Linux Open-iSCSI Initiator Development Mailing List ()

liquidhorse | 8 Jun 2012 21:19

iscsistart reconnect to same target, new host?

Hi!

I am attempting to have a diskless iscsi-based server stay alive
during a target host failover.  Basically, my situation is as follows:
  * The iscsi target is on a 2-node HA cluster.
  * I can boot the initiator against either machine in the node.
  * I can disconnect and reconnect the network cable on the initiator
node, and it resumes communication with its chosen node.
  * Migrating the iscsi target from one node to the other causes the
connection to completely die.

I have configured other iscsi targets and initiators that are able to
reconnect in the event of a resource migration or target-node
failover.  However, I suspect there is something to the fact that my
diskless server is using iscsistart, and not iscsid, to make the
connection.  Obviously, starting iscsid after boot brings terrible
evil to my server, causing it to go insane and die for lack of file
system access.

I am using Ubuntu 11.10 Server; the iscsi-install went beautifully,
the machine can boot, run, and reboot without issue.  Only when the
target is failed-over from one backing node to the other do I see this
failure occur.  (I have the IP address migrating with the target, and
verified that everything works correctly for at least booting - as I
said, I've done this before with non-boot iscsi initiators, and it
works wonderfully there.)

Any ideas on what I could/should be doing to make this work correctly?
Thanks!
-- Matthew

(Continue reading)

headers

Mike Christie | 9 Jun 2012 05:28

Re: iscsistart reconnect to same target, new host?

On 06/08/2012 02:19 PM, liquidhorse wrote:
> Hi!
> 
> I am attempting to have a diskless iscsi-based server stay alive
> during a target host failover.  Basically, my situation is as follows:
>   * The iscsi target is on a 2-node HA cluster.
>   * I can boot the initiator against either machine in the node.
>   * I can disconnect and reconnect the network cable on the initiator
> node, and it resumes communication with its chosen node.
>   * Migrating the iscsi target from one node to the other causes the
> connection to completely die.
> 
> I have configured other iscsi targets and initiators that are able to
> reconnect in the event of a resource migration or target-node
> failover.  However, I suspect there is something to the fact that my
> diskless server is using iscsistart, and not iscsid, to make the
> connection.  Obviously, starting iscsid after boot brings terrible
> evil to my server, causing it to go insane and die for lack of file
> system access.

You must have iscsid running when there are network/connection/iscsi
errors. iscsid is the component that will reconnect to the target when
the problem is resolved.

Running iscsid after iscsistart should work ok. When iscsid starts it
will force the existing sessions to be dropped and to reconnect so we
can sync up on the state. When iscsid starts like this it will check the
FS for some files, but it should be a FS that is up and mounted and
running. iscsid is not going to try and disconnect a target that we are
then going to try and access.

(Continue reading)