HPSCTerrSys / TSMP2

CMake-based TerrSysMP
https://github.com/HPSCTerrSys/TSMP
MIT License
2 stars 2 forks source link

Building errors #34

Closed s7yoewer closed 5 days ago

s7yoewer commented 6 days ago

Since the last maintanance on JUWELS, I get some weird errors when trying to run the model with many ensemble members (buffer overflow detected). I have already written the support and they are still searching for the cause. The support team has written me that my error while running does not seem to occur when either using Stages/2024 or with OpenMPI in Stages/2023. Now I want to recompile / clean compile my model and I get some errors cause by missing modules dependencies:

CMake Error at /p/software/fs/juwels/stages/2023/software/CMake/3.23.1-GCCcore-11.3.0/share/cmake-3.23/Modules/FindPkgConfig.cmake:659 (message): pkg-config tool not found Call Stack (most recent call first): /p/software/fs/juwels/stages/2023/software/CMake/3.23.1-GCCcore-11.3.0/share/cmake-3.23/Modules/FindPkgConfig.cmake:825 (_pkg_check_modules_internal) cmake/FindMPIFortran.cmake:14 (pkg_check_modules) CMakeLists.txt:37 (find_package)

CMake Error at /p/software/fs/juwels/stages/2023/software/CMake/3.23.1-GCCcore-11.3.0/share/cmake-3.23/Modules/FindPackageHandleStandardArgs.cmake:230 (message): Could NOT find MPIFortran (missing: MPI_Fortran_INCLUDE_DIRS MPI_Fortran_LIB_DIR) (found version "3.1") Call Stack (most recent call first): /p/software/fs/juwels/stages/2023/software/CMake/3.23.1-GCCcore-11.3.0/share/cmake-3.23/Modules/FindPackageHandleStandardArgs.cmake:594 (_FPHSA_FAILURE_MESSAGE) cmake/FindMPIFortran.cmake:21 (find_package_handle_standard_args) CMakeLists.txt:37 (find_package)

Is the same issue causing this? I also tried to install everything new without any of my changes in eCLM or PDAF (the components that I use) and am facing this error with the quickstart as well as when using cmake.

s-poll commented 6 days ago

The error should be fixed with commit f8c1ab1. You will find more information about the error in #32. Please retry with these changes. Thanks.

DCaviedesV commented 6 days ago

Yes, the buffer overflow is a bit of an issue, and the changes in environments can be tricky. We are working on this. See this pull request, which I understand solves your specific issue.

There's also work to move everything into Stages/2024. @rigor, is something missing in the PR strictly for Stages/2024?

kvrigor commented 6 days ago

There's also work to move everything into Stages/2024. @Rigor, is something missing in the PR strictly for Stages/2024?

Only the minimal working CI which should build eCLM-ParFlow successfully. This build already succeeds on my local Ubuntu machine; last roadblock is to make it work on the Ubuntu CI job without breaking the standard JSC build. Hopefully I could iron out all remaining build errors in the next 1-2 weeks.