Live-Migration fails for an instance with 5 or more non-bootable volumes (only FC volumes) attached when live-migration operation is performed back and forth.

[Sumit Girish Shatwara]

Pre-requisite:
Fibre channel zone manager must be used with Brocade zone driver and fabric details.

Steps to Reproduce:

1. Create 2 instances with bootable volumes (3parfc) of 15 GB (Make sure both the instances are on different compute nodes).
2. Create and attach 5 non bootable volumes (3parfc) to each instance.
3. Migrate instance1 to compute2. Migration happens here.
4. Migrate instance1 back from compute2 to compute1. Migration failed here. But I didnt observe any error in CLI output.
5. Repeat Step 5. Migration failed again but this time with below error:

stack@be11-cp1-c1-m1-mgmt:~$ nova live-migration instance1 be11-cp1-comp0001-mgmt
ERROR (ClientException): Unexpected API Error. Please report this at http://bugs.launchpad.net/nova/ and attach the Nova API log if possible. <class 'MigrationError_Remote'> (HTTP 500) (Request-ID: req-1203ea57-0c04-445b-8d4a-cb6cc1c4c7a0)

Additional Note:
1. When the instances are present on both the compute nodes with 0, 1, 2 or 3 non-bootable volumes attached to them, live-migration operation executed back and forth passes successfully. I have not tested with 4 non-bootable volumes attached to both the instance here in a same test case.
2. After discussion and scrutinizing different log files by different team members, it was concluded that it is failing due to Authentication failure of Brocade switch.

Please find the snippet of cinder-volume logs (from controller node, where we can see error while authenticating with Brocade switch) in attachments here.

[Pankaj Khandar (pankaj-khandar)]

Any update on this?

[Angela Smith (aallen-m)]

From the log snippet submitted, this is a Nova defect. Any issue with communication of Brocade switch would be documented in the cinder.log.