[Pacemaker] iscsi target mounting readonly on client

Joseph-Andre Guaragna joseph-andre at rdmo.com
Sun Apr 14 14:32:36 UTC 2013


Thanks for the informations. I give it a try monday and gave you a feedback.



2013/4/12 Felix Zachlod <fz.lists at sis-gmbh.info>:
> Hello Joseph!
>
>> -----Ursprüngliche Nachricht-----,
>> Von: Joseph-Andre Guaragna [mailto:joseph-andre at rdmo.com]
>> Gesendet: Freitag, 12. April 2013 17:19
>> An: pacemaker at oss.clusterlabs.org
>> Betreff: [Pacemaker] iscsi target mounting readonly on client
>
> You have to make two things absolutely shure.
>
> 1. Data that has been acknowledged by you iscsi Target to your initiator has
> hit the device and not only the page cache!
>
> If you run your target in fileio mode you have to use write trough- cause
> with write back you or your cluster manager can't ever tell if the writes
> have completed before switching the DRBD states.
> That will only perform good if you have a decent raid card with BBWC! BUT
> YOU MUST RUN WRITE TRHOUGH or blockio (which will be write trough too)
> running write back in such a constellation IS NOT SAFE you might risk
> SERIOUS DATA CORRUPTION when switching targets.
>
> 2. On your initiator side try to rise the /sys/block/sd*/device/timeout
> value. That is the time the block device will wait for a command to complete
> before handing an i/o error tot he upper layer- which will most probably
> lead to your filesystem remounting r/o.
>
> 3. This is just a side note: do not use iet. We were running a production
> target wit iet for about 2 year which caused horrible problems to us.
> Consider scst or lio (I personally do not have any experiences with lio but
> scst is running in our production environment for years now without any
> problems)
>
> regards
> Felix
>
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org




More information about the Pacemaker mailing list