Description of problem: Following up BZ 1986406, it seems our doc don't tell to migrate workloads off the to-be-replaced controllers nodes. This would be nonsensical in a libvirt context, but with Ironic, nova-compute services are running on the controllers, so the workloads that they host need to migrated. For example, in BZ 1986406, they cannot delete an old nova-compute service db entry because it has no compute nodes associated with it (as they've been rebalanced by Ironic to other nova-computes), but they still have instances that have the `host` field set to the old controller hostnames. Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
Once BZ 2017980 is done, we can update the documentation to state that for the Ironic controller replacement case, nova-compute services on the to-be-replaced controller needs to be disabled, and then the Ironic periodic rebalance needs to happen to re-assign any instances on that host to other hosts. Until then, nothing much to do.
*** This bug has been marked as a duplicate of bug 1986406 ***