users@glassfish.java.net

Re: Excessive logging from ShoalLogger up to 1GB per day

From: Joseph Fialli <joe.fialli_at_oracle.com>
Date: Thu, 19 May 2011 14:13:59 -0400

On 5/19/11 5:50 AM, forums_at_java.net wrote:
> Hi Joe,
>
> I've implemented the logging tweaks, however it only stopped after I
> restarted both clusters and the DAS, but this is probably a known issue.
>
> I'd like to send you some log files but the question is will it be
> helpfull
> as I can't seem to find the logfile containing the start of the problem,
> we've been deleting them from time to time... so only got about 2GB of
> logs
> of the past 2 days.
Please file an issue and attach the log files that you do have.
Please be sure to include the DAS server logs.

> I could re-enable logging but if this requires me to
> bring down the cluster it will take some time as we're not fond of
> bringing
> down a live system during business hours.
>
> Regarding your other suggestions, yes we are on Linux (ubuntu), but
> we're on
> beefy machines with 8 CPUs and a 24GB RAM, so the machines are hardly
> used
> for the moment, 0.40 loadavg.
>
> # cat /proc/sys/net/core/rmem_max
>
> 131071

The above value is too small for a server.

Given that you have 3 instances running on one machine, one being the DAS,
the above size is too small.

I would recommend the following values:

net.core.rmem_max=1024000
net.core.wmem_max=1024000
net.core.rmem_default=102400
net.core.wmem_default=102400

>
> # cat /proc/sys/net/core/rmem_default
>
> 124928
>
> Our UDP buffers might need tuning for heavyer loads, which I expect in
> the
> future with other apps deployed, but I would not expect the current
> load to
> require larger buffersizes, but I'm no expert in this area so I could be
> wrong here.
>
There is a flurry of UDP traffic when starting a cluster and stopping a
cluster, so the larger value
is recommended so the UDP loss during startup is not high. There is a
reliable rebroadcast for
dropped UDP messages that require delivery. Based on your reported
error messages, there
was a high level of drops, which I would expect given the small UDP
buffer size AND
having a DAS with 2 clusters and two instances all running on one machine.

Running the DAS on its own machine with a 500k UDP buffer would be
sufficient.
But to keep your current configuration, I would follow above recommendation.

-Joe Fialli

> Thanks for your assistance.
>
>
> --
>
> [Message sent by forum member 'jayv']
>
> View Post: http://forums.java.net/node/803337
>
>