HORNETQ-469 is making the server leak

[clebert.suconic]

I have played with and without the changes for 469, and the server is running out of memory after 1 hour running.

Also, the testsuite seems unstable after 469. I could get a clear run fairly easily on my desktop on Branch_2_1 (which is basically trunk - HORNETQ-469)

[timfox]

If you have any more info I'll take a look tomorrow.

[clebert.suconic]

I didn't have time to look at why...

All I have done is sending and consuming at a sustained rate for more than 1 hour. I have been using Ronnys's tests (http://community.jboss.org/message/558905?tstart=0#558905) without the restarts part, but our performance tests should also fail.

You could add -verbose:gc to your run.sh and you will see that the memory will start growing at every FULL-GC. up to the point you get an OME.

[clebert.suconic]

Just to make sure it's clear.. I already determine that the server will run fine without any issues when I reverted HORNETQ-469

I just didn't have time to find where's the leak (well it would be fixed if I knew, right?)

[clebert.suconic]

Tim, did you do any investigation on this? I'm just wondering if you have any idea on what could be the cause.

I will take over this if you don't plan on doing it any time soon.

[timfox]

I've been on holiday for the past couple of weeks.

I'll take a look when I return on Monday

[clebert.suconic]

I knew that and I didn't mean to "patronize" you. (I'm not finding a better english term now)

I just wanted to confirm if you were expecting me to work on it or if you would handle it.

I have another task to complete anyway.. so I will let this one for you :-)

[timfox]

Is there a simple test I can run to replicate this?

I'd rather not have to run this complex load test program.

[timfox]

I'm going to wait for someone to provide a simple test case to replicate this before I look at it

[clebert.suconic]

you just need to run our performance tests as I said. Have it running for some time with 1k message body sizes.

[timfox]

Which performance tests? Can you be more specific?

Do you mean the example you added? This was added in 2_1 but not in TRUNK.

[clebert.suconic]

>> "Which performance tests? Can you be more specific?"

No, I meant the performance tests we used on our recent benchmark. I don't know any other performance tests.

>> Do you mean the example you added? This was added in 2_1 but not in TRUNK.

I'm not sure what you mean:

http://anonsvn.jboss.org/repos/hornetq/trunk/examples/soak/tx-restarts/

[timfox]

Clebert Suconic wrote:

>> "Which performance tests? Can you be more specific?"

No, I meant the performance tests we used on our recent benchmark. I don't know any other performance tests.

Ah, the Sonic Test Harness tests? Is there any specific one I should run to replicate the issue?

[clebert.suconic]

It should fail with anyone as long as you send 1K messages or something bigger (< minLargeMessageSize)

[timfox]

OK, you lost me here. I just need to send messages? What about consuming? If I just send messges it's going to run out of memory anyway.

If you can please be more specific on how to replicate this it would help a lot...

Right now it's all a bit vague and hand-wavy