Comment 2 for bug 1450658

For one thing, the cinder code isn't logging the reason for the failure here:

2015-04-29 07:26:02.815 21775 ERROR cinder.volume.targets.lio [req-fec5ff89-5465-4e76-b573-6a5afc0d4ee2 a9d0160b91f2412d84972a615b7547dc 828348052e494d76a401f669f85829f3 - - -] Failed to delete initiator iqn iqn.1994-05.com.redhat:734fb33285d0 to target.

That's catching a ProcessExecutionError but not logging stdout/stderr information so it's going to be hard to debug.

That raises ISCSITargetDetachFailed which is just wrapped into the generic VolumeBackendAPIException in volume manager terminate_connection which raises it up to the API which returns a 500 in the response, which isn't helpful.

So a few things you have to unwind if you want a more specific error, and that starts with why the detach fails down in cinder.volume.targets.lio, translate that to a specific error, raise that up and eventually get that information out in the cinderclient to raise a more specific error to nova.

But as Jay says, maybe nova should just handle the 500 from cinder and continue.