Part 3 of a series on building a Maglev load balancing system with VPP, focusing on the observability layer needed to make the system manageable. The author details three components: vpp-maglev exposing health state and VPP dataplane counters via Prometheus; nginx-ipng-stats-plugin, a custom nginx C module that attributes each request to the Maglev frontend that delivered it using IP_PKTINFO instead of SO_BINDTODEVICE (to preserve Direct Server Return); and nginx-logtail, a Go pipeline that ingests per-request log lines via UDP, maintains top-K ranked tables across six time windows, and exposes results via gRPC and a server-rendered dashboard. The post covers GRE6 tunnel setup, MSS clamping, lock-free counter design, high-cardinality DDoS-safe logging, aggregator backfill on restart, and Prometheus alerting rules for detecting silent Maglev frontends or dead VIPs.
