The result is not correct when retype storwize volume after replication promote

[yixuan zhang]

Test step:
1.SVC with Pool-1 and Pool-2 ,configure as stretched cluster.
2.Create volume type-1 backend is Pool-1, without replication.
3.Create volume type-2 backend is Pool-1, target pool for stretched cluster is Pool-2. with replication=TRUE.
4.Create a volume with type-2.there are copy0 and copy1 on Pool-1 and Pool-2.
5.Do "Cinder replication-promote", the primary volume change to Pool-2.
6.Retype volume from type-2 to type-1 to disable replication.
7.Check the volume from SVC, it was found the volume was in Pool-2.
8.Retype volume from type-1 to type-2 to enable replication again.
9.Check the SVC, it was found there are copy0 and copy1 on Pool-2, no copy on Pool-1.
We can never create a copy on Pool-1 from cinder after that.

Expected Result:
On step6, when retype volume from type-2 to type-1,the replication was disable, and the volume should go back to Pool-1.

[Jay Bryant (jsbryant)]

Adding Tao.

[TaoBai (baitao2020)]

This is complicate user case, I need to discuss with Ronen. I am wondering whether we really need to migrate to another pool after replica-promote

[TaoBai (baitao2020)]

To perfect solve this issue, we need to implement "multi-pool schedular awareness" first. This feature will be K task for storwize.So I will also postpone this ticket to K

[TaoBai (baitao2020)]

This is a feature:
Storwize should support
1. Mulipool with in one backend
2. Multipool scheduler awareness
3.Replica-promote pools change

[TaoBai (baitao2020)]

I create a bp for this https://blueprints.launchpad.net/cinder/+spec/storwize-support-muli-pool-within-one-backend-relative-features
Storwize should support
1. Mulipool with in one backend
2. Multipool scheduler awareness
3.Replica-promote pools change

K release no plan to work on this feature

[TaoBai (baitao2020)]

BTW, point 3 should depend on point 1,2