Cloud-native Patterns

This repository contains patterns that can be used to design complex, cloud-native applications. The paper A Cloud Native Platform for Stateful Streaming contains an in-depth description of these patterns.

Patterns

Controller

Kubernetes defines controllers as "control loops that tracks at least one resource type". We constrain that definition further: in cloud native Streams, a controller is a control loop that tracks a single resource type. Controllers take some action on creation, modification and deletion of a resource type. As with regular resources, custom resources can be monitored using controllers.

Conductor

In contrast to controllers, the conductor pattern observes events from multiple resources and does not save state updates in a local cache. Instead, they are concurrent control loops that maintain a state machine that transitions based on resource events, all towards a final goal. Conductors do not own any resources. Rather, they register themselves with existing controllers as generic event listeners which receive the same notifications that each controller does.

Coordinator

When asynchronous agents need to modify the same resource, we use the coordinator pattern. The coordinator pattern implements a multiple-reader, single-writer access model by granting ownership of the resource to a single agent and serializing asynchronous modification requests coming from other agents. Coordinators are synchronous command queues that serially execute modification commands on resources. In cloud native Streams, this pattern means that the controller for a resource owns that resource, and other controllers which want to modify it must make requests to that controller.

Getting started

The repository contains a custom Job resource example that makes use of those patterns. It is located in the com.ibm.cnp.samples package.

Compiling the code

To compile the code, simply import the repository as a Maven project in your favorite IDE, or run the following command at the root of the repository:

$ mvn package

Running the example

You need first to install the Job custom resource definition into your Kubernetes cluster:

$ kubectl apply -f ${REPO_ROOT}/crds/job.yaml

Then, run the com.ibm.cnp.samples.Main class either through your IDE or by typing the following command at the root of the repository:

$ mvn exec:java