CourseAI System Design: The Complete Course · Module 11: Reliability · part 82 of 107
Part 82 · Module 11: Reliability

Incident: Retries kept a recovered service down

16 min read·
Module 11 · Incident file · About 35 minutes

What happened

CASE STUDY OpenAI, 3 February 2026 (PST) 12:12 errors begin about 13:01 gate disabled 13:03 most regions back 15:05 last region back one region: two more hours delayed propagation and a retry feedback loop in that region
The trigger was off within an hour (amber), most regions recovered at 13:03 (green), and one region took two more hours (red).

On 3 February 2026 OpenAI had "a significant service disruption that impacted login functionality and ChatGPT conversation availability", beginning at 12:12 PM PST. Its status page write-up lists the impact: elevated errors for ChatGPT conversations "across all plan types", login success rates that "dropped significantly at peak impact", fewer accounts created, and Free and Go tiers hit hardest.

The trigger was fixed quickly. "Once disabled, services immediately began recovering at around 1:01 PM PST", and most regions were mitigated by 1:03 PM. One region took until 3:05 PM. The write-up gives two reasons for that region's slow recovery: delayed configuration propagation, and a feedback loop made of retries. It gives no count of users affected and no cost.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.

Buy this course · ₹399See what is inside

Or open everything with a plan