|View Issue Details|
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0001358||Orfeo Toolbox (OTB)||General||public||2017-02-08 12:19||2017-08-16 17:39|
|Summary||0001358: FFT applications random Segfault|
|Description||The dashboard report random segfaults with apTvDomainTransform_fft_shift_fwd on hulk.|
|Additional Information||This problem seems to come from a conflict with MPI and FFTW since removing MPI_FINALIZE nor cleaning the FFTW threads solve (at least partially) reproducible Segfaults.|
From https://github.com/open-mpi/ompi/issues/1364, calling MPI_FINALIZE in a static class destructor could cause segfault on OSX and there is no guarantee in which order the destructors are called at exit.
So the random segfault could occurs when MPI and FFTW cleaning are in different order.
A solution could be to move the call of MPI_FINALIZE outside of the otb::MPIConfig destructor or even to not call MPI_FINALIZE at all since if a process terminates without calling MPI_FINALIZE, independent processes are not affected. Also from the issue : "A much cleaner approach is to not call MPI in your destructor (you're terminating the application anyway)."
|Tags||No tags attached.|
3 otbMPIConfig related test are failing (09/02/2017)
But the FFT segfault dissapear.
Still need to wait few day to see if the random segfault is really removed.
|No more segfault occurs since 2017-03-07 on dashboard.|
Bug appears today:
It's probably a synchronisation issue with develop. Monitor and close bug if that's the case.
||Status||new => assigned|
||Assigned To||=> Ludovic|
||Note Added: 0004018|
||Note Added: 0004068|
||Status||assigned => resolved|
||Resolution||open => fixed|
|2017-04-05 10:20||poughov||Note Added: 0004091|
|2017-04-05 10:20||poughov||Status||resolved => feedback|
|2017-04-05 10:20||poughov||Resolution||fixed => reopened|
|2017-08-16 17:39||rashadkm||Status||feedback => resolved|
|2017-08-16 17:39||rashadkm||Resolution||reopened => fixed|