Struct OtelSink 

pub struct OtelSink { /* private fields */ }

TelemetrySink that exports to OpenTelemetry providers.

Implementations

impl OtelSink

pub fn new( tracer_provider: SdkTracerProvider, meter_provider: SdkMeterProvider, logger_provider: SdkLoggerProvider, ) -> Self

Wrap already-built providers. This is the seam the tests use (with the SDK’s in-memory exporters); OtelSink::from_config is the OTLP path.

pub fn from_config(cfg: &ObservabilityConfig) -> Result<Self, String>

Build the OTLP HTTP/protobuf export pipeline from config + OTEL env fallbacks. Constructs a blocking HTTP client - call from a plain thread, not a tokio runtime thread (crate::build_sink does this).

pub fn tracing_log_layer(&self) -> LogLayer

A tracing-subscriber layer that forwards the process’s own tracing events into this sink’s OTLP logs pipeline (daemon-level log export).

These records correlate by resource attributes only - the daemon’s tracing events fire outside any run’s spans, so they carry no trace ids; the per-run TelemetryEvent::Log records are the correlated ones. Filtered to INFO, with the OTel/HTTP stack’s own targets silenced: an export failure that logged through this bridge would otherwise generate more exports. The filtering is done inside TargetFilteredBridge rather than Layer::with_filter because a Filtered layer swapped into a reload slot after subscriber construction has no registered FilterId and panics.

Trait Implementations

impl TelemetrySink for OtelSink

fn emit(&self, event: TelemetryEvent)

Record one event.

fn observe_lanes(&self, health: LaneHealth)

Record one daemon-wide health sample. Default: ignore it, so a sink that only cares about runs needs no changes.

fn observe_providers(&self, down: &[ProviderHealth])

Record which providers are currently out of service, sampled on the same re-drive tick as TelemetrySink::observe_lanes. Default: ignore it, so an existing sink keeps compiling unchanged.

fn force_flush(&self)

Flush any buffered export before shutdown. Default: nothing buffered.