Building a Single-Pane RED Dashboard for Microservices with OpenTelemetry and Semantic Conventions

Three months ago we had 34 microservices and 34 Grafana dashboards, each one hand-built by whoever owned that service at the time. Some had rate/error/duration panels. Some had CPU and memory but no latency. One had a pie chart of HTTP status codes that nobody had looked at in a year. When we had an incident that touched five services at 2am, the on-call engineer had to open five different dashboards, each with different label names (route vs path vs endpoint), different histogram bucket boundaries, and different naming for the same metric.