[CentOS] Possible to run iscsi-target and initiator on same server?

Sat Oct 10 01:42:17 UTC 2009
Ross Walker <rswwalker at gmail.com>

On Oct 9, 2009, at 6:14 PM, C Linus Hicks <linush at verizon.net> wrote:

>
> On Fri, 2009-10-09 at 17:53 -0400, Ross Walker wrote:
>> On Fri, Oct 9, 2009 at 5:37 PM, C Linus Hicks <linush at verizon.net>  
>> wrote:
>>> I am trying to install Oracle RAC in a two node cluster for testing
>>> purposes, so performance is not something that concerns me. I just  
>>> want
>>> to go through the process all the way to creating a database. I  
>>> have all
>>> the prerequisites except the shared storage and thought I'd give  
>>> this a
>>> try.
>>>
>>> I'm running:
>>> - CentOS 5.3 kernel 2.6.18-164.el5
>>> - iscsitarget-1.4.18-1
>>> - iscsi-initiator-utils-6.2.0.868-0.18.el5_3.1
>>>
>>> I can discover and login to the node using iscsiadm on the server  
>>> that
>>> is not also running iet, and see the new scsi disk.
>>>
>>> When I try to login using iscsiadm on the server that is running the
>>> ietd daemon, I see packets going over the lo interface in both
>>> directions, but I get error 8 - connection timed out. Wireshark is
>>> reporting TCP Previous segment lost errors with large sequence  
>>> numbers
>>> like 2999833 and 8999718.
>>>
>>> Has anybody tried this, and can anybody offer any help?
>>
>> You can only do that if you use blockio for your targets, otherwise
>> fileio will cause a deadlock on the page cache between the target's
>> usage of it and the initiator's usage of it.
>
> So for iet, that means /etc/iet/ietd.conf should define the Lun like
> this:
>
> Lun 0 Path=/dev/sdb,Type=blockio,ScsiId=asmdg,ScsiSN=dg0
>
> If that's all it takes to make it do blockio instead of fileio, then
> that is already the case. Deadlocks on the cache pages - wouldn't that
> only affect I/O and not logging in?

Ok, well at least that's covered.

Is iptables disabled?

Also the latest version of IET won't advertise the loopback on  
discovery unless you are discovering on that same same loopback  
address. You can still statically connect though, which I recommend  
for loopback connections.

-Ross