Build the Container Images#

This guide explains how to build the Dioptra container images from the source repository. After completing these steps, you will have local container images ready for deployment.

Note

Most users do not need to build images manually. Pre-built images (including GPU workers) are available for download from the GitHub Container Registry. See Download the Container Images for the faster approach.

Build images locally only if you need:

  • Custom modifications to the container images

  • To work in an environment without access to the GitHub Container Registry

Prerequisites#

  • Docker Engine installed and running

  • Git installed

  • A terminal with access to Docker commands

Building the Images#

Step 1: Clone the Repository#

Clone the Dioptra GitHub repository if you have not already done so.

git clone https://github.com/usnistgov/dioptra.git

git clone [email protected]:usnistgov/dioptra.git

Change into the cloned repository directory:

cd dioptra

Note

To build the latest development versions of the containers, switch to the dev branch:

git checkout -b dev origin/dev

Step 2: Add CA Certificates (Optional)#

If you are building the containers in a corporate environment that uses its own certificate authority, copy your CA certificates into the docker/ca-certificates/ folder before building.

cp /path/to/your/ca-certificate.crt docker/ca-certificates/

See the docker/ca-certificates/README.md file for additional information.

Step 3: Build the Container Images#

Use the Makefile to build the container images:

make build-nginx build-mlflow-tracking build-restapi build-pytorch-cpu build-tensorflow-cpu

Note

The PyTorch and TensorFlow images may take a while to build.

Tip

If make cannot find your Python executable, specify it manually by prepending PY=/path/to/python3 to the command.

Step 4: Build GPU Images (Optional)#

If you plan to run GPU-accelerated workers, build the GPU-enabled images. These images require a host machine with CUDA-compatible GPUs and the NVIDIA Container Toolkit to be useful.

make build-pytorch-gpu build-tensorflow-gpu

Step 5: Verify the Images#

Run docker images to verify that the container images are available with the dev tag:

docker images | grep dioptra

You should see output similar to the following:

REPOSITORY                  TAG       IMAGE ID       CREATED         SIZE
dioptra/nginx               dev       17235f76d81c   3 weeks ago     243MB
dioptra/restapi             dev       f7e59af397ae   3 weeks ago     1.16GB
dioptra/mlflow-tracking     dev       56c574822dad   3 weeks ago     1.04GB
dioptra/pytorch-cpu         dev       5309d66defd5   3 weeks ago     3.74GB
dioptra/tensorflow2-cpu     dev       13c4784dd4f0   3 weeks ago     3.73GB

Note

The IMAGE ID, CREATED, and SIZE fields will vary. Verify that the REPOSITORY and TAG columns match.

Warning

Locally built images have a different registry prefix than downloaded images. See Understanding Container Registry Prefixes for implications when configuring your deployment.

Next Steps#

Once you have finished building the container images, move onto the next step: Prepare Your Deployment

See Also#