Matti Kortelainen
Matti Kortelainen
Would someone be able to compare (again) the behavior of `cmsRun` (jemalloc), `cmsRunTC` (TCMalloc), and `cmsRunGlibC` (glibc malloc)?
@srimanob @jfernan2 Do you run your tests on an otherwise free machine, or is other activity possible? In https://github.com/cms-sw/cmssw/issues/42387 it was reported that in cases where the memory ends up...
> In our test cmsRunGlibC is running for the igprof part: e.g. igprof -d -mp -z -o step3_igprofMEM.gz -t cmsRunGlibC cmsRunGlibC step3_igprof.py > > but gives no difference in results...
I guess this issue could be closed now?
> Tagging framework team, @Dr15Jones @makortel can we avoid throwing in these cases where the framework is already shutting down due to an exception? It could indeed make sense to...
> > Tagging framework team, @Dr15Jones @makortel can we avoid throwing in these cases where the framework is already shutting down due to an exception? > > It could indeed...
Written that, I wonder what the other exception `HcalDigisProducerGPU::acquire()` is. Oh wait, the destructor of `ScopedContextAcquire` is not marked as `noexcept(false)` https://github.com/cms-sw/cmssw/blob/0efbba9e38c6e067e9327a6aac257083586cbaf5/HeterogeneousCore/CUDACore/interface/ScopedContext.h#L123 which I'd bet is the cause for `std::terminate()`...
@fwyzard Do you agree?
Ping @cmsbuild
assign Configuration/PyReleaseValidation