Title: A Cloud Native Platform for Stateful Streaming

Abstract: We present the architecture of a cloud native version of IBM Streams, with Kubernetes as our target platform. Streams is a general purpose streaming system with its own platform for managing applications and the compute clusters that execute those applications. Cloud native Streams replaces that platform with Kubernetes. By using Kubernetes as its platform, Streams is able to offload job management, life cycle tracking, address translation, fault tolerance and scheduling. This offloading is possible because we define custom resources that natively integrate into Kubernetes, allowing Streams to use Kubernetes' eventing system as its own. We use four design patterns to implement our system: controllers, conductors, coordinators and causal chains. Composing controllers, conductors and coordinators allows us to build deterministic state machines out of an asynchronous distributed system. The resulting implementation eliminates 75% of the original platform code. Our experimental results show that the performance of Kubernetes is an adequate replacement in most cases, but it has problems with oversubscription, networking latency, garbage collection and pod recovery.