Define the spanmetrics connector in the connectors block with explicit_buckets histogram configuration listing latency bucket boundaries in seconds
Set dimensions to pull specific span attributes (such as http.method and http.status_code) as metric label dimensions
Add namespace and metrics_flush_interval to control metric naming and emission cadence
Wire a traces pipeline to export to the spanmetrics connector, then wire a metrics pipeline to receive from it
Add a Prometheus exporter to the metrics pipeline and scrape the resulting RED metrics from its endpoint
Known gotchas
The spanmetrics connector generates high cardinality if dimensions include attributes with many distinct values such as user IDs or URLs; prefer low-cardinality attributes
Using exponential histograms instead of explicit buckets requires setting histogram: exponential: max_size in the connector config, not in the SDK
The connector emits metrics only after receiving spans; a cold pipeline with no traffic produces no metrics, which can cause no-data alerts to fire
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