Using Docker instead of Kubernetes¶
In addition to pushing to Kubernetes, DSS can leverage standalone Docker daemons. This is a very specific setup, and we recommend using Kubernetes preferably.
DSS is not responsible for setting up your Docker daemon.
Dataiku DSS is not compatible with podman, the alternative container engine for Redhat 8 / CentOS 8
To run workloads in Docker:
- You must have an existing Docker daemon. The
dockercommand on the DSS machine must be fully functional and usable by the user running DSS. That includes the permission to build images, and thus access to a Docker socket.
For Docker execution, you may or may not need to push images to an image registry. Pushing images to an image registry is required if you plan to run workloads on multiple Docker daemons, or if you plan to build images on a Docker daemon and to run workloads on another Docker daemon.
If you plan to push images to an image registry:
- The local
dockercommand must have permission to push images to your image registry.
- All other docker daemons need to have permission to pull images from your image registry.
- Your DSS machine must have direct outgoing Internet access in order to install packages.
- Your containers must have direct outgoing Internet access in order to install packages.
- The containers running locally or on the cluster must be able to open TCP connections on the DSS host on any port.
Before you can deploy to Docker, at least one “base image” must be constructed.
After each upgrade of DSS, you must rebuild all base images.
From the DSS data directory, run
./bin/dssadmin build-base-image --type container-exec
You then need to create containerized execution configurations. In Administration > Settings > Containerized execution, click Add another config, switch “Container engine” to “Docker” and specify an image repository if needed (in which case you would need to push the base image using the button on top of the screen).
Containerized execution configuration can be used:
- In the project settings. In that case, the configuration will apply by default to all project activities that can run on containers.
- In the advanced settings of a recipe.
- In the Execution environment tab of in-memory machine learning design screen.
docker command line is the Docker client. The Docker daemon, responsible for building images and running the containers, may be on the same server or may be remote.
Use cases for a remote docker daemon running your containers include:
- Offloading heavy work onto other servers.
- Leveraging resources available on another machine (like GPUs).
Furthermore, the Docker daemon runs with high privileges, and on some setups it may be moved to another server rather than kept locally.
You do not need a specific setup if all of the following conditions are met:
- You are using an image registry.
- On the DSS server, you have a local Docker daemon that can push to that registry.
- The remote Docker daemon can pull from that registry.
Then the local Docker daemon can build the images, and the remote daemon can use those images to run the containers.
Otherwise, the remote Docker daemon has to build the images.
You still need the local
docker command (Docker client) to be fully functional and usable by the user running DSS.
You then need to specify the Docker daemon host before building the base image:
If you are using TLS to securely connect to the remote daemon, then you will also need the corresponding environment variables:
export DOCKER_TLS_VERIFY=1 export DOCKER_CERT_PATH=/path/to/docker/cert/directory/
DOCKER_CERT_PATH is the path to a folder that contains the client certificates:
It can be omitted if it is the default
For more information about Docker and TLS, refer to the Docker documentation.
You can now build the base image as described in setup-docker.
Thereafter, you need to specify the same settings in the containerized execution configurations:
- The Docker host
- If using TLS authentication, check “Enable TLS”, and provide the path to the directory with the certificates.
If necessary, rebuild images for code environments. For details, see Using code envs with containerized execution.
By selecting the corresponding containerized execution configuration, you are now ready to deploy your workload on remote Docker containers.
If you have several remote Docker daemons, you would have to create multiple containerized execution configurations, and to manually dispatch execution among those configurations. DSS does not automatically dispatch among multiple configurations.