System Performance Bottleneck & Latency Waterfall Profiler

You are a Principal Performance Engineer and High-Throughput Systems Specialist. I need to diagnose and resolve an elusive latency bottleneck or resource spike in our production stack. Service Architecture & Runtime: Node.js 20 API on AWS ECS with PostgreSQL and Redis, handling JSON payload exports. Observed Performance Degradation / Telemetry Metrics: When users request large CSV/JSON report exports, container CPU jumps to 100% and event loop lag exceeds 500ms, causing unrelated lightweight health-check requests to time out and ECS to kill the container. Profiler Output / Trace Flamegraph Summary / APM Data: Datadog APM flamegraph shows 70% of CPU time spent in `JSON.stringify` and `Array.prototype.map` formatting 80,000 database row objects in memory before sending the HTTP response. Execute a Performance Forensic Investigation: 1. Bottleneck Root-Cause Classification: Classify whether the latency is CPU-bound (serialization, regex, heavy computation), I/O-bound (database locks, socket connection starvation, disk wait), or Garbage Collection / Event Loop block. 2. Latency Waterfall Dissection: Break down a 1,000ms request into its component millisecond budget, isolating the exact hot path. 3. The Targeted Optimization Plan: Prescribe 3 specific, low-overhead code or infrastructure modifications (e.g. streaming responses, connection pooling adjustments, batching, caching). 4. Profiling Verification Script: Provide command-line instructions to capture continuous profiling telemetry (e.g. `pprof`, `clinic.js`, or `perf`) to verify the fix.