There are reasons why an OS image build may fail for reasons beyond the control of either the image build process or the cluster admin which may not necessarily indicate a problem with the build itself. For example, if the node that a build is running on is drained, the build pod would be evicted. Ideally, the BuildController could identify that the pod was evicted and recreate the pod so the build will run until completion.

Done when:

On eviction

the build is rescheduled 

On builder failure

the build is failed

(It does not have to be a complete list)

Done When:

• identify the cause of a build failure to determine if it was truly the build that failed or something else that caused the failure (e.g., evicted build pod, node failure, etc.).
• identified what build failures should be retryable without cluster admin intervention and they've been enumerated either in this card or elsewhere.