Skip to content

Latest commit

 

History

History
97 lines (71 loc) · 3.94 KB

04-Rscript-function.md

File metadata and controls

97 lines (71 loc) · 3.94 KB

Running R Non-Interactively with Rscript

Working interactively in RStudio is ideal for exploration and development, but most production R workloads on Hyak Klone are run non-interactively using the Rscript command.

Rscript allows you to:

  • Execute a single R expression, or
  • Run an entire .R script from the command line

When combined with a container and Slurm, Rscript enables long-running or resource-intensive analyses to run unattended on compute nodes.

The Example R Script

We’ll start with a very simple R script called test.R:

cat test.R

This script doesn’t do much computationally, but it’s enough to demonstrate:

  • That R is running correctly
  • That packages inside the container are available
  • Where output from a batch job ends up

The Slurm Submission Script

Next, we’ll use a Slurm submission script to run this R script on the cluster. This file is called rscript-submit.slurm.

cat rscript-submit.slurm

Let’s walk through this script section by section.

Slurm Directives Explained

All lines starting with #SBATCH are directives to Slurm, not shell commands.

Job identification and placement

  • --job-name=rscript-submit Sets the job name (used in queue listings and output files).
  • --account=uwit Charges compute usage to this allocation.
  • --partition=compute Specifies which partition to run on. Use hyakalloc to see partitions you can access.

Resource requests

  • --time=5:00 Maximum wall time for the job.
  • --nodes=1 Number of nodes requested.
  • --ntasks=1 Number of tasks (maps to cores).
  • --mem=4G Total memory requested for the job.

Output handling

  • --signal=USR2 This directive tells Slurm to send the USR2 signal to your job shortly before it reaches its wall-time limit. This gives the application a chance to clean up, save state, or exit gracefully.
  • --output=%x_%j.out Writes standard output and standard error to a file named jobname_jobID.out.

This is where the output from test.R will appear.

Optional Side Quest: Email Notifications These lines configure email alerts:

#SBATCH --mail-type=BEGIN,END,FAIL
#SBATCH --mail-user=your_netid@uw.edu

They tell Slurm to email you when:

  • The job starts
  • The job finishes
  • The job fails

This is especially useful for long-running jobs that you don’t want to monitor manually. Add this to rscript-submit.slurm if you wish.

Running R Inside the Container

The final line is the actual command Slurm executes:

apptainer exec --bind /gscratch r-w-libs.sif Rscript test.R

This does the following:

  • Runs the command inside your custom container (r-w-libs.sif)
  • Binds /gscratch so your data and scripts are accessible
  • Executes Rscript test.R non-interactively

Because the container already includes the required R packages and system libraries, the job does not depend on your personal R library or an interactive session.

Submitting the Job

To submit the job to Slurm, run:

sbatch rscript-submit.slurm

Slurm will return a job ID. Once the job completes, you can view the output with:

cat rscript-submit_<jobID>.out

You should see the result of 5 + 5 and confirmation that tidyverse loaded successfully.

Key Takeaway: This pattern, custom container + Rscript + Slurm, is the standard way to run scalable, reproducible R workloads on Hyak Klone. Interactive RStudio sessions are for development; batch jobs are for getting real work done.

At this point, you should start to imagine how powerful this pattern becomes when your R script runs for hours, ingests large datasets, and writes results to disk without requiring your attention. By combining a custom container, Rscript, and Slurm, you can move seamlessly from interactive development in RStudio to fully unattended, production-scale analyses on Hyak Klone.

In the next section, you’ll put this into practice by running a real analysis end-to-end using Open OnDemand.