Skip to content

Latest commit

 

History

History
69 lines (45 loc) · 2.89 KB

containers.md

File metadata and controls

69 lines (45 loc) · 2.89 KB

Building the containers

We will need 3 containers:

  • one to run the training pipeline on Pachyderm
  • one to run the deployment pipeline on Pachyderm
  • one to serve predictions on Seldon

These containers require several parameters that need to be provided in the form of:

  • script parameters (i.e. parameters that will be provided in the Python's command line)
  • environment variables
  • Kubernetes secrets (like credentials, connection tokens, service account, etc...)

The image below depicts, for each component of the architecture, which parameters it needs and where it expects them:

screenshot

All files needed to create these containers are located inside the pachyderm-seldon/container folder (there is a subfolder for each image). As we are using Google Cloud, we will store our images into the Google Cloud Registry (GCR for short).

The training image

We just need to enter its folder and run a couple of Docker commands: the first one will create the image locally and the second one will push the image to GCR. Here are the commands:

cd pachyderm-seldon/container/train
docker build --tag gcr.io/PROJECT/pachyderm-seldon/train:0.0.1 .
docker push gcr.io/PROJECT/pachyderm-seldon/train:0.0.1

Of course, you have to replace PROJECT with the name of your Google Cloud project.

Pay attention that we are tagging the image with the 0.0.1 version. In the future, if you need to rebuild the image (i.e. you had to change the Python code) you need to use a newer version (like 0.0.2) and you also have to specify this newer version inside the training pipeline descriptor (which is training-pipeline.json).

After this step, you should see the new image in your local repository:

$ docker image list
REPOSITORY                                       TAG        IMAGE ID       CREATED         SIZE
gcr.io/determined-ai/pachyderm-seldon/train      0.0.1      72dde3ac3a41   8 days ago      1.31GB

The deployment image

This is created in the same way as the previous one:

cd pachyderm-seldon/container/deploy
docker build --tag gcr.io/PROJECT/pachyderm-seldon/deploy:0.0.1 .
docker push gcr.io/PROJECT/pachyderm-seldon/deploy:0.0.1

Again, if you need to rebuild the image, remember to update the deployment pipeline descriptor (which is deployment-pipeline.json).

The serving image

This is created in the same way of the previous ones and the same considerations apply:

cd pachyderm-seldon/container/serve
docker build --tag gcr.io/PROJECT/pachyderm-seldon/serve:0.0.1 .
docker push gcr.io/PROJECT/pachyderm-seldon/serve:0.0.1

In this case, if you need to rebuild the serving image, remember to update the deployment pipeline descriptor with the correct image version (here are two different images: one for the pipeline and another one passed as a parameter to the script).

Up | Next