Explore using `targets` + `crew` with UA HPC

usa-npn / cales-thermal-calendars

Estimating trends in phenology in the northeastern US

https://usa-npn.github.io/cales-thermal-calendars/spatial-trends-report.html

MIT License

0 stars 0 forks source link

Explore using `targets` + `crew` with UA HPC #3

Closed Aariq closed 4 months ago

Aariq commented 5 months ago

It may be desirable to move this project to the HPC at some point, and I haven't yet figured out how to use crew with the HPC here.

Things I know:

Absolutely no R code can run on login nodes, so pipeline would need to be run from interactive session
Open On Demand gives you a RStudio GUI to work with, but you can't start SLURM jobs from it (although you can request a session with many cores and treat it like a multi-core local machine)

Things I don't know:

How to set up a crew controller for SLURM
What UA specific SLURM settings are required
Where the _targets/ store would live on the HPC (Does USA-NPN have a S3 bucket we could use?)

Aariq commented 5 months ago

https://github.com/cct-datascience/targets-uahpc

Aariq commented 5 months ago

Still need to test this with spatial data, which may be an issue since geotargets might not yet be correctly working with multiple workers.

Aariq commented 5 months ago

HPC is missing system dependencies for the terra package, namely geos-devel, proj-devel, and proj-epsg. I'll look into getting these installed.

Aariq commented 5 months ago

Ok, able to run geospatial targets on HPC now (needed module load gdal). Working on getting this specific repo to work on HPC next.

Aariq commented 5 months ago

data/prism needs to be stored in group/thereasm where there is more storage. Need to create a symlink or something for that.

Aariq commented 4 months ago

Not 100% working yet. Will open other HPC-specific issues as I encounter them

Aariq commented 4 months ago

Working 100% now. Just needed more memory per worker