Code Room
On-callHard
Question
Under a big launch your service autoscales from 60 to 150 instances. The instances boot and pass their own health checks, but the load balancer only ever sends traffic to ~100 of them; the rest stay healthy-but-idle and your effective capacity plateaus while latency climbs. The target group's registered-target count is pinned at a flat 100 no matter how many instances run, and the LB control plane logs show registration requests being rejected. There was no deploy to the LB. How do you triage and mitigate the launch?
What a strong answer looks like
Stop the bleeding first (mitigate), then form hypotheses from real signals. Separate root cause from symptom, communicate status as you go, and close with what prevents a repeat.
Learn the concepts
Loading whiteboard…
Run or narrate your approach, then ask the coach.