Telemetry.Metrics definitions for RpcLoadBalancer.call/5, RpcLoadBalancer.cast/5, and node-selection events.

Consumers register these in their telemetry supervisor — e.g. through PrometheusTelemetry — to expose rpc_load_balancer.* series.

# in your_app/application.ex
children = [
  {PrometheusTelemetry,
   exporter: [enabled?: prod?()],
   metrics: [
     RpcLoadBalancer.Metrics.metrics(),
     PrometheusTelemetry.Metrics.VM.metrics()
   ]}
]

Emitted events

• [:rpc_load_balancer, :rpc, :start]
• [:rpc_load_balancer, :rpc, :stop] — measurement :duration (:native)
• [:rpc_load_balancer, :rpc, :exception]
• [:rpc_load_balancer, :node_selected] — count: 1, members_count: pool_size
• [:rpc_load_balancer, :node_selected, :empty] — count: 1

Metadata

RPC events:

• :type — :call or :cast
• :node — target node atom
• :module — inspect/1'd module name (string, low-cardinality)
• :function — function name atom
• :load_balancer — load balancer name atom, or nil for direct calls

• :status (stop only) — :ok | :error | atom() mapped from ErrorMessage code

Selection events:

• :algorithm — full algorithm module atom (e.g. Random, RoundRobin)
• :load_balancer — load balancer name atom
• :node — selected node atom (omitted on :empty event)

Prometheus series

• rpc_load_balancer.rpc.request.start.count
• rpc_load_balancer.rpc.request.stop.count
• rpc_load_balancer.rpc.duration.milliseconds
• rpc_load_balancer.node.selected.count
• rpc_load_balancer.node.selected.empty.count
• rpc_load_balancer.node.pool_size