Skip to main content
Guidance for scaling platform components as traffic grows.

​
Horizontal scaling

  • WebSocket Gateway: ~1 replica per 1,000-1,500 peak concurrent connections (PCC)
  • Chat API: scale out when average CPU exceeds ~60%
  • Kafka: increase partition counts to raise throughput and parallelism
  • Redis: enable Redis Cluster mode when deployments exceed ~200k MAU

​
Vertical scaling

  • Raise file descriptor limits
  • Tune kernel network queues (somaxconn, netdev_max_backlog)
  • Increase application worker processes and thread pools where supported
  • Example file descriptor tuning:
sudo tee -a /etc/security/limits.conf <<'EOF'
* soft nofile 500000
* hard nofile 500000
root soft nofile 500000
root hard nofile 500000
EOF
echo "DefaultLimitNOFILE=500000" | sudo tee -a /etc/systemd/system.conf
echo "DefaultLimitNOFILE=500000" | sudo tee -a /etc/systemd/user.conf
sudo reboot
ulimit -n

​
When to migrate to Kubernetes

  • MAU exceeds ~200k
  • Multi-region deployments or failover are required
  • Sub-50 ms latency targets are critical
  • Dynamic autoscaling and elasticity are operational priorities