silverstream server hangs

[ihbrune]

Hello,

since more then half a year we try to find a nasty bug in our silverstream server application.

the bug causes the server to "hang", that means the server process is still running (but with 0.X CPU load), users can connect with their browsers to the server, but no content is given back anymore.

has anyone made similar experiences ? Did anyone found a solution ?

regards,
Henning

 

[stevepalmer]

I am working with a situation that sounds similar. Did you get to solution? Help.

 

[ihbrune]

Not yet. Still do load tests on our 3.7.4 server. I was able to reproduce the behaviour on a "bare bones" system without any deployed applications. But silverstream support does not believe me.

Will try to move to version 4 in februrary and try if this behaves better.....

 

[stevepalmer]

ihbrune,,
Thank you for responding so soon.
I think my team is close to concluding that root cause is a silverstram limit on inbound connections of around 200, plus or minus some factor due to other tuning. I see that 200 limit (per instance of SS) mentioned in a couple other forums. And in those notes, it appears to not matter what config (H/W, monitoring software, etc.) is being used to run the app server
We're testing all weekend to nail this and deal with this limit, if it is reality.
Might you be experiencing what we're seeing?

 

[jwenting]

We're now encountering much the same.
Strangely the server ran for months without any problems at all, but the day we take the site life on the net the server hangs solidly every hour or so requiring a restart.
We're running 3.7.4 (inherited from another project we originated from).

nothing is logged, first sign is that browsers are not getting returned content from the server.

 

[ihbrune]

Hello jwenting,

we found a solution for our problem: the hanging server is caused by a bug in silverstreams database connection pool: When the pool has not yet allocated all possible connections and there are so many parallel requests that the server tries to allocate all missing connetions at once some kind of deadlock occurs.

With the help from SilverStream consultants we fixed this with a small business object that is open all db connections on server startup. Since then we did not encounter this problem any more.

 

[abc1260]

ihburne

What is this small business object you used. I am getting a pause on my server also. It does not hang entirely, but pauses after several user requests.

http://www.aikiin.com/pchauvet

 

[ihbrune]

we use a server start triggered business object in every application database in our server (not in the silvermaster). This is the code:

public void serverStarted(AgoServerStartEvent evt)
{
funktion(evt.getDatabase()) ;
}

private void funktion(AgiDatabase db) {
AmbryWebbase webBase = (AgWebbaseMgr.getWebbaseMgr()).getWebbase(db.getName());
// AmbryWebbase webBase = (AgWebbaseMgr.getWebbaseMgr()).getWebbase("Test"

// webBase.dumpConnectionPools(new PrintStream(System.out));


AgiConnectionSet connSet = webBase.getConnectionSet();

if( connSet instanceof AgDataSource )
{
int conncnt = 0;
conncnt = connSet.getConnectionCount() + connSet.getUnallocatedConnectionCount();

System.out.println("setting min connections for db "+ db.getName()+" to " + conncnt);
((AgDataSource) connSet).setMinConnCount(conncnt);
}

// webBase.dumpConnectionPools(new PrintStream(System.out));
return;
}

 

[jwenting]

We've now solved our problem. Was due to a webapp (WAR) having sessions set to 300 minutes in the deployment descriptor (which value according to the documentation we have is calculated in seconds, so it should have been 300 seconds) leading to the server running out of memory for opening sessions.

<warJarOptions>
	<warJar>
		<warJarName>b123admin.war</warJarName>
		<isEnabled>true</isEnabled>
		<sessionTimeout type=&quot;String&quot;>5</sessionTimeout>

sessionTimeout according to SilverStream's own J2EE courseware is measured in seconds, but the 3.7.4 at least interprets it as minutes