Cloud Application Slowness: When Every Team Says ‘It’s Not My Problem’

This article details a production outage in a retail ERP system after scaling from 3,000 to 10,000 stores, where standard dashboards reported healthy metrics despite widespread service failures. The root cause wasn’t CPU, memory, or bandwidth, but the EC2 instances hitting a packets-per-second (PPS) limit, causing silent packet drops and TCP retransmissions that standard monitoring failed to detect. The incident highlights the importance of cross-layer correlation of metrics – specifically network, application, and database telemetry – to identify issues beyond simple resource utilization and the need for operations teams to own data plane configuration in the cloud.