Netflix’s System Design Interview Was Dirty. I Wasn’t Ready.

Midway through my answer, they killed an entire AWS region and watched.

The video is still there. The region that grants permission to play it is gone.

Your video can survive a region outage while the permission to play it doesn’t. Licence leases, retries with jitter and active-active failover, traced from one bad whiteboard.

I spent my whole prep making a streaming design bigger. Netflix spent the interview switching parts of it off.

Halfway through my playback design, the interviewer put a red X through one of my boxes and said:

“us-east-1 is gone. Forty percent of your viewers are mid-movie. What do they see?”

Then she stopped talking. No hint, no follow-up, just watching me trace the damage on my own diagram. That’s what felt dirty: she changed the problem mid-answer and let me find the holes myself. Looking back, it’s the most Netflix question there is, and I hadn’t prepared for it.

For Netflix, losing a region is a drill

Netflix has written publicly for years about breaking its own infrastructure on purpose. Chaos Monkey goes around killing instances in production. Chaos Kong goes bigger and fakes losing an entire AWS region, which is why their control plane runs active-active across…