- Important note
- With Singularity
- Prerequisites
- Get a container image
- Option: Download a release image (preferred)
- Option: Download image from the latest master-branch build
- Run OGS inside a Container (called from outside)
- Custom Python environment for the container
- Run the DataExplorer inside a Container
- With Docker
- Prerequisites
- Run OGS inside a Docker container
date = "2018-11-14T15:00:13+01:00"
title = "Running OGS in a container"
author = "Lars Bilke"
weight = 2
[menu]
[menu.userguide]
parent = "basics"
Important note
This page describes how to run OGS with the help of a Linux container (for users). To build OGS with the help of a container go to the developer guide (for developers).
With Singularity
Prerequisites
- A running installation of Singularity 3.0 or higher:
- Available on Eve (UFZ),
envinf1
/envinf2
(UFZ), Juwels (SC Jülich), Taurus (TU Dresden). - See the developer guide for install instructions.
- Available on Eve (UFZ),
Get a container image
Option: Download a release image (preferred)
Simply download an image from the releases page.
Option: Download image from the latest master-branch build
Simply download an image from the latest master-branch build page.
Run OGS inside a Container (called from outside)
# Linux only:
singularity exec --app ogs ogs-6.2.2-serial.sif ogs some/path/project.prj
This starts the container, mounts your home directory inside the container, passes the current working directory and runs the ogs executable (in your home directory which is mounted inside the container) with the passed project file. Everything works as expected and is transparent to the user. When ogs finishes the container stops and returns to the host system.
The --app ogs
selects a predefined execution environment in the container (i.e. setting the PATH
to /scif/apps/ogs/bin
in which all the executables are located). You could also (and on macOS you have to) run without the --app
-parameter but then you had to specify the full executable path in the container:
# Works on macOS too:
singularity exec ogs-6.2.2-serial.sif /scif/apps/ogs/bin/ogs ...
Running a benchmark:
# Create output directories
mkdir -p _out _out_mpi
# Run serial benchmark
singularity exec --app ogs ogs-6.2.2-serial.sif ogs -o _out [ogs-sources]/Tests/Data/Mechanics/Linear/disc_with_hole.prj
# Run serial benchmark with output validation (via vtkdiff)
singularity exec --app ogs ogs-6.2.2-serial.sif ogs -o _out -r [ogs-sources]/Tests/Data/Mechanics/Linear [ogs-sources]/Tests/Data/Mechanics/Linear/disc_with_hole.prj
# Run parallel benchmark with MPI
mpirun -np 4 singularity exec --app ogs ogs-6.2.2-openmpi-2.1.2.sif ogs -o _out_mpi [ogs-sources]/Tests/Data/Mechanics/Linear/disc_with_hole.prj
You can run other contained executables as well, e.g. vtkdiff
:
singularity exec --app ogs ogs-6.2.2-serial.sif vtkdiff --help
You can interactively explore the container with singularity shell
(you can see that you are in the container because of the Singularity [container image file]:...>
prefix of the shell):
# Shell into container
singularity shell ogs-6.2.2-serial.sif
# List files in the container
Singularity ogs-6.2.2-serial.sif:...> ls /scif/apps/ogs/bin
... ogs tetgen vtkdiff
# Exit the container and get back to your hosts shell
Singularity ogs-6.2.2-serial.sif:...> exit
Custom Python environment for the container
For certain benchmarks or tutorials you may need additional Python packages. You can create a Python virtual environment inside the container (stored on your host) and install packages via pip
(inside the container):
mkdir my-working-directory && cd my-working-directory
singularity shell my-container.sif
# Now in the container
virtualenv .venv
source .venv/bin/activate
pip install numpy # or whatever you need
# run ogs
exit
# Now outside the container
# The virtualenv-directory .venv still persists
# If you want to run something in the container with exec, source the venv before:
singularity exec --app ogs my-container.sif bash -c 'source .venv/bin/activate && ogs ...'
Run the DataExplorer inside a Container
- Get a Singularity container with the DataExplorer (has
-gui
in its name) singularity exec --app ogs ogs-xxx-gui-xxx.sif DataExplorer
You may use this container on e.g. envinf1
with X11 forwarding (ssh -XY envinf1
).
With Docker
Although Singularity is the preferred container runtime you can use Docker too.
Prerequisites
- A running installation of Docker
Run OGS inside a Docker container
- Get the container:
docker pull registry.opengeosys.org/ogs/ogs/ogs-serial
- Start interactive container session:
docker run --rm -it registry.opengeosys.org/ogs/ogs/ogs-serial
- Run ogs:
/scif/apps/ogs/bin/ogs --version
- Exit the container:
exit
You will notice that the interactive session in your container is isolated from your host, i.e. you do not have access to files on your host. You need to explicitly mount them on docker run
:
mkdir ~/ogs_out
docker run --rm -it -v $HOME/code/ogs6/ogs/Tests/Data:/tmp/data:ro -v $HOME/ogs_out:/tmp/out registry.opengeosys.org/ogs/ogs/ogs-seria
/scif/apps/ogs/bin/ogs -o /tmp/out /tmp/data/Elliptic/cube_1x1x1_SteadyStateDiffusion/cube_1e4.prj
exit
ls ~/ogs_out
# [shows ogs generated output files]