users@glassfish.java.net

Re: shoal / GMS sniffer?

From: Dick Davies <rasputnik_at_hellooperator.net>
Date: Fri, 21 Nov 2008 17:02:01 +0000

On Thu, Nov 20, 2008 at 3:46 PM, Shreedhar Ganapathy
<Shreedhar.Ganapathy_at_sun.com> wrote:
>
>
> Dick Davies wrote:
>>
>> I've been sat down crosseyed in front of snoop and tcpdump lately seeing
>> what our cluster is up to, and wondered if there were any higher level
>> tools
>> for looking at cluster GMS traffic?
>>
>
> What specifically are you looking for ?
> You can set ShoalLogger (as a module-log-levels property in domain.xml) to
> FINE and see more data. It gets a lot more detailed and a bit noisy with
> debug information as you go down FINER and FINEST.

I can see multicast traffic (? it's addressed to the multicast address
I've configured, anyway)
 that seems to be related to cluster events (instances stopping/starting)
but nothing that looks like a session being replicated (or copied
over) - even when I
tweak the load balancer to send a client to one node, have them log in
and then point the
LB to the other node.

Clients have to login again when they hit the other node, so something
is not right I think, but
I can't understand why some traffic seems to make it across and other
traffic doesn't.

Do both sessions and cluster events use GMS? Are there any differences?
(URLs are fine at this point :) - I've read
https://shoal.dev.java.net/ShoalDesignDocument.html
which was quite useful but not GF specific).

Will have a look at btrace, thanks.

Does anyone know of any introductory reference material on writing a
GMS client?
Thanks again.

>> From a quick glance over the shoal design docs, it looks like it would
>> be possible to add a SPECTATOR member to the cluster who could receive
>> messages
>> (and log/graph them etc.) without disrupting cluster operation or
>> requiring a lot of debug-level
>> logging on the cluster instances themselves.
>>
>
> Very good thought. A Spectator member collecting monitoring information
> about cluster members would be a very useful way to present information on
> cluster messaging and collecting historical information on cluster lifecycle
> events.
>>
>> Does anybody know of such a tool, or where I could look to get started
>> writing my own?
>>
>
> btrace comes to mind. http://btrace.dev.java.net
> We are beginning to use it for debug purposes in our load tests and I am
> hoping that will lead us to write a utility on the above lines.
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: users-unsubscribe_at_glassfish.dev.java.net
>> For additional commands, e-mail: users-help_at_glassfish.dev.java.net
>>
>>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe_at_glassfish.dev.java.net
> For additional commands, e-mail: users-help_at_glassfish.dev.java.net
>
>