For some tools, the shared Toolforge Redis instance is a performance/functionality bottleneck. T318479: Intermittent redis connection timeouts in Toolforge is one documented example of this.
Today it is possible for a tool to use the Build Service and its Apt builder to create a custom image running Redis. This image can then be paired with a manually created Service object to expose a task running the image to other Pods within the tool's namespace. See https://gitlab.wikimedia.org/toolforge-repos/wikibugs2-znc#deploy-on-toolforge for documentation of this pattern for a similar non-public Service. T348758: [jobs-api,jobs-cli] Support services in jobs envisions a future where exposing the service port is a more easily accomplished process with toolforge jobs.
In this task I propose Toolforge providing a 'redis' container image in the same general way that today it provides a 'python3.11' container image. This would make deploying a tool-local Redis service easier by eliminating the need for a tool to make its own container image. When T348758 is eventually implemented it will make running an isolated Redis for task queuing and other latency sensitive applications relatively trivial.