scratch: Working with local scratch space

In episphere/biowR: Helper functions for Biowulf Cluster

Working with local scratch space

Description

These function work with the local scratch disk on the compute node of biowulf. Ideally, when you use the sinteractive, sbatch, or swarm command, you remembered to use the --gres=lscratch:XXX option which allocates XXX GB of local scratch disk. Another things to remember is to use

 export TMPDIR=/lscratch/$SLURM_JOB_ID

in your batch file or issue the commnand interactively. If you do this, then tempfile() and tempdir() with create files appropriately.

Usage

get_scratch_dir()

Details

If you are using swarm, the swarm task id is affixed the to scratch directory to prevent the different processes from sharing a common scratch space. Since you cannot control whether or not all the swarm tasks are run on the same node, even using /tmp or /lscratch/JOB_ID cannot guarantee that all the processes will have the same local scratch space.

Note

I removed the clear_scratch_space function. You cannot delete the /tmp or /lscratch/JOB_ID files. It may do more harm than good. Clean the space yourself. Maybe work in a directory under the scratch directory.

I use to worry about setting the scratch dir to a directory in /lscratch/JOB_ID/PROC_ID. I would add the swarm array task id to the directory name. since every instance of R has a unique tempdir, I dont need to work about this. for simplicity just use the tempdir and scratch. The new scratch directory is /lscratch/JOB_ID/RtmpXXXX/scratch. Each instance is unique.

episphere/biowR documentation built on Dec. 6, 2023, 1:44 a.m.