I don’t have specific details, but it was done likely to make sure the work happened before the core masters. Since for RU things are done in a rolling fashion, the descriptor change were made (largely) up front, and the services coming before it didn’t have any new descriptors to worry about.
As far as orchestration, the groups are processed in reverse, so if they need to be in a different spot then a group will have to be added specifically for Downgrade. Are you seeing an issue with behavior as it is today? Thanks, Nate On 3/1/17, 5:10 PM, "Di Li" <[email protected]> wrote: Hello folks, I am testing rolling ugprade in Ambari trunk code and I noticed that for rolling upgrade, update kerberos descriptors step is after Zookeeper, Ranger, Ranger KMS and Kafka. For rolling downgrade, update Kerberos descriptors step happens at the end of the entire downgrade just before the finalize downgrade step. I thought updating Kerberos descriptors should have happened before starting components. For example, express upgrade/downgrade runs that step before starting core and high level components. Are there any particular reasons that RU runs Kerberos descriptor update step the way it does ? Thanks. Di
