Containers bundle an application, the libraries and other executables it may need, and even the data used with the application into a portable, self-contained files called images. Containers simplify installation and management of software with complex dependencies and can also be used to package workflows. Singularity is a container application targeted to multi-user, high-performance computing systems. It interoperates well with SLURM and with the Lmod modules system. Singularity can be used to create and run its own containers, or it can import Docker containers.
Creating Singularity Containers
To create your own image from scratch, you must have root privileges on some computer running Linux (any version). Follow the instructions at the Singularity site. If you have only Mac or Windows, you can use the Vagrant environment. Vagrant is a pre-packed system that runs under several virtual-machine environments, including the free Virtualbox environment. Singularity provides instructions for installing on Mac or installing on Windows. Once you have installed Vitrualbox, you install Singularity's Vagrant image, which contains the prerequisites to author images. You can then follow the instructions for Linux to author your image.
You can also import existing Docker images with the singularity build command.
singularity build animage.simg docker://someaccount/animage
If you do not have the ability to create your own image for Rivanna or to use a Docker image, contact us for assistance.
Singularity on Rivanna
Singularity is available as a module, and a library of prepared Singularity container images is provided for popular applications as part of the shared software stack. Descriptions for these shared containers can be found via the module avail and module spider commands.
module load singularity module avail
The available singularity containers are listed under /apps/modulefiles/standard/container/singularity/2.4.1.
---------------------- /apps/modulefiles/standard/container/singularity/2.4.1 -------------------- cellprofiler/3.0.0 (D) danpos/2.2.2 (D) hydrator/0.0.2 tensorflow/1.6.0-py36
Loading any of these container modules produces an on-screen message with instructions on how to copy the container image file to the personal /scratch directory and how to run the container.
What is inside the containers?
To learn more about the applications and libraries installed in a container you can use the singularity help CONTAINER command, where CONTAINER is a placeholder for the actual container image file.
module load singularity module load tensorflow/1.6.0-py36 singularity help $CONTAINERDIR/tensorflow-1.6.0-py36.simg This container is backed by Anaconda version 4.4.0 and provides the Python 3.6 bindings for: * Tensorflow 1.6.0 * Keras 2.1.5 * PyTorch 0.3.1 * XGBoost * LightGBM * CUDA 9.0 * CuDNN 126.96.36.199
Running non-GPU Images
If your container does not require a GPU, all that is necessary is to load the singularity module and provide it with a path to the image.
module load singularity singularity <CMD> <OPTIONS> <IMAGEFILE> <ARGS>
- CMD defines how the container is used. There are three main commands:
- run: Executes a default command inside the container. The default command is defined at container build time.
- exec: Executes a specific application/command inside the container as specified with ARGS. This provides more flexibility than the run command.
- shell: Starts a new interactive command line shell inside the container.
- OPTIONS define how the singularity command is executed. These can be omitted in most cases.
- IMAGEFILE refers to the single Singularity container image file, typically with a .img or .simg extension.
- ARGS define additional arguments passed inside the container. In combination with the exec command they define what application/command to execute inside the container.
containerdir=~mst3k singularity run $containerdir/myimage.img
This executes a default application or set of commands inside the container. The default application or set of commands to execute is defined in the image build script and cannot be changed after the container is built. After execution of the default command, the container is closed.
singularity exec $containerdir/myimage.img python myscript.py
This is similar to singularity run but more versatile by allowing the specification of the particular application or command to execute inside the container. In this example it launches the python interpreter and executes the myscript.py script, assuming that Python was installed into the container image. After execution of the command, the container is closed.
singularity shell $containerdir/myimage.img
This opens a command line shell inside the container. Note that the command line prompt changes and includes the name of the container. You can navigate the container file system, including /scratch and /nv, and run any application that is installed inside the container. To leave the interactive container shell, type
Running GPU Images
Singularity can make use of the local NVIDIA drivers installed on the host. To use a GPU image, load the singularity module and add the --nv flag when executing the singularity shell, singularity exec, or singularity run commands.
module load singularity singularity <CMD> --nv <IMAGE_FILE> <ARGS>
$ containerdir=/share/resources/containers/singularity/gpu $ singularity run --nv $containerdir/tensorflow.2-1.4.1-py27.simg myscript.py
In the container build script, python was defined as the default command to be excuted and singularity passes the argument(s) after the image name, i.e. myscript.py, to the python interpreter. So the above singularity command is equivalent to
$ singularity exec --nv $containerdir/tensorflow.2-1.4.1-py27.simg python myscript.py
The tensorflow.2-1.4.1-py27.simg was built to include CUDA and cuDNN libraries that are required by TensorFlow. Since these libraries are provided by the container, we do not need to load the CUDA/cuDNN libraries available on the host.
Running Images Interactively
Start an ijob:
$ ijob -A mygroup -p gpu --gres=gpu -c 1 salloc: Pending job allocation 12345 salloc: job 12345 queued and waiting for resources salloc: job 12345 has been allocated resources salloc: Granted job allocation 12345 $ module purge $ module load singularity $ containerdir=/share/resources/containers/singularity/gpu/tensorflow $ singularity shell --nv $containerdir/tensorflow-1.4.1-py27.simg singularity tensorflow-1.4.1-py27.simg:~> python myscript.py
Example SLURM Batch Script
#!/usr/bin/env bash #SBATCH -J tftest #SBATCH -o tftest-%A.out #SBATCH -e tftest-%A.err #SBATCH -p gpu #SBATCH --gres=gpu:1 #SBATCH -c 1 #SBATCH -t 00:01:00 #SBATCH -A mygroup module purge module load singularity containerdir=/share/resources/containers/singularity/gpu/tensorflow singularity run --nv $containerdir/tensorflow-1.4.1-py27.simg $containerdir/tensorflowtest.py
Interaction with the Host File System
Each container provides its own file system. In addition, directories of the host file system can be overlayed onto the container file system so that host files can be accessed from within the container. These overlayed directories are referred to as bind paths or bind points. The following system directories of the host are exposed inside a container:
In addition, the following user directories are overlayed onto each container by default on Rivanna:
Due to the overlay these directories are by default the same inside and outside the container with the same read, write, and execute permissions. This means that file modifications in these directories (e.g. in /home) via processes running inside the container are persistent even after the container instance exits. The /nv and /project directories refer to leased storage locations that may not be available to all users.
Disabling the Default Bind Paths
Under some circumstances this default overlay of the host file systems is undesirable. Users can disable the overlay of /home, /scratch, /nv, /project by adding the -c flag when executing the singularity shell, singularity exec, or singularity run commands.
$ containerdir=~mst3k $ singularity run -c $containerdir/myimage.img
Adding Custom Bind Paths
Users can define custom bind paths for host directories via the -B/--bind option. The -B/--bind option can be used in combination with the -c flag.
For example, the following command adds the /scratch directory as an overlay without overlaying any other user directories provided by the host:
$ singularity run -c -B /scratch $containerdir/myimage.img
To add the /home directory on the host as /rivanna/home inside the container, execute this command:
$ singularity run -c -B /home:/rivanna/home $containerdir/myimage.img