Prometheus Plugin Memory Bloat
warningKong's Prometheus plugin is a memory hog consuming +500MB per worker and adding 2ms latency per request. It periodically scans all entities, causing CPU spikes that can peg CPU at 100%.
Monitor per-worker memory usage and CPU utilization. If Prometheus plugin is enabled and memory usage exceeds baseline by 500MB+ per worker, or if periodic CPU spikes occur without traffic changes, the plugin is the culprit. Check worker thread CPU usage with 'top -H -p $(pgrep nginx)' during spikes.
Disable Prometheus plugin if not actively used. If metrics are required, export them via StatsD or use global response transformer to disable Kong-Debug headers in production. Consider dedicated observability data planes to isolate Prometheus overhead from production traffic. Use StatsD plugin as lighter alternative.