ConScapeSlurm

SLURM job management for ConScape.jl

Runs a ConScape.jl BatchProblem as array jobs on a slurm cluster.

user/problem.jl holds the problem specification, loaded with using ConScapeSlurm. user/datasets.csv holds file paths and parameters for one or multpiple dataset that use this Problem specification.

The scripts folder holds paired julia .jl and bash .sh scripts. The .sh scripts can be edited to change SLURM settings, and set memory and time requirements, and the number of jobs to run.

Generally, a ConScape workflow looks like this.

1. Clone this repository, and make a branch or a fork to work in
2. Edit the datasets.csv and problem.jl files to match your needs.
3. rsync files to the path specified in dataset.csv on your cluster.
4. Run sbatch instantiate.sh to install the exact project dependencies specified in Manifest.toml.
5. Run sbatch assess.sh mydatasetname to calculate the number of jobs and windows for each dataset (row) in datasets.csv.
6. Run sbatch estimate.sh mydatasetname to estimate computation and memory needs.
7. In run.sh set array to e.g. 0-9 and other SLURM parameters, to see if things are working with sbatch run.sh mydatasetname.
8. Update array for the rest of the job and rerun sbatch run.sh mydatasetname, and repeat for each dataset.
9. Run sbatch reassess.sh mydatasetname and check the data map in the slurm file to make sure it is empty. If there are remaining jobs for some reason, update runs.sh with array=0-myremainingjobs and run sbatch run.sh mydatasetname.
10. Repeat until the map is empty, fixing any issues that may cause the same jobs to fail repeatedly.
11. Run sbatch mosaic.sh mydatasetname for each dataset to mosaic the outputs into rasters.
12. rsync output_* files from the paths in your datasets.csv back to your local machine.

If you have any problems following this, please make an issue in the issues tab above.