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

[Stephan Budach]

Hi Dan,

Am 07.03.16 um 15:41 schrieb Dan McDonald:
>> On Mar 6, 2016, at 9:44 AM, Stephan Budach <stephan.budach at JVM.DE> wrote:
>>
>> when I noted that one node would panic,
> AS A RULE -- if you have an OmniOS box panic, you should save off the corefile (vmdump.N) and be able to share it with the list.  I understand this may be an RSF-1 panic, BUT if it's not, it'd be nice to know.
>
> You can upload it to uploads.omniti.com if you wish, just request an upload token.
>
> Dan
>
thanks - I will keep that in mind and I actually had a core dump 
available, but since I was testing around, I didn't mean to occupy 
anyone's time more than absolutely necessary and so I dumoed them.
Speaking of that incident, I have lowered the iSCSI connection timeout 
to 60s, which seems to be the lowest value supported by issueing a

iscsiadm modify initiator-node -T conn-login-max=60

and afterwards I used stmfadm offline target on the storage node to cut 
the target off. This time, the initiator timed out after 60s and that 
particular zpool changed it's status to degraded without anything 
happening. I still have to test that under load, but I will probably 
push that to next weekend.

Thanks,
Stephan