Restart workers after certain number of events processed

ROOT
1

Hi,

Is it possible to limit the number of events processed by each worker instance? It’s common and powerful workaround for memory leak issues and it’s widely used in, for instance, web servers.

–
Thanks,
Evgeny Boger

2

Hi,

Limiting the number of events processed by a worker goes against the PROOF paradigm.
If you have problems with memory you can try to set upper limits on the memory usage.
See root.cern.ch/drupal/content/limi … nd-workers .

The best would be of course to find the leaks.

G. Ganis

3

Thank you, this feature will be useful in our case.

Just to clarify: when using that limits, if the program has a significant and constant memory leak (say, 10MB/1000evts), PROOF system will keep to kill workers until all of them will be dead? I mean, is it possible for PROOF to allocate new workers when some of them are killed?

–
Evgeny Boger

4

Hi,

Yes, that’s right.

Adding workers to a running query is something that will come but is not yet in. However, if there are memory leaks I think one should really make an effort to understand them: they may bring other subtle misbehaviors and it is not good, in my opinion, to leave them around.
We made an effort to integrate valgrind runs in PROOF (see root.cern.ch/drupal/content/runn … y-valgrind) and I believe it now works reasonably well to give good hints. We use regularly when we have weird problems and it always provides good input to solve the problems.

G. Ganis