Skip to main content
RPC Core’s dashboard shows provider connectivity, circuit state, block freshness, upstream attempt rates, success rates, and latency distributions. These observations belong to a node and its available observation window. An unavailable value is not a measured zero. GET /api/metrics/:chain returns JSON metrics for the node-local public profile. Its provider and method measurements describe upstream attempts; a client request can make several attempts during failover. See the metrics API for its fields, units, and nullable values. GET /metrics exposes exact per-route request totals (lasso_rpc_route_*), bounded sampled routed-completion latency histograms, non-success upstream attempt diagnostics, sampled completion failovers, circuit state and recovery, routing’s own per-transport head lag, WebSocket continuity budgets, and BEAM runtime pressure. The sampled lasso_rpc_requests_total series by chain, provider, method, and outcome occupies at most 4,096 series slots; unknown methods use other, and successful diagnostics above 256/s per profile, chain, and origin are excluded (counted in lasso_rpc_route_detail_sampled_out_total) while the exact route totals are not sampled. Circuit and lag samples scan at most 2,048 configured routes per scrape. Missing lag is unknown, not zero. Scrape each Core node separately and retain an instance label in Prometheus when you aggregate nodes. Restrict /metrics to your collector at the proxy or private network boundary: Core does not authenticate that endpoint, and labels can include configured provider identifiers. Import the versioned Grafana dashboard, and see the Prometheus operator guide for the full metric catalog, labels, scrape setup, and incident queries. The dashboard request tester generates real upstream traffic. Its success, error, and latency counters describe that browser’s test run. Open WebSocket connection counts do not establish that every subscription succeeded; the activity feed records confirmations and errors separately. The request pipeline also emits [:lasso, :rpc, :request, :stop] with a millisecond duration measurement and metadata for chain, method, strategy, provider, transport, result, and failovers. This event describes routed request completion. Locally handled methods, invalid requests, and pre-dispatch failures follow other paths, so it is not a complete ingress counter. An embedding deployment can attach its own telemetry handler and export measurements to its monitoring system. Optional OpenTelemetry tracing adds HTTP, routed-request, and upstream-attempt spans with W3C context propagation and JSON-log trace correlation. Set LASSO_OTEL_ENABLED=true to enable it; tracing is disabled by default and never affects routing results. VM/system metrics are disabled by default. Set LASSO_VM_METRICS_ENABLED=true to collect and display them in the dashboard; Prometheus’s lasso_vm_* BEAM totals export without that flag. Protect metrics and dashboard routes at your ingress when deployed publicly. See the released observability reference for the event schema and metrics API for response examples.

Learned routing

RPC Core v0.7.0 exposes node-local counters for applied ordering changes and hints withdrawn by contradictory success. See provider learning for their names, fixed labels and counting boundaries.