Prometheus metrics

Air Pipe exposes a Prometheus /metrics endpoint and lets actions emit custom metrics. For distributed tracing see OpenTelemetry tracing.

On Air Pipe managed

Managed configs don't serve /metrics. Their request, error and latency metrics are collected and stored for you: see Hosted metrics.

Enable it

Set expose_metrics: true at the config root. /metrics is served in self-hosted modes and is unauthenticated (Prometheus convention — isolate it by network).

name: my-api
expose_metrics: true
interfaces:
  # ...

As of engine 1.7.2, /metrics is served on a dedicated bind, default 127.0.0.1:9090 — not the public API port — so it isn't exposed alongside your API. Configure with --metrics-address (env AIRPIPE__METRICS_BIND):

Value	Effect
127.0.0.1:9090 (default)	loopback — scrape locally or via a sidecar
0.0.0.0:9090	all interfaces — firewall the port
off	disable the endpoint (and the per-request instrumentation)

(Before 1.7.2, /metrics was served on the API port.)

Scrape config:

scrape_configs:
  - job_name: airpipe
    static_configs:
      # engine >= 1.7.2: the dedicated metrics bind (default 127.0.0.1:9090).
      # If Prometheus runs elsewhere, start AirPipe with --metrics-address 0.0.0.0:9090.
      - targets: ["127.0.0.1:9090"]

Built-in metrics

Metric	Type	Labels
airpipe_interface_requests_total	counter	config, interface
airpipe_interface_duration_ms	summary	config, interface, quantile
airpipe_http_action_duration_ms	summary	config, interface, action, status, quantile
axum_http_requests_total	counter	endpoint, method, status
axum_http_requests_duration_seconds	histogram	endpoint, method, status, le
axum_http_requests_pending	gauge	endpoint, method

• Per-route numbers come from the airpipe_* series, labelled by interface. The axum_* series see every config route as one endpoint, /{*catchall}, so use them for totals, status codes and overall latency.
• The two _ms series are summaries: quantiles (0, 0.5, 0.9, 0.95, 0.99, 0.999, 1) over roughly the last minute, plus _sum and _count. Read a quantile directly, for example airpipe_interface_duration_ms{quantile="0.95"}, rather than with histogram_quantile. Summaries from several engines can't be combined into one quantile, so chart them per engine.

Custom metrics

Emit your own metrics from an action with emit_metric. Full fields: EmitMetric reference.

actions:
  - name: RecordSignup
    emit_metric:
      name: signups_total       # must be a valid Prometheus metric name
      type: counter             # counter (default) | gauge | histogram
      value: 1
      labels:
        plan: a|body::plan|

• counter — increments by 1 when value is omitted.
• gauge / histogram — require a value.
• Air Pipe auto-attaches config, interface and action labels to every custom metric, in addition to any labels you set.