17.51Interview answer

“I'd decompose the latency first. I'd compare end-to-end p95 with Salesforce and other dependency latency, retries, connection saturation, runtime CPU/memory, traffic and recent deployment changes. Then I'd trace representative requests by correlation ID. I wouldn't increase timeouts until I knew which segment was responsible.”

Strong.