Comment 2 for bug 1440201

Mitsuhiro,

Thank you for confirming this. Did you have to do anything special to get this to recreate in devstack? I haven't had any luck getting it to recreate in my devstack environment.

So you are thinking we are hitting some sort of timing issue between when the libvirt removes the block device info and when disconnect_volume() is called?

I wonder if we could add a retry in disconnect_volume around 'devices = self.connection._get_all_block_devices()' to put a sleep in and retry for cases where that does actually return devices.

The other question is if there is something we can do to detect and disconnect the orphaned iscsi session when this situation is encountered at attach time. Looking at _connect_to_iscsi_portal, I guess that be easier said than done given that we don't have access to the XML indicating what volumes are actually being used.

Will keep trying to recreate here so I can play around with this more. Let me know what you figure out.

Thanks for your help!