[Gluster-users] 120k context switches on GlsuterFS nodes

[mabi]

I have been using GlusterFS now for over a year in production and what made me write this mail initially is that it is the first time I see such high peeks in terms of context switches/interrupts but else to reply to your question everything seems to work just fine so far.

What has changed from my side is that I was using 3.7 until end of April and now I am on the latest 3.8.11 version of GlusterFS. So as you mention it could be something related to 3.8.

-------- Original Message --------
Subject: Re: [Gluster-users] 120k context switches on GlsuterFS nodes
Local Time: May 17, 2017 7:49 PM
UTC Time: May 17, 2017 5:49 PM
From: jlawrence at squaretrade.com
To: mabi <mabi at protonmail.ch>, Gluster Users <gluster-users at gluster.org>

> On May 17, 2017, at 10:20 AM, mabi <mabi at protonmail.ch> wrote:
>
> I don't know exactly what kind of context-switches it was but what I know is that it is the "cs" number under "system" when you run vmstat.
>
> Also I use the percona linux monitoring template for cacti (https://www.percona.com/doc/percona-monitoring-plugins/LATEST/cacti/linux-templates.html) which monitors context switches too. If that's of any use interrupts where also quite high during that time with peaks up to 50k interrupts.

You can't read or write data from the disk or send data over the network from userspace without making system calls. System calls mean context switches. So you should expect to see the CS number scale with load - the whole point of Gluster is to read and write and send data over the network.

As far as them being "excessive", I don't know how to think about that without at least a comparison , or better, some evidence that something is doing more work than it "should". (Or best, line numbers where unnecessary work is being performed.)

Is there something other than a surprising number to make you think it isn't behaving well? Did the number jump after an upgrade? Do you have other systems doing roughly the same thing with other software that performs better? Keep in mind that, say, a vanilla NFS or SMB server doesn't have the inter-gluster-node overhead, and how much of that traffic there is depends on how you've configured Gluster.

-j
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170517/39be281f/attachment.html>