WebRTC.ventures scaling strategy solves horizontal clustering gap for Janus server
WebRTC.ventures details the architectural requirements for scaling the open-source Janus WebRTC Server, which lacks native clustering capabilities. The guide explains implementing a 'Media Resource Broker' using Redis and Kubernetes to manage room state persistence across multiple service instances.
Key Takeaways
- Janus rooms are strictly instance-local, meaning media paths and session state are pinned to a single server's memory.
- The MRB pattern uses a stateless router and Redis directory to map specific rooms to stable instance handles like Kubernetes pod ordinals.
- Scaling in requires a mandatory 'drain-before-remove' discipline to prevent dropping active calls during cluster updates.
- Janus's admin HTTP interface is disabled by default, often causing silent failures in automated orchestration layers.
- Horizontal scaling can be automated using KEDA to resize the pool based on real-time session counts gathered via the admin API.
Why It Matters
While Janus is favored for its modular C core, its inability to share state between instances has long been a hurdle for enterprise-grade scaling. This Media Resource Broker approach provides a standardized blueprint for building high-concurrency Selective Forwarding Unit (SFU) clusters without the overhead of complex, homegrown coordination logic. For the broader ecosystem, it reinforces the shift toward Kubernetes-native media infrastructure where orchestration is handled at the platform level rather than the application level. Strategists should monitor the upcoming integration with STUNner, which aims to simplify NAT traversal and media delivery within private Kubernetes environments.
Additional Context
The expansion of Janus architectural patterns comes as the WebRTC industry increasingly prioritizes specialized Selective Forwarding Unit (SFU) footprints over traditional Multipoint Control Units (MCU). According to reporting from Fora Soft in late 2025, SFUs have become the 2026 default for group video, as they offer 4-10x lower compute costs compared to MCUs by forwarding media streams without server-side transcoding. This shift has placed higher importance on external orchestration layers like the Media Resource Broker, as SFUs must manage room-to-server pinning more strictly than stateless web services. Recent developments in the Kubernetes ecosystem are also simplifying these deployments. Per OneUptime in February 2026, the graduation of KEDA (Kubernetes Event-Driven Autoscaling) as a CNCF project has introduced improved Prometheus integration, allowing media servers to scale based on granular application metrics rather than just CPU or memory stress. Furthermore, Meetecho, the maintainers of Janus, released v1.3.3 in October 2025, which included critical refactors for keyframe buffering and RTSP reconnection—features that bolster the server's stability in the large-scale, high-concurrency environments that WebRTC.ventures' MRB architecture is designed to support. In addition to scaling, networking remains a technical bottleneck. Documentation from STUNner in March 2026 highlights that the traditional 'hostNetwork: true' hack for running media servers in Kubernetes is being replaced by media gateways that expose server pools via private IP networks. This aligns with WebRTC.ventures’ current roadmap to integrate Janus with STUNner for cleaner NAT traversal. Industry analysts at Red5 noted in July 2026 that while cluster-based architectures are now standard for virtually unlimited scale, the 'drain' period remains the most critical guardrail for maintaining quality of service during automated scale-in events.
Read full article at webrtc.ventures
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source