[ClusterLabs] SCSI persistant reservation fencing

Vladimir-M. Obelic v at obeli.ch
Thu Jul 16 11:19:41 EDT 2015


On 30 June 2015 at 17:48, Kristoffer Grönlund <kgronlund at suse.com> wrote:
> "Vladimir-M. Obelic" <v at obeli.ch> writes:
>
>> Hi,
>>
>> I'm trialling SLES 12 w/ HAE to run a Fileserver cluster consisting of
>> two nodes. Idea was to use SCSI persistant reservation as a fencing
>> method using the fence_scsi script from stonith fence agents. Two
>> nodes (a, b) are connected via FC to the same LUN which is then
>> exported via NFS from the active node only.
>>
>> According to fence_scsi resource info page, only the 'action'
>> parameter is obligatory, while the 'nodename' and 'key' parameters
>> aren't.
>> Yet without those, it fails. Seems to me this is a bug?
>> I've seen a similar issue in RHEL: https://access.redhat.com/solutions/1421063
>
> Yes, it looks like you are encountering this precise issue. Please file
> an issue with SUSE about this!
>

Already did, we'll see what happens.


>>
>> This is an example from RHEL that takes care of the whole thing, no
>> additional constraints (and it works!)
>>
>>  pcs stonith create my-scsi-shooter fence_scsi devices=/dev/sda meta
>> provides=unfencing
>>
>> (https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Configuring_the_Red_Hat_High_Availability_Add-On_with_Pacemaker/s1-unfence-HAAR.html)
>>
>> Note that SLES12 still uses crm while RHEL uses pcs. Also in SLES the
>> meta attribute 'provides' doesn't exist. Is there a way to translate
>> the RHEL pcs command to SLES?
>
> I would recommend against following guides for RHEL when configuring
> SLES. While the core software is the same, the versions are different,
> they are patched differently and the tools around pacemaker are
> different.
>
> The pcs command creates pretty much the same resource as the crm
> command, there are two problems however: The first is that you are
> encountering the above bug, and the second is that the "provides" meta
> attribute is not known by crmsh.
>
> The meta attribute can can still be created, however you will get a
> warning.
>
> // Kristoffer

Actually i've had a hint from a colleague who did a similar setup but on CentOS.

He also had the same issue and discovered by chance that there was a
change in fence_scsi script.
The one that was delivered with fence-agents-4.0.10 was written in
python and that one DIDN'T work with corosync-2.3.4 and
pacemaker-1.1.12.
He copied an older version (4.0.2) of fence_scsi written in perl (the
new one is a complete rewrite?!) and that one worked for him.
Although both of the scripts take the same parameters apparently the
newer script doesn't cope with corosync.
Does anyone actually use fence_scsi with recent corosync version or
can you tell me which corosync/pacemaker version works with the python
version of the script?

I mean I even tried on OpenSuse 13.2 (same issue...). This is
obviously an upstream issue.


>
>>
>>
>> Here the complete config:
>> crm config http://pastebin.com/mqxge6jm
>> corosync.conf http://pastebin.com/M5sr7htC
>>
>> corosync 2.3.3
>> pacemaker 1.1.12
>>
>> Any help appreciated!
>>
>> Regards,
>> Vladimir
>>
>> _______________________________________________
>> Users mailing list: Users at clusterlabs.org
>> http://clusterlabs.org/mailman/listinfo/users
>>
>> Project Home: http://www.clusterlabs.org
>> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>> Bugs: http://bugs.clusterlabs.org
>>
>
> --
> // Kristoffer Grönlund
> // kgronlund at suse.com




More information about the Users mailing list