tacc-sing-jhub

Launches a JupyterHub instance on TACC HPC in a Singularity container.

NOTE: If your workflow is I/O intensive, please be sure to locate your working directory on the appropriate file system. Please see the TACC documentation for details.

Usage

1. From a login node, clone this repo and cd to root of this repo

git clone [email protected]:eho-tacc/tacc-sing-jhub.git tacc-sing-jhub

# alternatively, clone one of the application-specific branches, for instance DeepLabCut
git clone --single-branch --branch dlc [email protected]:eho-tacc/tacc-sing-jhub.git tacc-sing-jhub

cd tacc-sing-jhub

2. Configure environment by editing config.mk
   • ALLOCATION and PARTITION are passed to sbatch command as -A and -p options respectively. See man sbatch for details.
   • DOTENV is a path to an env-formatted file. It contains variables that will be exported before launching the Singularity container.
     • To set environment variables within the container, prefix variable names in the DOTENV with SINGULARITYENV_.
     • For instance, to set the variable SHELL within the container:

     # my_env.env
     SINGULARITYENV_SHELL=/bin/bash

     • dotenv/work2cache.env is an example that installs python packages on /work2
   • Singularity will pull the Docker IMAGE to POSIX path SIF.
     • For instance, the following config.mk will singularity pull ./my_image.sif docker://docker/whalesay:latest:

     IMAGE ?= docker/whalesay:latest 
     SIF ?= ./my_image.sif 

     • The Docker IMAGE must have the binary jupyter-notebook in its $PATH
   • Any of these variables can be overwritten when running make commands.
     • For instance, to overwrite PARTITION:

     PARTITION="p100" make jupyter-mav2

     • See GNU Make documentation for details.
3. Optional: manually pull Singularity image
   1. Enter idev session
   2. Pull image from Dockerhub to .sif file

   module load tacc-singularity
   make sif

   3. Exit idev session
4. Submit job via sbatch, using make jupyter-frontera or make jupyter-mav2:

$ make jupyter-frontera
...
tail -f jupyter.out

job 2822590 execution at: Tue Apr 13 14:35:22 CDT 2021
TACC: running on node c196-012
TACC: using DOTENV=work2cache.env
SINGULARITYENV_PYTHONUSERBASE=/work2/06634/eho/jupyter_packages
SINGULARITYENV_JUPYTER_PATH=/work2/06634/eho/jupyter_packages/share/jupyter:
SINGULARITYENV_JUPYTER_WORK=/work2/06634/eho/jupyter_packages
TACC: using singularity version 3.6.3-2.el7
TACC: using IPYTHON_BIN singularity exec --nv --home /work2/06634/eho/support/tacc-sing-jhub --bind /work2 ./tacc-ml.sif jupyter-notebook
TACC: using jupyter command: singularity exec --nv --home /work2/06634/eho/support/tacc-sing-jhub --bind /work2 ./tacc-ml.sif jupyter-notebook --config=/home1/00832/envision/tacc-tvp/server/scripts/frontera/jupyter.tvp.config.py
TACC: got login node jupyter port 11296
TACC: created reverse ports on Frontera logins
Your jupyter notebook server is now running!
Please point your favorite web browser to https://vis.tacc.utexas.edu:11296/?token=53ba17ctokengoeshere3b5ef243aa

6. Check the job output file ./jupyter.out for URL to JupyterHub, or for errors
7. Please remember to scancel from your SSH session after you are done using JupyterHub. It frees up the node for others to use :)

Troubleshooting

WIP

• Check $HOME/.jupyter/${NODE}.log for additional debugging logs. The $NODE is the compute node that the job is running on; you can retrieve it using the squeue command.