Kakao Analyzes Prometheus Metric Bottleneck in Ingress Nginx Controller

Kakao's Ad Recommendation Team analyzed Prometheus metric collection failures and memory spikes in Ingress Nginx Controller. During high-traffic peak hours, servers operated normally, but metric data was not collected, resulting in 'metric loss.' Analysis revealed a bottleneck in the Unix Socket path transmitting metrics from Nginx to the Controller. The handleMessages function in the Go-implemented Controller slowed down as traffic increased, causing continuous Goroutine accumulation, which was identified as the primary cause of the memory spike. This is Part 1; specific solutions will be covered in Part 2.