Containers support
Users can run containers with enroot. Enroot uses the same underlying technologies as docker but without the isolation.
Download container image
To get started, import a docker image from a repository. In the following example, we are downloading a docker tensorflow image from the docker repository.
[IS000G3@origami ~]$ srun --partition=student --mem=2G --cpus-per-task=4 enroot import docker://tensorflow/tensorflow:latest-gpu
In the current working directory, a .sqsh will be created. This is the image file which was downloaded from the previous command.
In our example, a tensorflow+tensorflow+latest-gpu.sqsh is created
[IS000G3@origami ~]$ ls -lrt tensorflow+tensorflow+latest-gpu.sqsh
-rw-r--r--. 1 IS000G3 IS000G3 5958070272 Dec 21 15:17 tensorflow+tensorflow+latest-gpu.sqsh
Create container on cluster
Create a container from the .sqsh image. In our following example, we are creating a container named tensorflow from the tensorflow+tensorflow+latest-gpu.sqsh file
[IS000G3@origami ~]$ srun --partition=student --mem=2G --cpus-per-task=10 enroot create --name tensorflow "tensorflow+tensorflow+latest-gpu.sqsh"
When the command completes, check that the container has been created
Tip
You may remove the .sqfs file after creating the container to conserve home directory space
Using the container
With the container created, you may run it interactively with the srun command or submit it as a batch job using sbatch
srun example
The following example starts the container and places you in a interactive bash shell environment
[IS000G3@origami ~]$ srun --pty --partition=tester --mem=8G --cpus-per-task=4 --gres=gpu:1 enroot start tensorflow bash
sbatch example
The following example requests for a GPU resource from the scheulder to start a tensorflow container. It mounts the entire user home directory into the /tf directory on the container and executes the python code "echo.py"
#!/bin/bash
#SBATCH --ntasks=1
#SBATCH --time=00:30:00
#SBATCH --nodes=1
#SBATCH --mem=16GB
#SBATCH --cpus-per-task=4
#SBATCH --qos=studentqos
#SBATCH --partition=student
#SBATCH --gres=gpu:1
#SBATCH --job-name=containerJob
srun --pty --partition=student --mem=8G --cpus-per-task=4 --gres=gpu:1 enroot start --mount ~:/tf tensorflow python echo.py
Removing a container from your account
To remove a container, execute the following command