Skip to content

Parallel Processing Coordinator. Splits dataset processing to run parallel cluster jobs and aggregates outputs

License

Notifications You must be signed in to change notification settings

DiamondLightSource/ParProcCo

Repository files navigation

ParProcCo

Requires a YAML configuration file in grandparent directory of package, CONDA_PREFIX/etc or /etc

--- !PPCConfig
allowed_programs:
    rs_map: msmapper_utils
    blah1: whatever_package1
    blah2: whatever_package2
url: https://slurm.local:8443
extra_property_envs: # optional mapping of properties to pass to Slurm's JobDescMsg
    account: MY_ACCOUNT # env var that holds account
    comment: mega job
valid_top_directories: # optional mapping of top directories accessible from cluster nodes
                       # (used to check job scripts, log and working directories)
    - /cluster_home
    - /cluster_apps

An entry point called ParProcCo.allowed_programs can be added to other packages' setup.py:

setup(
...
    entry_points={PPC_ENTRY_POINT: ['blah1 = whatever_package1']},
)

which will look for a module called blah1_wrapper in whatever_package1 package.

Testing

Tests can be run with

$ pytest tests

To test interactions with Slurm, set the following environment variables:

SLURM_REST_URL  # URL for server and port where the REST endpoints are hosted
SLURM_PARTITION # Slurm cluster parition 
SLURM_JWT       # JSON web token for access to REST endpoints

The environment can be set up and managed by running the create_env task in VSCode. This will read the token from ~/.ssh/slurm.tkn but will not check or generate the key. The resulting file .vscode/.env is used by the python.envFile setting to propagate these values automatically.

On the initial run, SLURM_REST_URL and SLURM_PARTITION will need to be given values manually (unless already set as environment variables). Those values will be kept whenever the task is rerun, with only the token being updated. As .vscode/.env is ignored by git, it is safe to save these values in that file.

If you are not using VSCode, running .vscode/create_env.sh will create the env file, and the variables can be exported using:

set -a
source ".vscode/.env"
set +a

About

Parallel Processing Coordinator. Splits dataset processing to run parallel cluster jobs and aggregates outputs

Resources

License

Stars

Watchers

Forks

Packages

No packages published