[Gluster-users] glusterfsd process thrashing CPU

[Lindsay Mathieson]

On Tue, 18 Nov 2014 02:36:19 PM Pranith Kumar Karampuri wrote:
> On 11/18/2014 01:17 PM, Lindsay Mathieson wrote:
> > On 18 November 2014 17:40, Pranith Kumar Karampuri <pkarampu at redhat.com> 
wrote:
> > 
> > However given the files are tens of GB in size, won't it thrash my
> > network?
> 
> Yes you are right. I wonder why thrashing of the network is never
> reported till now.

Not sure if you are being sarcastic or not :) But from what I've observed, 
sync operations seem to self throttle, I've not seen them use more than 50% of 
bandwidth, and given most setups have a dedicated network for the servers 
maybe they just don't notice if it takes a while?

> I still need to think about how best to solve this problem.

Setup a array of queues for self healing, sorted by size maybe?

> 
> Let me tell you a bit more about this issue:
> there are two processes which heal the VM images:
> 1) self-heal-daemon. 2) Mount process.
> Self-heal daemon heals one VM image at a time. But mount process
> triggers self-heals for all the opened files(VM image is nothing but an
> opened file from filesystem's perspective) when a brick goes down and
> comes backup.


Thanks, interesting to know.

> So we need to come up with a scheme to throttle self-heals
> on the mount point to prevent this issue. I will update you as soon as I
> come up with a fix. This should not be hard to do. Need some time to
> choose the best approach. Thanks a lot for bringing up this issue.

Thanks you for looking at it!

Cheers,


-- 
Lindsay
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part.
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20141118/8f0629ab/attachment.sig>