Three Logging Practices That Bridge the Gap Between Green Dashboards and Real Reliability

A software engineer with years of JVM backend production experience argues that healthy-looking infrastructure dashboards can mask real user-facing failures, a pattern they documented through repeated incidents. The core fix proposed involves logging three specific signals per request: a correlation ID that persists across all services, meaningful distributed trace spans tied to business operations, and SLOs based on customer-experienced outcomes rather than machine metrics. On the technical side, the author recommends assigning correlation IDs at the network edge using a servlet filter, storing them in thread-local MDC context, and critically, clearing that context after each request to prevent ID leakage across reused threads. For distributed tracing, the author cautions that auto-instrumented spans from tools like OpenTelemetry are insufficient on their own, and that manually defined spans around meaningful business operations — tagged with relevant context like account IDs — are far more useful during incidents. The overarching argument is that closing the gap between infrastructure health and system reliability requires no new tooling, only more intentional logging strategy.