[Gluster-devel] RPC, DHT and AFR logging errors that are expected, reduce log level?

[Pranith Kumar Karampuri]

On Thu, Sep 29, 2016 at 7:51 PM, Niels de Vos <ndevos at redhat.com> wrote:

> Hello,
>
> When NFS-Ganesha does an UNLINK of a filename on an inode, it does a
> follow-up check to see if the inode has been deleted or if there are
> still other filenames linked (like hardlinks) to it.
>
> Users are getting confused about the errors that are logged by RPC, DHT
> and AFR. The file is missing (which is often perfectly expected from a
> NFS-Ganesha point of view) and this causes a flood of messages.
>
> From https://bugzilla.redhat.com/show_bug.cgi?id=1328581#c5 :
>
> > If we reduce the log level for
> > client-rpc-fops.c:2974:client3_3_lookup_cbk there would be the
> > following entries left:
> >
> > 2x dht-helper.c:1179:dht_migration_complete_check_task
> > 2x afr-read-txn.c:250:afr_read_txn
> >
> > it would reduce the logging for this non-error with 10 out of 14
> > messages. We need to know from the AFR and DHT team if these messages
> > are sufficient for them to identify potential issues.
>

Updated the bug from the perspective of AFR as well:
https://bugzilla.redhat.com/show_bug.cgi?id=1328581#c12

"I am not sure how an inode which is not in split-brain is linked as 'no
read-subvolumes' case. That is something to debug."


>
> Thanks,
> Niels
>



-- 
Pranith
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-devel/attachments/20160930/083cf280/attachment.html>