
Scaling Strategies and Why Auto-Scaling Fails

Live streaming at world-record scale with Ashutosh Agrawal (ex-Jio / Disney+ Hotstar)
Topic 8 of 11
Published Feb 12, 2025 · Clip 44:32–49:50 (5:18)
The guest explains that compute scales horizontally while databases are preemptively scaled pessimistically to avoid data-layer changes under load. Auto-scaling fails for this workload because sudden surges - like the post-innings-break rush - exceed its response speed and it lacks awareness of user journeys: when a key batsman gets out, users hit the back button, causing a spike on homepage APIs that were seeing little traffic. Instead, the team built custom scalers that scale every service against a single golden metric - concurrent users - with triangulation to verify the concurrency number's accuracy.