Question

You cut over a JVM trading-data API from the blue environment to green via a 100% load-balancer switch at market open. The instant the switch lands, p99 latency on green spikes to 12s for ~90 seconds, error rate hits 15% (mostly client-side timeouts), then everything settles to normal and stays healthy. Green had been deployed and idle for 30 minutes. CPU on green shows a brief 100% spike correlated with the latency. No code differs between blue and green. Triage and mitigate.