Repository navigation
[release/2.1] Proper fixes for EPIC batch install script - #2211
Conversation
… EPIC Tier 1 hosts (JCSDA#2043) Add a batch install script for EPIC Tier 1 hosts, and update mirrors.yaml files for those hosts to define source and binary cache locations. File util/epic/intel_oneapi-2026.0.0_install.sh is added for potential future need / use. Both scripts shamelessly borrowed from NRL host scripts. (cherry picked from commit c17b15c)
…JCSDA#2115) Add required module pathing and loading for MSU hosts to successfully build oneAPI envs. GCC envs are not effected. Module pathing: - specify module use pathing for the EPIC-installed oneAPI compilers - specify module use pathing for the system-installed oneAPI MPI and MKL Module loading: load module git-lfs so that packages that require it are successfully fetched This is accomplished by adding configs/sites/tier1/{hercules, orion}/setup.sh scripts and referencing them in the top-level spack setup.sh script. This allows for removing incomplete module pathing commands from util/epic/batch_install.sh and instead place them in site-specific setup scripts. Placing these updates in configs/sites/tier1/{hercules, orion}/setup.sh ensures that they're available to both manual and automated builds. Note that a module pathing command for site derecho in util/epic/batch_install.sh was removed because it's already in configs/sites/tier1/derecho/setup.sh (cherry picked from commit b2f2346)
(cherry picked from commit 0f6a74d)
|
@EdwardSnyder-NOAA - thank you for making this change! How did you test these approaches? |
For the past few weeks, I have been using the |
@EdwardSnyder-NOAA - do you mean the commit from #2210? |
@natalie-perlin - I was referring to commit #2210. Are you planning on cherry picking this commit to release/2.1 then? I'm fine having this commit to future release branches. I'd figure since we are versioning control the |
@EdwardSnyder-NOAA - oh, sure, older branches such as release/2.1 may not need to be updated. |
…ors with a version tag (JCSDA#2210) spack bootstrap mirror writes metadata/sources and metadata/binaries at fixed paths. Rerunning it for a newer Spack (e.g., when doing spack-stack v2.2.0 install or spack-stack-nvhpc-dev) and attempting to use mirrors overwrites the metadata (including clingo.json, gnupg.json, patchelf.json) that installs from older spack-stack versions rely on. Change (util/epic/batch_install.sh): - Build a metadata key from Spack's bootstrap release: -. The tag is from BINARY_TARBALL in spack/lib/spack/spack/cmd/bootstrap.py, e.g. v2.2; the hash is the first 8 characters of the sha256 of the three github-actions-v2/*.json files. - Keep generating the mirror in a temp directory. Merge bootstrap_cache/ into the shared cache, which is safe because its files are stored by checksum and by name. Copy the metadata to metadata// instead of metadata/. - Rewrite url: in the copied metadata.yaml files to the absolute bootstrap_cache path, since the relative ../../bootstrap_cache no longer resolves one level deeper. - Register local-sources and local-binaries from metadata//{sources,binaries}. (cherry picked from commit bc146a5)
|
@natalie-perlin - I tested this PR by running both modes of the |
Use of generative AI
Please refer to our temporary spack-stack AI policy for further information.
Description
A few git commits from the dev branch to get the
batch_install.shscript working with EPIC GitHub Actions workflow for the release/2.1 branch.Dependencies
None.
Issues addressed
None.
Applications affected
None.
Systems affected
None.
Testing
Checklist