Lapin Kubes

A demonstration of a peculiar networking issue when running Lapin + RabbitMQ within a Kubernetes environment.

The code here is a simple adaptation of the Lapin minimal example merely splitting it into two components, client and server, in order to have more control over the requests being sent.

Usage

Build as a Docker container and deploy into a Kubernetes cluster. An example configuration YAML file is supplied in k8s/lapin-kubes.yml.

Issue

Under Kubernetes the server stays connected and runs, producing regular heartbeats, but will not actually process any messages after ~10s of idle activity, or around ~40s after starting regardless of activity.

This does not happen in Docker on the same host/OS. It does not happen on the host operating system.

This does happen within Kubernetes to both Kubernetes-hosted RabbitMQ, and an external (non-Kubernetes) RabbitMQ server.

Tested using Kubernetes v1.22.1 on Docker 20.10.8 using Flannel as a CNI.

See also:

• Issue #347 - Kubernetes
• Issue #348 - Delivery stalls