-
Notifications
You must be signed in to change notification settings - Fork 524
-
Notifications
You must be signed in to change notification settings - Fork 524
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
IOError: Unable to create file (Mpi_err_other: known error not in list) #434
Comments
I'm not sure... there's something HDF5 doesn't like about your MPI installation. What version of MPI are you using? |
Hi Andrew. I ran mpicc --version in terminal and got Red Hat 4.4.7-4 And for mpirun --version Open MPI 1.6.2 |
I suspect that h5py is using the parallel disabled (default) install of HDF5 and not the parallel enabled I got through the dorsal script(s) (https://github.com/FEniCS/dorsal/blob/master/FEniCS/packages/hdf5.package). I have added the path for this install in my .bashrc file : export HDF5_DIR=/uio/home/student-u29/imranal/Work/FEniCS/ But when I run h5cc -showconfig , I see that Parallel HDF5 is set as no. Could this be the issue ? In which case, how can I make sure that the parallel enabled install is used in the above specified folder. |
Yes, that could certainly be the issue. Setting HDF5_DIR won't affect h5cc as that flag is h5py-only. |
According to the h5cc file : http://www.hdfgroup.org/ftp/HDF5/releases/hdf5-1.8.2/bin/linux/utilities/h5cc I can modify the install path. I did a locate on h5cc and got /usr/bin/h5cc Unfortunately I can not change the h5cc-64 file to include the correct install location, since I do not have write access to this file. How can I make sure that the parallel enabled HDF5 install is used ? (I might have forgotten to mention that I do not have root access.) |
|
I did the following (after cleaning up my python site-packages for h5py) : $ export CC=mpicc Running the tests again resulted in the same error message : ................EE................................................................................................................sss...........................x.......................s.......................................x.................................s...............................................ERROR: test_mpi_atomic (h5py.tests.test_file.TestDrivers) Traceback (most recent call last): ====================================================================== Traceback (most recent call last): Ran 306 tests in 0.382s FAILED (errors=2, skipped=5, expected failures=2) |
Now that I think about it, you must be already using a parallel build of HDF5 or you would have gotten an unresolved symbols error when compiling or loading h5py. Unfortunately I'm not sure what this error means. You could try asking on the HDF-Forum mailing list, to see if there's some incompatibility between your MPI setup and HDF5. I'll leave this issue open for the moment in case we learn more from that. |
Thanks, I will come back with an update once I find a solution. |
I finally managed to resolve the issue. According to the openmpi user forum : http://www.open-mpi.org/community/lists/users/2014/05/24288.php Using MPI.File.Open(,) would cause the crash. I was using a version of openMPI which was not thread safe. I was encouraged to install openmpi 1.8.1. That fixed the issue. |
What does this error imply and how can I fix it ?
Running the tests :
import h5py
h5py.run_tests()
I receive the following error message :
The text was updated successfully, but these errors were encountered: