Unit 4: OpenTelemetry — distributed traces across Pilot tunnels

[TeoSlayer]

Scope

Distributed tracing across agent-to-agent communication. One trace ID follows a request across multiple Pilot hops.

Deliverables

• Trace context propagation in Pilot packet headers (W3C Trace Context format)
• Span creation per tunnel hop: source daemon → tunnel → destination daemon
• Span attributes: source address, destination address, port, encryption status, RTT, packet loss
• Metrics export: connection counts, bytes transferred, retransmission rates
• OTLP exporter: send traces/metrics to any OpenTelemetry Collector
• Jaeger/Zipkin compatibility via OTLP

Files

• pkg/protocol/packet.go — trace context in packet header (optional extension)
• pkg/daemon/telemetry.go — span creation, metrics recording
• tests/ — trace propagation tests

Priority: MEDIUM

[TeoSlayer]

Context: Without distributed tracing, debugging a multi-agent pipeline is guesswork. OpenTelemetry is vendor-neutral — traces can go to Jaeger, Zipkin, Datadog, Honeycomb, or any OTLP-compatible backend. Trace context in the packet header means the trace follows the data through the tunnel without application-layer changes.