Stateful Connections in Kubernetes: The Scaling Secrets Nobody Talks About#

André Mocke & Rodrigo Fior Kuntzer, Miro — Wildcard

• Why: Long-lived, stateful TCP/WebSocket connections are the awkward case Kubernetes' pod-churn model wasn't designed for, and this is a specific, non-networking-standards account of solving it in production for real-time collaboration at scale — a useful counterpoint to the mostly-stateless assumptions elsewhere in the program.
• What: How Miro built and operates a custom WebSocket connection manager in Kubernetes for real-time collaboration, handling connection rebalancing, draining, and graceful shutdown under enterprise compliance constraints. (from abstract only)
• How: Custom Kubernetes operators manage the WebSocket connection lifecycle — rebalancing live connections and draining them gracefully during scaling/deployment events instead of dropping them. (from abstract only)
• Where applicable: General for any Kubernetes workload with long-lived stateful client connections (real-time collaboration, gaming, trading); Miro-specific compliance constraints aside, the connection-lifecycle patterns generalize.