Jira ticket:(see https://hysds-core.atlassian.net/browse/HC-26)

Introduction

This guide will teach you how to setup and configure your HySDS cluster to use a docker registry and S3 storage backend to host the cluster's docker images.

The following AWS architecture diagram provides an overview of the concept of the PAL (Processing Algorithm Library). The PAL is essentially a storage location that stores the container images needed by compute workers (verdi) when executing HySDS jobs. Each worker downloads the container image prior to executing the HySDS job.

Open

The following diagram details the default mechanisms by which container images are stored in the PAL (by the CI/CD subsystem) and loaded from the PAL (by the compute worker subsystem). Note that docker images are exported to tarballs and stored in the PAL as normal objects (files). As a result, there is overhead in the cluster's handling of docker images since operations involving importing and exporting of docker images involves compression or extraction of all layers of a particular image.  

Open

With the advent of the container registry feature of HySDS, the CI/CD and compute worker subsystems (verdi) can operate much more efficiently by utilizing the native layering capabilities of the docker daemon. The following diagram details the registry-backed mechanisms by which container images are stored in the PAL (by the CI/CD subsystem) and loaded from the PAL (by the compute worker subsystem).

With this feature enabled, the cluster's handling of docker images is optimized in that only image layers that are needed are transferred. Additionally, the worker processes on verdi will always pull the latest version of a particular container and tag from the registry. This is especially useful on development clusters where you would have to delete the cached docker image on the test worker prior to running a test job.

Requirements

• Configured HySDS Cluster (HySDS framework v3.0.0-rc.5 or greater)

• AWS S3 bucket for storage of the docker layers

Setup

1. After setup and configuration of your HySDS cluster, ensure that the docker registry image was published to your code bucket:

   
   
   

2. Log into your mozart instance.

3. Update the ~/.sds/config file. For example, if your code bucket was ops-code-bucket:

   CONTAINER_REGISTRY: localhost:5050 CONTAINER_REGISTRY_BUCKET: ops-code-bucket

   
   
   

4. Stop the cluster: sds stop all -f

5. Update the cluster's configuration: sds update all -f -c

6. Ship out the updated code/config tarballs: sds ship

7. Start the cluster back up: sds start all -f