Closed edwardhartnett closed 2 months ago
@edwardhartnett Is there a way to turn off Intel CI before you find a solution to solve this false alert?
No, instead we want to fix it.
For the last several Intel runs, I'm seeing "No space left on device" errors in the workflow summaries, e.g., https://github.com/NOAA-EMC/UPP/actions/runs/9909411518
It appears the logs are filling up: System.IO.IOException: No space left on device : '/home/runner/runners/2.317.0/_diag/Worker_20240712-142126-utc.log'
I would suggest turning off the verbose (-v
) option for spack install
, as there are other ways to obtain that information (I would suggest either saving off various log files as artifacts, or perhaps use the --show-log-on-error
option for spack install
so that detailed logs aren't spit out for the packages that install correctly).
@AlexanderRichert-NOAA thanks, I should have checked the action page to see the out of space error. However, now I am checking it and it's happening even after I removed the -v (which does drastically reduce log size).
Is there any way to get spack to use less space? Perhaps by doing the clean after each build instead of at the end?
I think the placement of the cache clean makes sense... One thing I'm noticing at the moment is that Spack is installing intel-oneapi-mpi, even though we've already installed that through apt. I think we can safely remove intel-oneapi-dev-utilities
and intel-oneapi-mpi-devel
from what gets installed by apt.
I'll see if I can find other ways to reduce the disk usage.
I will take those out of the apt install...
Taking out the apt install did not work:
38 f951: Fatal Error: Reading module ‘/home/runner/work/UPP/UPP/spack/o
pt/spack/linux-ubuntu20.04-zen2/gcc-10.5.0/intel-oneapi-mpi-2021.12.
1-ydoxetm3lceho5ctesx2d6pcds63cgpw/mpi/2021.12/include/mpi/mpi.mod’
at line 1 column 2: Unexpected EOF
39 compilation terminated.
40 make[2]: [src/CMakeFiles/nemsio.dir/build.make:130: src/CMakeFil es/nemsio.dir/nemsio_module_mpi.f90.o] Error 1 41 make[2]: Waiting for unfinished jobs.... 42 make[2]: Leaving directory '/tmp/runner/spack-stage/spack-stage-nems io-2.5.4-tx726tqt6lk7vxcztcdmjks2xydcxwal/spack-build-tx726tq' 43 make[1]: [CMakeFiles/Makefile2:917: src/CMakeFiles/nemsio.dir/al l] Error 2 44 make[1]: Leaving directory '/tmp/runner/spack-stage/spack-stage-nems io-2.5.4-tx726tqt6lk7vxcztcdmjks2xydcxwal/spack-build-tx726tq' 45 make: [Makefile:149: all] Error 2
That's weird, it appears to be building everything with gcc in that workflow... Can you try adding spack config add "packages:all:prefer:['%intel']"
somewhere before the concretization step?
I've done that.
It's now failing like this:
f951: Fatal Error: Reading module ‘/home/runner/work/UPP/UPP/spack/o
pt/spack/linux-ubuntu20.04-zen2/gcc-10.5.0/intel-oneapi-mpi-2021.12.
1-ydoxetm3lceho5ctesx2d6pcds63cgpw/mpi/2021.12/include/mpi/mpi.mod’
at line 1 column 2: Unexpected EOF
Are we sure that mpich is included in the intel environment?
To be clear, do you want to use Intel MPI or MPICH to provide MPI?
Intel
Since there are no plans for unit testing, there is no need to spend more time maintaining the CI build. It doesn't have any tests to run anyway. ;-)
I will close this issue.
Ends like this:
Not sure what is happening.
@AlexanderRichert-NOAA can you take a look?