Re: [Scst-devel] Active/Non-Optimized ALUA error, lost target port group id on Non-Optimized node disk

[Vladislav B. <vs...@vl...>]

Vladislav Bolkhovitin wrote on 08/18/2017 09:20 PM:
> Hi,
> 
> I'm afraid, I can hardly understand your problem. All you need in this mode are:
> 
> 1. Create a forwarding target for each your real ("optimized") target on the active
> side that is exporting the same SCST device and accepting redirected commands from the
> non-optimized passive node. This target is NOT visible to initiators, hence its ALUA
> state doesn't matter as soon as it is active.
> 
> 2. Connect to this target from the passive node and use imported from it SCST device to
> redirect all commands coming to the corresponding initiator-visible non-optimized
> target. Use scst_disk for that device with CONFIG_SCST_FORWARD_MODE_PASS_THROUGH enabled.

I accidentally came across my scripts I was using during testing this functionality, so
I added them in the README as examples. Everybody welcome to extend scstadmin of this
new functionality.

Vlad

> ?波 史 wrote on 08/16/2017 12:48 AM:
>> <https://outlook.live.com/owa/?path=/mail/AQMkADAwATNiZmYAZC1iZjJkLTEyZjItMDACLTAwCgAuAAAD19NTu8xoHkmNVIek5Mln7QEAMi7U5dfozk6Su7bNKKRg5wAAAMGV6PsAAAA%3D&realm=live.com>
>>
>> hello:
>>
>> For alua(Active - NonOptimized) setup with latest scst code,  we config it with the
>> mothod you pushed on Jun 7(commit msg: scst: update ALUA documentation
>> (Active/Non-Optimized)), and map device from active node to nonoptimized node through
>> iscsi without target group, but dont get expected result. On nonoptimized node, the
>> mapped disks dont have target port group id info (get scsi inquiry 0x83 vpd via sg_inq
>> cmd),so the host multipath cannot get the right NonOptimized disk info in pass-through
>> mode(use scst_disk handler, enable target forwarding and
>> CONFIG_SCST_FORWARD_MODE_PASS_THROUGH).
>>
>> This problem is caused by the internal device map from active node to NonOptimized
>> node. We have two methods for this(not tried):
>> 1.  map the device from active to standby node through the target group for host, the
>> mapped disks should have right target port group info on nonoptimized node. This method
>> need to create a lot of different internal iscsi targets equal to the number of target
>> groups, and may map one device to NonOptimized several times.
>> 2.  modify the scst_disk handler code, get the target port group info locally and
>> insert it to inquiry cmd(pass-through) response data, as the both two nodes have all
>> same target groups.
>>
>> Is there any other better way for this?
>> Thanks very much!
>>
>> our scst configuration is like this：
>>
>> node1 device group：
>>     scst device: disk1 (scst_vdisk)
>>          target group: tg1, active
>>             fc target: fc1     (local, lun1)
>>         target group: tg2, nonoptimized
>>            fc target: fc2    (peer, not real)
>>
>> node2 device group:
>>     scst device: H:C:I:L (scst_disk)
>>         target group: tg1, active
>>             fc target: fc1    (peer, not real)
>>         target group: tg2, nonoptimized
>>             fc target: fc2     (local, lun1)
>>
>> note: H:C:I:L is mapped from node1 disk1 via internal iscsi target.