Description of problem: Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. Have Ironic overcloud, with nova-compute running on the controllers. 2. Perform the controller replacement process, with the replacement controller having a different hostname from the original. Actual results: 'splosions. Instances with instance.host set to the replaced controller become unmanageable, as their instance.host now refers to a nova-compute service that no longer exists. Expected results: No 'splosions. Additional info:
Coming out of DFG:Compute tech call on October 27th [1], filing this BZ to track the existing WIP patch at [2] [1] https://docs.google.com/document/d/1vw9wJndmU_4U9mha4HhClEG4jAZPEItn4FysHEqhO_o/edit#heading=h.edqw67arx7si [2] https://review.opendev.org/c/openstack/nova/+/813897
Update: The upstream nova project has rejected automatic remedy of the the field value with the existing hash ring design. The forward only solution which was agreed upon during the last upstream PTG is to basically utilize a shard key to be added to ironic's API in order to create reliable pairing/matching of nodes. This is presently being worked in two separate spec documents upstream and contributors are already working on code to make this happen. * https://review.opendev.org/c/openstack/nova-specs/+/862833 * https://review.opendev.org/c/openstack/ironic-specs/+/861803
The work has merged in the Ironic project upstream. Community contributors are pushing forward the nova side of the equation which is anticipated to merge in the Bobcat cycle upstream. Given we were not able to negotiate this to be a bugfix, and the agreed upon solution is quite intertwined, I'm going to mark this as a future feature.