resolves: 103 create workflow for running benchmarks #104
Conversation
|
Merge #94 before. |
JavierCladellas
added
the
github_actions
JavierCladellas
changed the title
|
@prudhomm @vincentchabannes Benchmarks can now be triggered (after workflow is in master) via repository and workflow dispatch by providing the machines and benchmark json configs via a girder ID. Apptainer is not ready yet: bug related with mpi Parameters are:
|
prudhomm
left a comment
prudhomm
left a comment
There was a problem hiding this comment.
There are some hard coded paths, we need at some point in the future to change this
There was a problem hiding this comment.
I don't think that will work on eurohpc systems
There was a problem hiding this comment.
you didn't use the work in feelpp-kub it seems
| "reframe_base_dir":"/data/home/cladellash/benchmarking/build/reframe", | ||
| "reports_base_dir":"/data/home/cladellash/benchmarking/reports/" |
There was a problem hiding this comment.
the path should not be a personal one
| "output_directory": "/data/scratch/cladellash/feelppdb/toolboxes/heatlfuid/eye", | ||
| "use_case_name": "eye", | ||
| "options": [ | ||
| "--config-files /data/scratch/cladellash/feelppdb/input_data/eye/eye-{{parameters.position.value}}.cfg /data/scratch/cladellash/feelppdb/input_data/eye/pc_{{parameters.solver.value}}.cfg", |
There was a problem hiding this comment.
it should not be a personal path and probably relative to some root path for the framework
| "--heat-fluid.heat.scalability-save=1", | ||
| "--heat-fluid.fluid.scalability-save=1", | ||
| "--repository.append.np 0", | ||
| "--heat-fluid.json.patch='{\"op\": \"replace\",\"path\": \"/Meshes/heatfluid/Import/filename\",\"value\": \"/data/home/saigre/pulications/mesh.eye/mesh/Mr/{{parameters.meshes.value}}/Eye_Mesh3D_p{{parameters.nb_tasks.tasks.value}}.json\" }'" |
|
@JavierCladellas isn't it possible to split the json describing the benchmark and avoid duplicating eg the plot section between apptainer and builting ? |
This is already done in #109 . The benchmark configuration should now be modified as little as possible. Also, plots section will be moved to a separate file in #111 |
Main changes:
benchmark.ymlworkflow that launches the tests on a given machine (via configs on girder), and creates a PR (containing only docs) with the new benchmark reports.It has only been tested on gaya, with a builtin config and apptainer config. Apptainer config seems to have a bug concerning MPI bindings, but this bug should be independent of the pipeline.