Memory Usage Too High about factorytools

[lawrenceleejr]

Or actually maybe it is... I'm going to try and clean some stuff up.

-L

[lawrenceleejr]

I've narrowed it down to the RJigsaw calculator. Removing that algorithm from the job leads to stable memory usage.

[lawrenceleejr]

I no longer see a memory leak and feel like a crazy person. I guess the scouting job is what measures the required memory? I'm back at square one.

[lawrenceleejr]

Ok I can see that if I comment out all of the algorithms except BasicEventSelection, there's no memory growth. The memory leak starts with CalibrateST.

In the execute call of CalibrateST, I've narrowed it down to the Get* calls to ST:

https://github.com/rsmith54/FactoryTools/blob/master/Root/CalibrateST.cxx#L207

Once L207-246 are uncommented, the leaking starts...

Maybe this is something we need to bring up with Martin+Zach?

Investigating.

-L

[lawrenceleejr]

Haha! Appears to be taus. Damn taus...

I'm going to comment out the GetTau call in CalibrateST and test on the grid.

[lawrenceleejr]

this appears to be entirely from the GetTaus() call in the CalibrateST alg.

Commenting out this line appears to remove the memory leak.

Memory usage on lxplus node after 10k events
Only BasicEventSelection+CalibrateST (w GetTaus() call) - 4.7%
Only BES+ST (no taus) - 3.1%

Entire job (no taus) - 3.1% 

Without the leak, should be able to run stably within 1 GB of RAM.

[lawrenceleejr]

Fixed in #124