Installation

Orchest can be run on Linux, macOS and Windows (using the exact same steps!).

Prerequisites

  • Docker

If you do not yet have Docker installed, please visit https://docs.docker.com/get-docker/.

Linux, macOS and Windows

Simply follow the steps below to install Orchest. For Windows, please read the note at the bottom first.

git clone https://github.com/orchest/orchest.git && cd orchest
./orchest install

# The previous command will only install language dependencies for
# Python. To specify other dependencies you can, for example, use:
# ./orchest install --lang=all

# Verify the installation.
./orchest --help

Note

On Windows, Docker has to be configured to use WSL 2. Make sure to clone Orchest inside the Linux environment. For more info and installation steps for Docker with WSL 2 backend, please visit https://docs.docker.com/docker-for-windows/wsl/.

Build from source

You should expect the build to finish in roughly 15 minutes.

git clone https://github.com/orchest/orchest.git && cd orchest

# Check out the version you would like to build.
git checkout v0.3.0

# Build all Docker containers from source (in parallel).
scripts/build_container.sh

# Verify the installation.
./orchest --help

Tip

We recommend building a tagged commit indicating a release. Other commits cannot be considered stable.

GPU support

Linux (supported)

For GPU images the host on which Orchest is running is required to have a GPU driver that is compatible with the CUDA version installed in the image. Compatible version pairs can be found here.

The GPU supported image orchest/base-kernel-py-gpu includes CUDA Toolkit 10.1. Which requires the NVIDIA driver on the host to be >= 418.39.

To find out which version of the NVIDIA driver you have installed on your host run nvidia-smi.

nvidia-smi is also available from within the GPU enabled image. Please note that when run from within the container it reports the CUDA Toolkit version installed on the host. To find out the CUDA Toolkit version installed in the container image run cat /usr/local/cuda/version.txt.

Additionally, we require the nvidia-container package to make sure Docker is able to provide GPU enabled containers. Installation of the nvidia-container is done using apt-get install nvidia-container-runtime.

See also

Docker GPU documentation
Most up to date instructions on installing Docker with NVIDIA GPU passthrough support.

Windows WSL 2 (supported)

For WSL 2 follow the CUDA on WSL User Guide provided by NVIDIA.

Please note that the “Docker Desktop WSL 2 backend” (meaning, you’ve installed Docker not directly in the WSL 2 environment but on the Windows host itself) does not support CUDA yet.

macOS (not supported)

Unfortunately, nvidia-docker does not support GPU enabled images on macOS (see FAQ on nvidia-docker).

Note

Make sure you have installed our GPU images for the programming language you want to use. This can be done using the --gpu flag on installation and update.

./orchest update --lang=python --gpu

Run Orchest on the cloud

Running Orchest on a cloud hosted VM (such as EC2) does not require a special installation. Simply follow the regular installation process.

To enable SSL run scripts/letsencrypt-nginx.sh <domain> <email> and restart Orchest ./orchest restart.

Please refer to the authentication section to enable the authentication server, giving you a login screen requiring a username and password before you can access Orchest.