Ensure your instrumentation library records exemplars on histogram and counter observations with the current trace ID and span ID
Configure the Prometheus scrape endpoint to expose exemplars by enabling the exemplar feature in the metrics SDK
Enable the Prometheus server's exemplar storage by setting the appropriate feature flag at startup
In Grafana, configure the data source link between Prometheus and Tempo so exemplar trace IDs resolve to trace views
Verify exemplars appear on histogram panels in Grafana by enabling the exemplar display option
Known gotchas
Exemplars are only propagated when there is an active trace context at the time of the metric observation; background or batch jobs without spans will not produce exemplars
Prometheus's exemplar storage has a fixed ring buffer size per series; high-frequency metrics overwrite exemplars quickly, reducing the chance of capturing rare slow requests
The Prometheus remote-write receiver on some backends does not forward exemplars by default; check whether the receiving end stores and surfaces them
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp