pymapdl
pymapdl copied to clipboard
ansys
Supporting SLURM env vars for launching MAPDL configuration
As the title.
The idea is that on SLURM HPC clusters, PyMAPDL will read the SLURM job through the env vars that the SLURM manager creates, so it can launch MAPDL with the appropriate number of cores.
You can see in the code:
options = [
# 4, # Fall back option
SLURM_CPUS_PER_TASK * SLURM_NTASKS, # (CPUs)
SLURM_NPROCS, # (CPUs)
# SLURM_NTASKS, # (tasks) Not necessary the number of CPUs,
# SLURM_NNODES * SLURM_TASKS_PER_NODE * SLURM_CPUS_PER_TASK, # (CPUs)
SLURM_CPUS_ON_NODE * SLURM_NNODES, # (cpus)
]
nproc = max(options)
which is the way I use to decide how many cores should MAPDL instance use.
Same with memory:
if SLURM_MEM_PER_NODE:
# RAM argument is in MB, so we need to convert
if SLURM_MEM_PER_NODE[-1] == "T": # tera
ram = int(SLURM_MEM_PER_NODE[:-1]) * 10**6
elif SLURM_MEM_PER_NODE[-1] == "G": # giga
ram = int(SLURM_MEM_PER_NODE[:-1]) * 10**3
elif SLURM_MEM_PER_NODE[-1] == "G": # mega
ram = int(SLURM_MEM_PER_NODE[:-1]) * 10**0
elif SLURM_MEM_PER_NODE[-1].upper() == "k": # mega
ram = int(SLURM_MEM_PER_NODE[:-1]) * 10 ** (-3)
else:
ram = int(SLURM_MEM_PER_NODE)
I do not use the MAPDL -machines argument to specify the machines used. I don't think it is needed/compatible.
Thanks for opening a Pull Request. If you want to perform a review write a comment saying:
@ansys-reviewer-bot review
![ansys-reviewer-bot[bot] avatar](https://avatars.githubusercontent.com/in/383621?v=4 "Published by a pub.dev verified publisher"){kind=small}
Considering this https://github.com/koesterlab/setup-slurm-action for testing this PR.
However, sharing the MAPDL installation with the cluster seems a challenge.
GitHub
A github action to setup a small SLURM cluster for testing purposes. - koesterlab/setup-slurm-action
Considering this https://github.com/koesterlab/setup-slurm-action for testing this PR.
However, sharing the MAPDL installation with the cluster seems a challenge.
GitHub**GitHub - koesterlab/setup-slurm-action: A github action to setup a small SLURM cluster for testing purposes.**A github action to setup a small SLURM cluster for testing purposes. - koesterlab/setup-slurm-action
or use the pcluster pipelines to do it?
A github action to setup a small SLURM cluster for testing purposes. - koesterlab/setup-slurm-action
Considering this https://github.com/koesterlab/setup-slurm-action for testing this PR. However, sharing the MAPDL installation with the cluster seems a challenge.
or use the pcluster pipelines to do it?
I'm not sure if I want to have different parts of the testing in different repositories..... I can test from pymapdl repository using the pcluster pipelines??
A github action to setup a small SLURM cluster for testing purposes. - koesterlab/setup-slurm-action
cc @greschd . This might be related to your interest in launcher configuration
Wiz Scan Summary
| IaC Misconfigurations | |
| Vulnerabilities | |
| Sensitive Data | |
| Total | |
| Secrets | |
![wiz-inc-572fc38784[bot] avatar](https://avatars.githubusercontent.com/u/66023092?v=4 "Published by a pub.dev verified publisher"){kind=small}
Codecov Report
Attention: Patch coverage is 75.20000% with 31 lines in your changes missing coverage. Please review.
Project coverage is 86.88%. Comparing base (
93dd176) to head (36d33e6). Report is 9 commits behind head on main.
Additional details and impacted files
@@ Coverage Diff @@
## main #2754 +/- ##
==========================================
- Coverage 87.13% 86.88% -0.25%
==========================================
Files 55 55
Lines 9816 11070 +1254
==========================================
+ Hits 8553 9618 +1065
- Misses 1263 1452 +189
This PR should be complemented with https://github.com/ansys-internal/rep-orchestration-interfaces library.
GitHub is where people build software. More than 100 million people use GitHub to discover, fork, and contribute to over 420 million projects.
