On 10/19/2012 7:22 PM, [email protected] wrote:
note that setting maxclients lower isn't really a good option. It
means that when you hit maxclients, users trying to get to your site
get an error page.
Not in this case. We tested this, and when maxclients was hit, and the
server wasn't accepting any more connections, the load balancer just
sent the traffic elsewhere.
We are stuck with a really stupid load balancer right now (in process of
upgrading), which only has least-connection, round-robin or weighted
round-robin. Due to historical reasons, we are using the
least-connection method, which has no way to do any weighting.
JBB
you really should adjust your load balancer to send less traffic to
that server.
David Lang
On Fri, 19 Oct 2012, Jonathan Bayer wrote:
Date: Fri, 19 Oct 2012 14:24:52 -0400
From: Jonathan Bayer <[email protected]>
To: Aleksey Tsalolikhin <[email protected]>
Cc: [email protected]
Subject: Re: [lopsa-discuss] Linux problem, need some help
Hi all,
Just to update you all on this:
1. The host was running several other disk-intensive VMs. We shut
them down.
2. I reduced the MaxClients to a much lower value.
3. I also set the swappiness=0 on both the host and the vm
We ran some load tests on it, both before and after these changes to
confirm that they were helping.
What apparently happened was that MaxClients was too high. The
system got hit with a very large load, and it hit maxclients.
Swapping may also have been going on.
It has been in production since Wednesday and has performed very nicely.
Thanks again to all of you who responded.
JBB
On 10/18/2012 8:53 PM, Aleksey Tsalolikhin wrote:
On Mon, Oct 15, 2012 at 8:01 AM, Jonathan Bayer
<[email protected]> wrote:
The problem is that when the req/sec starts to climb on the VM, all
of a
sudden the load average skyrockets to 40-50 or even higher. This
kills the
server and makes it totally unresponsive.
Anybody have any suggestions?
What does "vmstat 1" show when the load starts to climb? I suspect
you are swapping,
in which case your "si" and "so" (swap in and swap out) columns will
be non-zero.
If so, the solution would be to increase RAM per VM instance, or add
more instances to
distribute the work more. (To shrink each VM's share of the workload.)
Best,
-at
_______________________________________________
Discuss mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/discuss
This list provided by the League of Professional System Administrators
http://lopsa.org/
_______________________________________________
Discuss mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/discuss
This list provided by the League of Professional System Administrators
http://lopsa.org/
