[OmniOS-discuss] RSF-1/ZFS panics node when offlining one iSCSI storage mirror

[Stephan Budach]

Hi,

I have set up a rather simple RSF-1 project, where two RSF-1 nodes 
connect to two storage heads via iSCSI. I have deployed one network and 
two disc heatbeats and I was trying all sorts of possible failures, when 
I noted that one node would panic, if I offlined an iSCSI target on one 
storage node and thus shutting down one side of a zpool mirror 
completely. Issueing a zpool status would't return and after a while the 
host got nuked.

I then onlined the target again and waited until the node returned and 
than removed the local iSCSI initiator on the RSF-1 node instead, which 
resulted in a degraded, but functional zpool and this time, the node 
didn't get nuked.

What is the difference between these two approaches and can I setup my 
systems such as that offlining a target doesn't lead to this behaviour? 
I'd imagine, that a target failure might as well occur as any other 
sofware fault.

Thanks,
Stephan