripped from the overcloud image. Expected results: Update should succeed, preferably with the automatic removal of ceph-mon and ceph-osd. Additional info:
While we work on a fix, I believe a workaround for this is to add to the deployment line an environment file with some custom hieradata, for example: parameter_defaults: ExtraConfig: ceph::params::packages: - ceph-base - ceph-mon
Verified on openstack-tripleo-heat-templates-5.3.10-23.el7ost.noarch
According to our records, this should be resolved by openstack-tripleo-heat-templates-5.3.10-23.el7ost. This build is available now.