users@glassfish.java.net

Re: Server stops responding due to Glassfish

From: Jeanfrancois Arcand <Jeanfrancois.Arcand_at_Sun.COM>
Date: Wed, 30 Apr 2008 10:34:09 -0400

Ryan de Laplante wrote:
>> After a few hours HTTP requests consistently stop reaching the web
>> application. The java application continues to run, but HTTP requests
>> are not being passed to servlets. If I go into the admin console and
>> fiddle with an HTTP Service setting, like enabling access logging for
>> example, POOF!! the site comes back to life. I can only assume that
>> doing that causes something to reset in Glassfish. Clearly this is a
>> Glassfish issue and nothing to do with the application code. Maddening
>> and unacceptable. My only option is to get off Glassfish and onto
>> something more stable. I am stopping/starting the domain on my
>> production servers constantly. If anyone has a suggestion to make this
>> problem go away I will be very grateful.
>> How quickly do the waters become muddy. This thread now has multiple
>> issues posted to it, which renders it pretty much useless to anyone
>> trying to actually track down and fix a bug in Glassfish.
> I would disagree. We are having the same exact issue, except I've been
> having it for 6 months in production and it has several symptoms:
> 1) after some period of time, users get a blank screen forever waiting
> for a response. We have to restart the Windows service to fix it.
> 2) With the same frequency, sometimes users get a "Maximum connections
> reached (4096)" error immediately. We have to restart the Windows
> service to fix it.

This is expected to see such exception, and this confirm that all
threads are blocked somewhere.

Unfortunalty, the thread dump you sent contains nothing about grizzly
threads. You need to increase the win console length to 99999 by
clicking properties on a cmd windows (ouf)...Can you try it?

Thanks

-- Jeanfrancois



> 3) Same symptom as #1, but restarting app server doesn't help and
> Windows has to be rebooted. This is the NP Pool bug in Windows 2003
> server, and it has only happened a few times.
>
> Through these emails, finally after 6 months someone has told me that
> the first two issues are NOT caused by the NP Pool problem in Windows
> 2003 Server. Sun is giving me other ideas now, such as disabling
> Grizzly and using Coyote with this JVM option:
>
> -Dcom.sun.enterprise.web.useCoyoteConnector=true
>
> These discussions used to happen off the mailing list because I was
> communicating with Sun through our support contract. Now the world gets
> to see these discussions. Nobody knows the answer to this problem yet,
> so these discussions are the best thing you've got. Not even a Sun
> support contract will help your resolve this issue.
>
> I don't use a database in the web tier either, that happens in the EJB
> tier. I'm not entirely convinced the problem is caused by database
> connections or my application. An older version of the application used
> to run on JBoss, same server, for over a year without these issues.
>
> Today the system went down again (9 hours since last time) and I
> remembered to try changing the HTTP Listener to see if it would come
> back to life like you suggested (new idea to me). I first disabled it,
> save, enable it, save. My app came back to life! I have a second HTTP
> listener that uses SSL for web services, and disabling/re-enabling made
> it come back with a response, but an HTTP 500 response. So I restarted
> the app server.
>
> This morning I've told the app server to use Coyote and we'll see how it
> goes. I also captured what I can of a thread dump (999 lines max in
> Windows console).
>
>
> Thanks,
> Ryan
>
>
>
>
> glassfish_at_javadesktop.org wrote:
>> The original issue that this thread was about has nothing whatever to
>> do with database connections or deadlocked threads. My application
>> doesn't even hit a database. The symptoms, and hints as to the source
>> of the problem are to be found in my original post at the top of this
>> thread. I wish the Glassfish team luck fixing that issue, because I
>> very much believe it to be a bug in Glassfish, unlike deadlocked
>> threads which is more likely to be application code.
>> [Message sent by forum member 'kevinmacdonald' (kevinmacdonald)]
>>
>> http://forums.java.net/jive/thread.jspa?messageID=272165
>>
>> ---------------------------------------------------------------------
>> 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