OpenTelemetry

Outpost exposes key performance metrics via OpenTelemetry. All metrics are exported as histograms.

Setup

You can configure OpenTelemetry Collector to export metrics to your observability platform.

OTEL_SERVICE_NAME=outpost
OTEL_EXPORTER_OTLP_ENDPOINT=http://otel-collector:4317

Signal Configuration

Setting OTEL_SERVICE_NAME enables all three signals: traces, metrics, and logs. Two sets of variables control them: the endpoint variables decide which signals are exported, and the exporter variables decide how an exported signal is written (or disable it with none — an exporter value can turn a signal off, never on).

Endpoints

The endpoint variables decide which signals are enabled:

• One or more per-signal endpoints (OTEL_EXPORTER_OTLP_TRACES_ENDPOINT, OTEL_EXPORTER_OTLP_METRICS_ENDPOINT, OTEL_EXPORTER_OTLP_LOGS_ENDPOINT) with no generic endpoint: only the signals with an endpoint are exported. Signals without an endpoint are disabled, regardless of any exporter setting.
• The generic OTEL_EXPORTER_OTLP_ENDPOINT: all three signals are exported.
• No endpoint at all: all three signals are exported to the SDK's default endpoint.

So exporting a single signal takes nothing but its endpoint — to export only metrics:

OTEL_SERVICE_NAME=outpost
OTEL_EXPORTER_OTLP_METRICS_ENDPOINT=http://otel-collector:4317

Exporters

Accepted values are otlp (the default), console (alias stdout) to print to standard output, and none to disable the signal. Because a signal without its endpoint is already disabled when per-signal endpoints are in use, console/stdout only takes effect with no endpoints or with the generic endpoint. For example, to export everything except logs:

OTEL_SERVICE_NAME=outpost
OTEL_EXPORTER_OTLP_ENDPOINT=http://otel-collector:4317
OTEL_LOGS_EXPORTER=none

Protocol

Each signal's OTLP protocol is resolved in precedence order:

1. OTEL_EXPORTER_OTLP_TRACES_PROTOCOL / OTEL_EXPORTER_OTLP_METRICS_PROTOCOL / OTEL_EXPORTER_OTLP_LOGS_PROTOCOL
2. OTEL_EXPORTER_OTLP_PROTOCOL
3. OTEL_PROTOCOL
4. grpc (the default)

Accepted values are grpc, http, and http/protobuf (http and http/protobuf are equivalent).

Available Metrics

delivery_latency

Latency from event publication to successful delivery at the destination.

delivered_events

Number of events successfully delivered.

delivery_connections

Outbound HTTP connections used for delivery, split by whether the connection came from the idle pool or had to be opened.

A high reused=false share means the idle connection pool is not holding connections long enough to be reused — expected at low traffic, but under sustained load it indicates the pool ceiling is binding.

published_events

Number of events published via the Publish API.

eligible_events

Number of published events that matched at least one destination.

api_response_latency

API endpoint response latency.

api_calls

Number of API calls received.

Note: When self-hosting, CPU, Memory and Disk usage are not exported by Outpost — monitor these via your VM or container runtime provider.