Project Description

Containers are great for scaling stateless services and making them robust. However, stateful services like databases continue to pose challenges. In particular, upgrading a running database service to a new version with near zero downtime is not trivial. Wouldn't it be great if a helm chart or K8S operator could automatically take care of this for you? This would allow you to fire up a database in your K8S cluster and consume it as-a-service. And put us one step closer to building a private cloud that runs anywhere, powered by pure open source.

Goal for this Hackweek

• Boot up an Amazon Kubernetes Service instance, deploy Rancher and register the cluster with Rancher
• Install a helm chart for an older version of a stateful service on the cluster, e.g. MariaDB, PostgreSQL, MongoDB or similar
• Put some data into the service and run a simulated load on it
• Naively upgrade the service to a newer version via helm and observe what, if anything, breaks
• Read database vendor instructions how human operators can achieve near zero downtime upgrades of the service
• Try and script these instructions to perform an automatic near zero downtime upgrade on the cluster
• Try and derive a general framework for near zero downtime upgrades of arbitrary stateful services (maybe as an extension to helm; needs to offer the right kind of service-specific extension points)

Resources

• Starting as an independent exploration
• Happy for anyone interested in Rancher, MariaDB and Helm to join!