Request fan-out¶

TL;DR. Counts distinct downstreams per inbound request and warns the trace when fan-out crosses your threshold. Catches chatty endpoints before they topple a downstream.

Some endpoints accidentally call thirty downstream services because of an unhelpful loop or a chatty serialiser. The first you find out is when one of those downstreams falls over and the noisy endpoint is the suspect.

Pulse counts the distinct downstreams per inbound request, so chatty endpoints show up before they become chatty incidents.

What you get¶

histogram_quantile(0.95,
  sum by (endpoint, le) (rate(pulse_request_fan_out_bucket[5m])))

The 95th-percentile fan-out per endpoint. An unexpected jump means a code change just made a previously-tight endpoint chatty — usually before it shows up as a downstream incident.

Turn it on¶

Nothing. Enabled by default alongside the dependency health map, which it shares the underlying outbound classification with.

What it adds¶

Metric	Type	Tags	Meaning
`pulse.request.fan_out`	Distribution summary	`endpoint`	Distinct logical dependencies called during the request

When to skip it¶

pulse:
  dependencies:
    fan-out:
      enabled: false

Conditional gating¶

Fan-out shares pulse.dependencies.enabled-when with the per-call RED metrics — one block, two consistent skips. See Dependencies → Conditional gating.

Source: io.github.arun0009.pulse.dependencies · Status: Stable since 1.0.0