I've got the some trouble on customer site where are using openstack iscsi cinder driver. (Environment: KVM with Ubuntu12.04, Havana release, iscsi cinder driver with FAS) Let me explain the symptom of the problem as below.
1. Each VM assigned 1 volume, MAP02 VM only assigned VOL05 additionally
2. TEST_VOL_MAP01 VM was live-migrated to COMP02 host. (No problem)
-> Lun-2 assigned to COMP02 as a lun id
3.TEST_VOL_MAP01 VM was live-migrated again to COMP01 host. (No problem)
-> Lun-0 assigned to COMP01 as a lun id
4. TEST_VOL_MAP01 VM was live-migrated again to COMP02 host. (Failed)
-> IO Error related with the existing mapper file.
-> The lun mapped as lun-2 at COMP02, on the other hand, VM is working at the existing host(COMP01), Storage perspectives 1 LUN mapped to two host at once.
5. After rebooting the TEST_VOL_MAP01 VM, the VM mounted lun-2(50GB) instead of lun-0(10GB)
In my guess, it seems that nova DB doesn't refresh the backend info correctly.
Does anyone have seen this kind of error before?
I attached cinder.log and nova.log files to refer.