OpenTelemetry¶
The server and both libraries export OpenTelemetry metrics and traces. Point them at your collector and the same series feed the Snodra dashboards and your own.
Server¶
docker run --rm -p 7000:7000 -p 7001:7001 \
-e SNODRA_SERVER_OTLP_ENDPOINT=http://otel-collector:4317 \
snodra/snodra-server:0.5.0
- Traces: one
hub.routespan per request, parented from the request'straceparent. The forwarded request keeps the original context, so the handler's span is a child of the sender's, not of the server's. Attributes name the type, kind, sender, handler and outcome. - Metrics: OTLP gRPC every 60 seconds by default;
OTEL_METRIC_EXPORT_INTERVAL(milliseconds) changes it. The resource isservice.name=snodra-serverandservice.instance.id=<SNODRA_SERVER_ID>, which becomesjobandinstancein Prometheus. - No traces or metrics endpoint on the server itself. It pushes; your collector receives.
An empty SNODRA_SERVER_OTLP_ENDPOINT disables export, which is what the local quick-start commands
do.
Libraries¶
The Rust library records through the OpenTelemetry API and installs nothing itself. Install the SDK
and a meter provider before building the app, with service.name set to the Snodra service name and
service.instance.id to the instance id the server sees:
let provider = SdkMeterProvider::builder()
.with_resource(
Resource::builder_empty()
.with_service_name(config.service_name.clone())
.with_attribute(KeyValue::new("service.instance.id", config.instance_id.clone()))
.build(),
)
.with_periodic_exporter(MetricExporter::builder().with_tonic().with_endpoint("http://otel-collector:4317").build()?)
.build();
opentelemetry::global::set_meter_provider(provider.clone());
The Kotlin starter uses Micrometer, so the service's normal Actuator and OTLP configuration applies. Both libraries export the same instrument names, so one dashboard works for either.
Signal names¶
| Prefix | Owner |
|---|---|
snodra_hub_* |
the server |
snodra_* |
the client library: outbox, processors, sagas, hub client |
snodra_event_browser_* |
the event browser service |
Dots in instrument names become underscores in Prometheus, and counters gain _total. Gauges from a
stopped instance need last_over_time(...); the shipped dashboards do that.
Dashboards¶
Set SNODRA_SERVER_GRAFANA_URL and the dashboard links to Grafana. The server's own dashboard stays
the live view; Grafana keeps history and the cross-service views. The queries and the full metric
reference are in Observability, and the signals worth alerting on are
in Monitoring.