Episode summary: Every media and streaming platform plans for scale around the obvious bottleneck: video delivery and CDN capacity. This conversation is about the less obvious places infrastructure actually breaks first, authentication systems under concurrent login spikes, recommendation services that were never load-tested against a live event audience, and billing systems that quietly fall over during a subscription surge.
What we cover
- Why live events, not steady-state viewership, are where most streaming infrastructure actually gets tested for the first time.
- The specific systems adjacent to video delivery that tend to fail first: auth, recommendations, and billing.
- How teams should actually load-test for a live event spike, versus the steady-state testing most infrastructure teams default to.
- What changes architecturally once a platform has been through one real scaling failure and is designing for the next one.
A few things that stood out
The recurring theme: teams that had already been through one major live-event failure talked about infrastructure completely differently than teams that had not. Not more cautious in a vague sense, specifically obsessive about identifying every system that touches the critical path during a spike, including ones that seem unrelated to video delivery, like customer support tooling that falls over under a spike in concurrent support tickets.
The other point worth noting: the fix in most of these post-mortems was not more infrastructure capacity. It was better isolation, making sure a spike in one system (support tickets, say) could not cascade into degrading an unrelated system (like authentication) that happened to share a database connection pool.
This episode is aimed at engineering leaders at media and streaming companies planning capacity for their next major live event or subscription push.