jmeter-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Al_Sweetman <al.sweet...@gmail.com>
Subject RE: Thread won't die issue
Date Wed, 07 Feb 2007 13:22:40 GMT

AFAIK, Green threads have been replaced/superseded by native OS threads since
version 1.2...  And the native OS threading now used has solved the
reliability (well, some anyway) issues.

Now: We're experiencing the same problem here - our platform is as follows:
testing a JSP based app on a WAS v5 server linking to an AS/400 RPG back
end; with 9 remote agents for the stress tests (currently running around 10
threads/agent).

We've done some investigation and for us it turns out that the threads were
hanging in the appserver: like you this was only experienced when using a
thread count in the hundreds - which is awkward as that's only a thread
count of 11 if, like us, you have 9 remote agents.  Considering each box is
a dual 4ghz/5gb ram etc. that's quite annoying!  

We're going to add dealys/waits to 'ape' human interraction, anyway, so with
any luck this may solve the problem as they're going to be random timings -
will let you know.  Not sure if this [websphere etc] is another red-herring,
however.

Watch this space I guess?

Glad we're not alone ;-)

Cheers,

Al.
 

git wrote:
> 
> Daniel,
> 
> I would doubt that the thread.stop would cause any trouble anyway as it
> is caused by possible deadlocking between interdependent threads.  Back
> in the late nineties when I first started messing with Java Threads was
> about the same time that native threads came in.  Back the you could
> tell the JVM to use green threads (non native) or native.  Under very
> heavy thread load, green threads were more stable.
> 
> But - that was 8 years ago - and things have moved on.  I have no idea
> if you can even use green threads any more.
> 
> Cheers
> 
> AJ 
> 
> On Mon, 2007-01-29 at 12:17 -0600, Daniel Kurtz wrote:
> 
>> AJ;
>>  
>> Well, as I mentioned the best that we've been able to work out so far is
>> to just limit the number of threads you run on one agent. Our target is
>> ultimately 500 users, and we're just going use 5 100-user agents to get
>> there, for this round at least.
>>  
>> I'm not a Java developer, but I had a look at the source to see what's
>> going on when the "thread won't die" is being logged and found the
>> thread.stop. I thought I might have stumbled on something when I read in
>> the Java reference that thread.stop is considered unsafe and shouldn't be
>> used any more. But in looking further I've realized that this isn't where
>> the failure's occurring. These threads are already toast at this point.
>> They're so toasty the thread.stop can't even kill them, which is why
>> JMeter throws the message.
>>  
>> Daniel Kurtz 
>> 
>> 
>> ________________________________
>> 
>> From: git [mailto:git@cubicalland.com]
>> Sent: Mon 1/29/2007 10:25 AM
>> To: JMeter Users List
>> Subject: RE: Thread won't die issue
>> 
>> 
>> 
>> Daniel,
>> 
>> I've seen this - but have no idea what causes it or what to do about it!
>> 
>> You're not being ignored...
>> 
>> Cheers
>> 
>> AJ
>> 
>> On Mon, 2007-01-29 at 09:40 -0600, Daniel Kurtz wrote:
>> 
>> > Well, since no one responded to this I guess that no one else is seeing
>> it. What we're seeing here is that the threads are actually dying in some
>> way. They stop returning samples and for all practical intents and
>> purposes are 'hung' after that point. When we issue the 'stop' from the
>> console, JMeter sees all these threads still running and issues a
>> thread.stop command for each. It comes back a little later to each thread
>> to see if it has, indeed, stopped. It hasn't, so JMeter returns the
>> "thread won't die" message.
>> > 
>> > What we've found is that, on any of the hardware we've tried it on
>> (which represents a pretty broad range of capabilities) this appears to
>> be related to an upper threshold on the number of threads that can be run
>> per agent. I place the figure at about 150, though for safety's sake
>> we're recommending to our users that they try not to go above 100 threads
>> per agent. That includes the console itself when running local tests.
>> > 
>> > Daniel Kurtz
>> >
>> >
>> > ________________________________
>> >
>> > From: Daniel Kurtz [mailto:dkurtz@olenick.com]
>> > Sent: Mon 1/22/2007 11:10 AM
>> > To: jmeter-user@jakarta.apache.org
>> > Subject: Thread won't die issue
>> >
>> >
>> >
>> > We find that invariably we end up with more or less a number of threads
>> that don't go into an 'ending' status. Of course the test itself does not
>> end in these circumstances, but will just hang out there forever. When we
>> force a stop from the console, the end of jmeter.log ends up looking
>> like:
>> >
>> > ...
>> > 2007/01/22 10:52:44 INFO - jmeter.threads.JMeterThread: Thread Thread
>> Group 1-500 is done
>> >
>> > 2007/01/22 10:52:44 INFO - jmeter.engine.StandardJMeterEngine: Ending
>> thread 499
>> >
>> > 2007/01/22 10:59:10 INFO - jmeter.threads.JMeterThread: Stopping Thread
>> Group 1-271
>> >
>> > 2007/01/22 10:59:15 INFO - jmeter.engine.StandardJMeterEngine: Thread
>> won't die: Thread Group 1-271
>> >
>> > 2007/01/22 10:59:15 INFO - jmeter.engine.StandardJMeterEngine:
>> Notifying test listeners of end of test
>> >
>> > 2007/01/22 10:59:15 INFO - jmeter.gui.util.JMeterMenuBar:
>> setRunning(false,local)
>> >
>> > 2007/01/22 10:59:15 INFO - jmeter.engine.StandardJMeterEngine: Test has
>> ended
>> >
>> > Suggestions as to what we should look at to resolve this?
>> >
>> > Thanks.
>> >
>> > Daniel Kurtz
>> >
>> >
>> >
>> >
>> >
>> > ---------------------------------------------------------------------
>> > To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
>> > For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
>> 
>> www.cubicalland.com
>> www.nerds-central.blogspot.com
>> 
>> 
>> 
>> 
>> 
>> 
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
>> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
> 
> www.cubicalland.com
> www.nerds-central.blogspot.com
> 
> 
> 
> 
> 

-- 
View this message in context: http://www.nabble.com/Thread-won%27t-die-issue-tf3059035.html#a8845300
Sent from the JMeter - User mailing list archive at Nabble.com.


---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org


Mime
View raw message