Each Domino installation comes with a standard Environment known as the Domino Analytics Distribution (DAD). Periodically, Domino publishes a new set of standard Environments with updated libraries and packages. These Environments include some common data science packages and libraries pre-configured for use in Domino.
Domino also makes a Domino Minimal Distribution (DMD) available that includes only the packages required to work in Domino. These are a good option if you want to build a Domino-compatible Environment from scratch. This helps speed Environment build times and execution start times.
Domino Cluster Environments work like a Domino Standard Environments, but have additional libraries to support specific types of clusters. They contain the same workspace tools and general packages as the DSE to support general data science workflows. You can use these Environments with any Domino execution like a Workspace or Job, but they are best used in a distributed compute cluster alongside a Domino Cluster Environment. These are smaller and simpler Environments and are only suitable for use in the worker nodes of the cluster. DSEs cannot be used for worker nodes in a cluster.
Domino Environments (abstractions on top of Docker images). Users can create Environments, use those provided by Domino, or edit existing ones.
- Domino Analytics Distribution (DAD)
-
The Domino Analytics Distributions can handle most of what a typical data science workflow needs out of the box. They include the most common Python and R packages along with an installation of CUDA that is required for using NVIDIA GPUs.
See the available Dockerfile and descriptions here: Domino Base Images.
quay.io/domino/base hosts the built images unless otherwise stated in the READMEs for the corresponding image. The version of the Domino Environment indicates the version of Domino that the Environment is shipped with, but typically any image will work on any version of Domino.
- Domino Minimal Distribution (DMD)
-
While the DAD includes most of what a data scientist needs to do their work, the DMD includes only the bare necessities required to work in Domino.
The DMD provides an image that allow you to:
-
Open Jupyter and Jupyterlab.
-
Batch run Python and R jobs.
-
Host a Shiny web app.
-
Publish a Python and R Model API.
-
Use Domino’s Git integration.
-
Install Python and R packages.
To shrink the DMD, remove workspaces that you won’t use or remove Python or R.
You can review the Dockerfile and descriptions here: Domino Base Images.
The images are hosted on quay.io/domino/base unless otherwise stated in the READMEs for the corresponding image. The version of the Domino Environment indicates the version of Domino that the Environment is shipped with, but typically any image will work on any version of Domino.
By default, the DSE includes fewer packages than the DAD, giving it a smaller footprint and making it faster and easier to work with. See Environments for more information about how to add packages to the DSE. The DSE comes with a GPU version that includes CUDA support and common packages for taking advantage of GPUs. WARNING: Domino recommend using an Environment with explicit GPU support when using GPU hardware tiers.
The version of the Domino Environment indicates the version of Domino that the Environment is shipped with, but typically any image will work on any version of Domino.
- Domino Minimal Environment (DME)
-
The DME has fewer packages and is lighter weight than the Domino Standard Environment (DSE).
The Domino Minimal Environment includes Jupyter and JupyterLab workspace support, but does not include several packages that are included in the DSE. Domino recommends using the DME if you will be doing several custom installations on top of a base Environment image, because its smaller size speeds build times and avoids conflicting dependencies. See Environments for information about how to add packages to the DME.
The version of the Domino Environment indicates the version of Domino that the Environment is shipped with, but typically any image will work on any version of Domino.
The following Environments are designed to be used with compute cluster Environments. You can use these Environments with any Domino execution like a Workspace or Job, but they are best used in a distributed compute cluster alongside a Domino Compute Cluster Environment.
Compute Cluster Environments work like the Domino Standard Environments, but they have additional libraries to support a specific cluster type. They contain the same workspace tools and general packages as the DSE to support general data science workflows.
A cluster won’t work correctly if the worker nodes are not using the appropriate Domino Cluster Environment for cluster workers and compatible Domino Compute Environment for Job or Workspace.
- Domino Spark Environment
The Domino Spark Environment is built specifically for workspaces that control a Spark cluster. It includes Scala and Spark on top of the typical DSE functionality. This Environment is best used alongside a Spark cluster Environment. To ensure compatibility between the Spark compute Environment and Spark cluster Environment, the Spark and Python versions must match across Environment images.
The version of the Domino Environment indicates the version of Domino that the Environment is shipped with, but typically any image will work on any version of Domino.
- Domino Ray Environment
The Domino Ray Environment is built specifically for workspaces that control a Ray cluster. It includes Ray on top of the typical DSE functionality. This Environment is best used alongside a Ray cluster Environment. To ensure compatibility between the Ray compute Environment and Ray cluster Environment, the Ray and Python versions must match across Environment images.
The version of the Domino Environment indicates the version of Domino that the Environment is shipped with, but typically any image will work on any version of Domino.
- Domino Dask Environment
The Domino Dask Environment is built specifically for workspaces that control a Dask cluster. It includes Dask on top of the typical DSE functionality. This Environment is best used alongside a Dask cluster Environment. To ensure compatibility between the Dask compute Environment and Dask cluster Environment, the Dask and Python versions must match across Environment images.
The version of the Domino Environment indicates the version of Domino that the Environment is shipped with, but typically any image will work on any version of Domino.
- GPU Environment flavors
The Domino Standard Environment also includes a GPU version with support for CUDA and common GPU specific libraries like Torch and Tensorflow. These GPU-enabled Environment images are larger, so Domino recommends that you use them only if you are using a GPU-enabled hardware tier.
- FUSE Environment flavors
The DSE Environment also has a version that includes FileSystem in Userspace (FUSE) binaries to enable Goofys and SSHFS support. You can add these commands to your Environment Dockerfile to enable FUSE functionality:
USER root
# Goofys
ADD https://github.com/kahing/goofys/releases/download/v0.24.0/goofys /usr/bin/
RUN chmod a+x /usr/bin/goofys
# SSHFS
RUN apt-get update && apt-get install -y sshfs &&
sed -i "s/^#user_allow_other/user_allow_other/" /etc/fuse.conf
USER ubuntu- How can I tell which image I’m currently using?
-
The URI for the image will be listed on your compute Environment’s overview page. If your Environment is built on top of another Environment, you may need to click through to the parent Environment before seeing the underlying docker image.
- I have a third-party Docker image, can I use that in Domino?
-
Maybe, but not likely without some customization. The DSE and DME are tested and configured to meet the Domino platform requirements and conventions. For example, by convention Domino uses /mnt as the default working directory. It is much easier to use the DME as your base Environment to build on top of than it is to try to get a 3rd party Environment to work directly in Domino.
However, this is not the case for Environments for compute cluster worker nodes. In most cases, these Environments can be plugged directly into Domino with no modifications, as they do not need to support the same workflows as the Domino Compute Environments.