Failed ResourceGroup members may not be replaced on rolling update
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Heat |
Fix Released
|
Medium
|
Zane Bitter |
Bug Description
When performing a rolling update of a ResourceGroup, if any of the members of the ResourceGroup are in a FAILED state they may (depending on their index in the group) be removed instead of being updated/replaced.
This is because the rolling update code assumes that resources are only added or removed at the end of the list, and names for any resources being created are selected on this basis. However, the valid_resources list of existing resources omits not only blacklisted resources (which will never be replaced) but also FAILED resources, which should be replaced. Because the names selected for new resources are merely the highest indices, and don't account for gaps, in some cases the definitions of those higher-indexed resources will be overwritten and the gap at a lower index left, so the ResourceGroup will have too few members.
Changed in heat: | |
milestone: | none → mitaka-1 |
tags: | removed: liberty-backport-potential |
Changed in heat: | |
status: | Fix Committed → Fix Released |
Fix proposed to branch: master /review. openstack. org/238306
Review: https:/