Skip to content

debug-perf.md

Latest commit

 

History

History
61 lines (36 loc) · 2.37 KB

debug-perf.md

File metadata and controls

61 lines (36 loc) · 2.37 KB

API: debug and performance

Where you are: docs → reference → api → debug-perf Read this first: api.md See also: metrics.md · subsystems/control-plane.md

TL;DR Five endpoints cover operational introspection: a full debug snapshot (diagnostic dump), live performance metrics with optional baseline comparison, baseline save and delete, and an internal connection-counter endpoint called by sf-agent on loopback for idle-session detection.

Endpoints

GET /api/debug/snapshot

Purpose: return a full diagnostic snapshot (all subsystems). Only registered when a debug collector is attached. The response is heavy (all state + ring buffers); intended for support tickets. Handler: debug.Collector.HandleSnapshot, wired by control/api.go → (*API).registerDebugRoutes.

Response 200: JSON with per-subsystem state and event log.

GET /api/perf

Purpose: return live performance stats (pipeline ns/frame, E2E latency, mix-cycle duration) over a 60 s ring buffer. Optional ?baseline=<name> compares the current window against a saved baseline. Handler: perf.Sampler.HandlePerf.

Response 200: perf.WindowStats with optional delta when baseline= is supplied.

POST /api/perf/baseline

Purpose: save the current perf sample window as a named baseline for regression tracking. Handler: perf.Sampler.HandleSaveBaseline.

Request body: { "name": "1080p60-baseline" }.

DELETE /api/perf/baseline/{name}

Handler: perf.Sampler.HandleDeleteBaseline.

GET /api/internal/connections

Purpose: return active connection counts by transport. Called by sf-agent on loopback (the agent reads process-level state, not tenant-level). Suitable for idle-session detection in a SaaS deployment. Handler: control/api.go → (*API).handleInternalConnections.

Response 200:

{ "webtransport": 3, "srt": 2, "moq": 1, "total": 6 }

Returns zeroed stats when no connection counter is registered.

Related docs