Memory leak

[Volker Halle]

Wim,

o.k. - not a memory leak in pthreads.

You could use SDA> SHOW PROC/PHD to look at which VA address space is increasing (P0 or P1). Then use SDA to just dump the most recent ;1000 bytes and look for patterns.

Volker.

[Wim Van den Wyngaert]

I did something like that yesterday. Found some patern in it and gave the conclusions to the developper.

Who will probably do nothing with it because we only have 2% of 2006 activity left and his bos says it will be finished by year end.

Wim

[GuentherF]

Sorry, it's me slow again.

Consuming what type of memory: physical or virtual?

What is the error message/status when the application/process fails?

What about this "process tree"?

If it is virtual memory - and the error message would tell - then a purge workingset does nothing here. If processes run in a a job tree then they have to share certain quotas like PGFLQUOTA comes to mind. Same amount of processes per job?

I may have missed some of this information in an earlier reply.

/Guenther

[Wim Van den Wyngaert]

Gunther,

Consuming what type of memory: physical or virtual?
- Virtual

What is the error message/status when the application/process fails?
- none. Hang of the proces

What about this "process tree"?
- a master process creationg about 10 sons. Normally memory consumtion is stable at 43% of 3Mpages.

If it is virtual memory - and the error message would tell - then a purge workingset does nothing here. If processes run in a a job tree then they have to share certain quotas like PGFLQUOTA comes to mind. Same amount of processes per job?
- yes I thought that too. It's PGFLQ that is the limiting factor. Increasing the PGFLQ is no solution because it's gone within minutes
(found interprocess traffic of about 10 GB per hour, it's about at this speed that memory is going.

I may have missed some of this information in an earlier reply.

- no problem.

Yesterday I moved the priority from 6 to 15 and didn't have the problem. But may be the markets are just quieter.

Wim
(BTW : Jan : the transactions are no longer done on VMS. Only a few M bucks left)

[GuentherF]

Wim,

that sounds like the application tries something smart - handling the exceeded pagefile quota by wait-and-retry.

So you increased the accounts PGFLQUOTA and the size/number of pagefiles?

There is a good chance that the application is doing expand P0 region and leaving unused virtual memory blocks behind. Maybe it is just a couple of virtual memory pages away from being happy? I would like to see the processes exhaust P0 space.

Time to let someone who knows the code look at it (open files, channels in use, application logs, exam to VA memory).

/Guenther

[Wim Van den Wyngaert]

It seems that nobody can conclude something based upon the cpu sampling.

As 750 MB is taken in 2 minutes, giving PGFLQ will only extend life of the process a little. So we restart it when it reaches 1 GB. And hope the exchanges will calm again (better for my money too).

Wim