Gluster users - May 2014 - Repeated file stats to non-existent files causing gluster to consume excessive bandwidth.

Hello,

I am currently using gluster replication to share a common filesystem
across multiple nodes. Part of my application is a directory watcher that
will end up doing a stat on a glob of file patterns. What I am seeing is
that gluster will start using roughly 300kBps per watched pattern. Will
gluster ever back off, or continue to try and force consistency even though
there are no files?

Here's a small bit of bash that will repro this:

for i in {1..1000}; do stat /path/to/files/that/do/not/*.exist; sleep 2;
done


Thanks!

-Max Murphy
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20140506/d30f80d2/attachment.html>

Sorry for the lack of platform and version details in my previous message.

Amazon Linux 2013.09 which is a centos 6.4 derivative
Gluster 3.4 and 3.5

Share is mounted over the fuse client.

Thanks,
Max


On Tue, May 6, 2014 at 5:08 PM, max murphy <max.murphy at gmail.com>
wrote:
> Hello,
>
> I am currently using gluster replication to share a common filesystem
> across multiple nodes. Part of my application is a directory watcher that
> will end up doing a stat on a glob of file patterns. What I am seeing is
> that gluster will start using roughly 300kBps per watched pattern. Will
> gluster ever back off, or continue to try and force consistency even though
> there are no files?
>
> Here's a small bit of bash that will repro this:
>
> for i in {1..1000}; do stat /path/to/files/that/do/not/*.exist; sleep 2;
> done
>
>
> Thanks!
>
> -Max Murphy
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20140507/0f9d09da/attachment.html>