Swelly

The Problem

Building a music bot for Discord that serves a few hundred servers is relatively straightforward. However, scaling to handle thousands of concurrent voice connections and processing real-time audio requires fundamentally rethinking infrastructure. Issues like memory leaks, WebSocket disconnections, and audio buffering latency become exponentially harder to solve at scale.

Architecture & Solution

To handle the immense load and provide 99.9% uptime, I architected a distributed, containerized microservices ecosystem over Google Cloud Platform (GCP).

• Load Balancing: Sharded the Discord.js instance into separate processes, reducing global failover risk.
• Audio Streaming Nodes: Separated the command-handling logic from the heavy FFmpeg audio processing by spinning up dedicated Lavalink/audio-nodes.
• Redis Caching Layer: Placed Redis at the core for session management, queue tracking, and rapid command rate-limiting, achieving ultra-low latency.
• Docker & K8s: Containerized the whole infrastructure to allow immediate auto-scaling when server traffic spikes during peak weekend hours.

Challenges Faced

Managing memory consumption on Node.js engines processing massive voice payloads frequently hit node limits. I tackled this by utilizing aggressive garbage collection strategies and load-shifting intense I/O operations directly to compiled C++ bindings. The second massive challenge was avoiding YouTube API rate limits; requiring round-robin proxy fleets to fetch video streams securely over prolonged periods.

The Result

Swelly runs smoothly across 32,000+ Discord servers reaching an aggregated 6 million global users. It effectively manages hundreds of uninterrupted parallel streams while providing AI-driven recommendations.