Mathematics and Statistics Tools#

MATLAB singularity with License server enabled#

At allocated compute node; activate singularity and run MATLAB image

$module load singularity
$singularity run /app/matlab2023bLic.sif
or
$singularity run /scratch/matlab2023bLic.sif

This software license is depended on MUIT MATLAB license.

RStudio on Jupyter with The Rocker Project#

Without complicated setup, we can run container and work on jupyter notebook environment.

How to#

On any node: activate singularity

 $ module load singularity
 $ singularity run /app/rstudio.4.4.2.sif

Connect to RStudio Server via browser::
       http://compute_node.mahidol.ac.th:port

Reference: The Rocker Project

RStudio Server(Optin II)#

One time setup:

Create the required directories first:

$ TMPDIR=~/rstudio-tmp
$ mkdir -p $TMPDIR/var/lib
$ mkdir -p $TMPDIR/var/run
$ mkdir -p $TMPDIR/tmp
$ mkdir -p $TMPDIR/var/lib/rstudio-server
$ mkdir -p $TMPDIR/var/run/rstudio-server

Generate a secure cookie key (often the missing piece):

$ uuid > $TMPDIR/tmp/rstudio-server/secure-cookie-key 2>/dev/null || python3 -c "import uuid; print(uuid.uuid4())" > $TMPDIR/tmp/rstudio-server/secure-cookie-key
$ mkdir -p $TMPDIR/tmp/rstudio-server
$ chmod 600 $TMPDIR/tmp/rstudio-server/secure-cookie-key

Create an rsession.conf and database.conf, rsession.conf (can be empty or minimal):: $ touch $TMPDIR/rsession.conf

The database.conf - tells RStudio to use SQLite in a writable location:

$ echo "provider=sqlite directory=$TMPDIR/var/lib/rstudio-server" > $TMPDIR/database.conf

Allocate resource(without gpu):

$ salloc -w node_name -t 1:0:0

Allocate resource(with gpu):

$ salloc -w node_name -t 1:0:0 --gres=gpu:1

Troubleshouting Rstudio::#

Check Hostname:

$hostname

⎿ zeta

Check RStudio Server:

$ singularity exec /app/geospatial_latest.sif which rserver 2>&1

⎿ /usr/local/bin/rserver

Check if port 8989 is already in use:

$ ss -tlnp 2>/dev/null | grep 8989 || netstat -tlnp 2>/dev/null | grep 8989 || echo "Port 8989 appears free (or no permissi…)

⎿ Port 8989 appears free (or no permission to check)

Check for stale rstudio lock/pid files:

$ ls -la ~/rstudio-tmp/var/run/rstudio-rserver/ 2>&1

stalled files:

srw-rw-rw- 1 snit.san snit.san 0 Feb 13 14:18 session-server-rpc.socket

There’s a stale socket file from the previous run. Let me also check the rstudio server logs and try a clean run.

Check for any rstudio log files:

$ find ~/rstudio-tmp -name "*.log" -o -name "*.err" 2>/dev/null; ls ~/rstudio-tmp/tmp/rstudio-server/ 2>&1

⎿ secure-cookie-key

Clean Stal files::#

The fix: cleaned up the stale files:

$ rm -f ~/rstudio-tmp/var/run/rstudio-rserver/session-server-rpc.socket
$ rm -f ~/rstudio-tmp/var/lib/rstudio-os.sqlite*

Start container ::#

Singularity run command:

$ TMPDIR=~/rstudio-tmp && singularity exec -B $TMPDIR/var/lib:/var/lib/rstudio-server -B $TMPDIR/var/run:/var/run/rstudio-server -B $TMPDIR/tmp:/tmp -B $TMPDIR/database.conf:/etc/rstudio/database.conf -B $TMPDIR/rsession.conf:/etc/rstudio/rsession.conf /app/rstudio_4.4.2.sif rserver --www-address=0.0.0.0 --server-user=$USER --www-port=8989 --auth-none=1 --server-data-dir=$TMPDIR/var/run/rstudio-server    --secure-cookie-key-file=$TMPDIR/tmp/rstudio-server/secure-cookie-key

Don’t care about following message for now!::: TTY detected. Printing informational message about logging configuration. Logging configuration loaded from ‘/etc/rstudio/logging.conf’. Logging to ‘syslog’.

Reonnect to RStudio Server via browser:

http://compute_node.mahidol.ac.th:port

RStudio Customization Library#

Author: Snit Sanghlao and Claude Code (Anthropic)

Install RStudio Server in user space on Ubuntu-based HPC cluster nodes, expose it via Jupyter proxy, and run it alongside a GPU-enabled Conda environment.

Create a private software folder and unpack the RStudio Server binaries locally (no sudo or root privileges required).

# 1.1  Create your permanent software directory and a temporary working folder
mkdir -p ~/software/rstudio-server
mkdir -p ~/software/temp-rstudio
cd ~/software/temp-rstudio

# 1.2  Download the Ubuntu 22.04 (Jammy) RStudio Server .deb package
#      Check https://posit.co/download/rstudio-server/ for the latest link.
wget https://download2.rstudio.org/server/jammy/amd64/rstudio-server-2026.01.1-403-amd64.deb

# 1.3  Extract the .deb file locally (no root required)
dpkg -x rstudio-server-2026.01.1-403-amd64.deb .

# 1.4  Move the core binaries to your permanent folder and clean up
mv usr/lib/rstudio-server/* ~/software/rstudio-server/
cd ~/software
rm -rf temp-rstudio

Create the ~/privatemodules/rstudio/ directory and write a modulefile so that the Environment Modules system can expose the rserver binary.

# 2.1  Create the modulefile directory
mkdir -p ~/privatemodules/rstudio

# 2.2  Create the modulefile
nano ~/privatemodules/rstudio/2026.01.1

Paste the following Tcl content into the file, then save with Ctrl+O → Enter and exit with Ctrl+X:

#%Module1.0
proc ModulesHelp { } {
    puts stderr "Loads RStudio Server 2026.01.1 for Jupyter Proxy"
}

module-whatis "RStudio Server (User Space)"

set RSTUDIO_DIR /home/snit.san/software/rstudio-server

# Add the binary to your PATH so Jupyter proxy can find 'rserver'
prepend-path PATH $RSTUDIO_DIR/bin

Install JupyterLab, Jupyter Server Proxy, and the RStudio proxy extension via Conda. The rstudio Conda package is intentionally omitted because the user-space binary installed in Step 1 is used instead.

# 3.1  Create the GPU-enabled Conda environment
conda create -n rstudio_gpu -c conda-forge \
    python=3.10 \
    r-base=4.3 \
    r-essentials \
    r-hdf5r \
    jupyterlab \
    jupyter-server-proxy \
    jupyter-rsession-proxy

4.1 Load the Slurm module:

module load slurm

4.2 Request an interactive GPU node:

srun --partition=defq --gres=gpu:1 -t 0:1:0 --pty bash

4.3 On the compute node, set up the environment:

# Tell the cluster where your private modules live
module use ~/privatemodules

# Load your custom RStudio module (adds rserver to PATH)
module load rstudio/2026.01.1

# Activate the Conda environment
conda activate rstudio_gpu

# Force the Conda environment's bin directory to the front of PATH
export PATH=$CONDA_PREFIX/bin:$PATH

4.4 Verify that Jupyter resolves correctly:

which jupyter

4.5 Confirm the proxy extensions are registered:

jupyter server extension list

4.6 Launch JupyterLab:

jupyter lab --no-browser --ip=$(hostname -i)

Open the URL printed in your terminal in a local browser (you may need an SSH tunnel). In the JupyterLab Launcher, click the RStudio button. The proxy will locate the rserver binary from Step 1, connect it to the Conda R 4.3 environment, and start your RStudio session.