Re: postgresql is unable to create a table in gluster volume

Vijay Bellur <vbellur@xxxxxxxxxx> · Thu, 4 May 2017 22:44:24 -0400

On Thu, May 4, 2017 at 3:19 AM, Jiffin Tony Thottan <jthottan@xxxxxxxxxx> wrote:

    On 04/05/17 02:03, Praveen George
      wrote:

              Hi
                Team,

              We’ve
                been intermittently seeing issues where postgresql is
                unable to create a table, or some info is missing.

              Postgresql
                logs the following error:

              ERROR:
                 unexpected data beyond EOF in block 53 of relation
                base/16384/12009
              HINT:
                 This has been seen to occur with buggy kernels;
                consider updating your system.

              We
                are using the k8s PV/PVC to bind the volumes to the
                containers and using the gluster plugin to mount the
                volumes on the worker nodes and take it into the
                containers.

              The
                issue occurs regardless of whether the  k8s spec
                specifies mounting of the pv using the pv provider or
                mount the gluster volume directly.

              Just
                to check if the issue is with the glusterfs client, we
                mount the volume using NFS (NFS on the client talking to
                gluster on the master), the issue doesn’t occur.
                However, with the NFS client talking directly to _one_
                of the gluster masters; this means that if that master
                fails, it will not failover to the other gluster master
                - we thus lose gluster HA if we go this route. 

    If you are interested there are HA solutions available with NFS. It
    depends on NFS solution which u are trying, if it is gluster
    nfs(integrated nfs server with gluster) then use ctdb and for
    NFS-Ganesha , we already have an integrated solution with
    pacemaker/corosync

    Please update ur gluster version since it EOLed, you don't receive
    any more update for that version.

Do you notice any errors in the fuse client logs when postgresql complains about the error? 

It might be useful to turn off all the performance translators in gluster and check if the problem persists.

Regards,
Vijay
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
http://lists.gluster.org/mailman/listinfo/gluster-users