Files
Piik/docs/status.md
T
AAA 0af8e90284 feat(observability): expose local media path evidence (#166)
Add local-only SFU, encoder, A/V sync, and selected candidate evidence without changing media routing or wire contracts.
2026-08-22 04:23:55 +08:00

9.2 KiB

Current Status

Last updated: 2026-08-22

Phase

Production is exact 083c111437485064a766368944a7938b79e07365 at https://share.bonfire.icu, release 083c11143748, wire screener-v5. SQLite v3/five rooms keeps SHA-256 aef724ae52c4107be8ec8791e72f8dcaa11b0ec849fb432d022e2cfb9cfe0974, owner/mode screener:screener/0600. screener, LiveKit, coturn, and nginx are active; local/public health are green and screener has NRestarts=0. The previous 27ad90ddf9f8 release is the rollback target. Ordinary ICE is STUN-only; limits are SFU roots <=2, one peer-selected attempt excluding Host ingress, and selected UDP TTL 120. No browser/media/SFU/TURN/performance canary ran; exact-16f6eab Host-ingress proof is separate. Stable SFU, route/transport truth, preview notice, entry/nickname UI, selected-edge lease authority, and corroborated bad-relay demotion are live. MAX_PEER_RELAY_DOWNSTREAM_EDGES is unset, so Web endpoints use the parameterized default of two. Native is source-only.

Execution Principle

Parallel flagship work; use minimum relevant checks and rollback proof. Target-device/production evidence drives performance decisions; ordinary-PC synthetic runs prove only correctness/interoperability. Git owns history; delete no-consumer layers.

Current Snapshot

  • Capture precedes rooms; source/quality changes and AV pause preserve peers. Defaults are balanced, automatic codec, and 128 kbps screen audio. Clarity/fluid plus H.264/VP8 stay explicit. The source candidate adds 64/128/256 kbps audio presets under Share advanced settings; codec and audio quality are locked during a share. Browser degradation, audible route proof, and Host SFU A+B remain open.
  • Deployed screener-v5 admits private fragment grants directly; code-only Viewers need site access, and private rooms additionally need their password. Credentials remain unstored. Web presence fields remain opt-in and Native does not subscribe to them.
  • Deployed UI has nickname editing, room entry, Viewer roster, duplicate-only ID suffixes, pause notice, and session-bound SFU first-media truth that clears on track/route/session loss. P2P/SFU/TURN labels require evidence. Guarded relay evidence reaches opted-in Web Host rosters without Host-local transport claims or false parent-edge proof.
  • Native v5 is source-only (memory rooms, 300s reclaim); it does not subscribe to Web presence, accepts the shared absolute three-child bound, and fails unsupported SFU/selected ingress boundedly. Production uses the default endpoint cap of two.
  • Production roots/Web relay <=2; Viewer upstream1. Source parameterizes endpoint cap at default2/range1-3 while SFU roots remain <=2; production stays2. One-root healthy-SFU MBB is deployed/media-unverified; source defers one cooldown ready for expiry reproof. Capacity 0 -> 1, multi-root and browser gates remain open. Room 1 is historical; HTTPS/WSS is TLS/TCP.
  • Source-only connection details read selected-pair addresses from the current page's local stats, hide them when incomplete, and never add them to relayed evidence, signaling, server state, logs, or persistence.
  • After an answer, a generation-bound 15s initial-connect deadline enters ICE restart; success/replacement/disposal cancels it. It is deployed but not mobile-verified.
  • C+B is uncalibrated. Production compares relay-child FPS with fresh parent input; two corroborated children can pause one Viewer relay for 30s and trigger peer-only reassignment. No candidate keeps the old edge, and admission rescue remains deterministic.
  • Deployed screener-v5 selected-TURN authority separates immutable grant revision from current route authority. Unrelated revisions may carry only the same in-budget edge, while topology/session/share/advertised-budget loss revokes it before active authority. Browser/media proof remains open.

Verified Evidence

  • Chrome 151 synthetic 1/3/5/8 and 720p30 runs kept bounded fanout/decoding. A five-Viewer cap2 run had one relay serve two children (+80 decoded frames/~1.1 MB each), all decoded and no fatal. Functional only.
  • Chrome 151 + local LiveKit 1.13.5/client 2.22.0 SFU/UDP passed after q,f -> q,h: one root, decoded 3 -> 23, rendered 26, Host edge one, clean leaves, no TURN. Functional only.
  • Host A+B/P2P Viewer C are sanitized, generation-bound/read-only. Source adds Host-local current SFU h snapshot, cleared on identity change; SFU C stays fail-closed/unverified.
  • The a11a73d built-in TURN canary is rejected: local allocation passed, but direct Host/Pion Viewer failed before forced relay. Full rollback restored STUN-only client ICE and zero allocations.
  • Chrome 151 loopback proves H.264 negotiation/decode interoperability, not performance. Production orders H.264 first and keeps browser repair/fallback codecs.
  • Native WGC/MF used no browser capture/encoder. One Chrome Viewer rendered 203 1280x720 frames and received 500 Opus packets; process/LUID-correlated VideoEncode was nonzero, with no fatal/encoder errors.
  • Native Win11 audio is source-only/default-off: target isolation was 4018x and one Viewer got 495 Opus packets. Download, game sync, Win10, and other routes remain open.
  • Local gate proves pre-DOM fragment clearing, room-scoped session isolation, no site/grant transport-log hits, and no raw room-password SQLite sentinel; rotate/revoke tests pass. Headless loopback only.
  • Native wire/session tests enforce at most two authoritative children, no presence-created edge, stale-revision ignore, and one bounded unsupported-route failure per revision.
  • Production caps pending/answered peer-selected at one per room; Host ingress is independent. Focused admission/release/rollback/STUN-only tests pass; earlier Chrome proves active Host-ingress function only.
  • Exact 083c111 deployed with the endpoint-cap environment unset, so the parameterized default remains two; local/public health are 200, all four services are active, and screener reports zero restarts.
  • Exact 6ccb516a passed typecheck/build/hygiene and 32/489; deploy gates preserved artifacts, config/nft/DB, zero shared inodes and four zero-restart services. No browser/media/SFU/TURN/performance canary ran.

Unverified Boundaries

  • The 083c11143748 cutover has service/health proof but no browser/media/SFU/TURN/performance canary. Earlier 16f6eab proof covers Host ingress only; healthy reselection, audible quality, last mile, external cohorts, resources/performance and game blur remain open.
  • Mobile is Viewer-only. Its persistent audible media element keeps media/signaling active while the page lives; iOS lock-screen, reclamation and background reconnection remain device gates. AirPlay/system mirroring is local output; live WebRTC srcObject has no portable in-app TV-output contract.
  • Silent partitions can wait 30 to 60 seconds for heartbeat detection before the default 5-second grace; this remains unverified.
  • Production startup blur remains open. One local Chrome synthetic same-balanced A/B kept route/PC/SSRC/track and coincided with 720p -> 1080p without loss/freeze, but natural ramp prevents causal, production, or SFU proof.
  • ADR-0006 has one Viewer proof; two-edge/FIFO, hardware, endurance, public/downloaded-package, and browser-diversity proof remain open.
  • Production fanout is Host2/Viewer2 and rejects child3. Relay resource behavior remains open.
  • Active SFU already publishes q,h; only BWE/resource/zero-child/root gates are unverified. No local per-leaf UDP shaper; resume with Linux tc or public canary, never CDP. Web P2P/SVC shortcuts remain no-go.
  • Access lacks headful/production proof for fragment consumption, rotate/revoke and request/nginx/journal/SQLite leakage; its source gate is loopback-only. No accounts.
  • Split large modules only at proven consumer boundaries, never by file length.

Next Milestone

Next: representative production-room UDP/frame/cap/stop canary; room 1 is history. TURN performance follows later. Recovery: ICE restart -> same parent -> alternate peer -> SFU. Source-only manual recovery keeps the route: peer restarts; SFU reauth rebuilds its subscriber. Next: exactly-two/Dynacast-off zero-child SFU BWE, then evacuation. Native WGC/MF passed one-Viewer hardware; package download, Viewer2/FIFO and game A/V remain. Use representative target-device or production evidence for performance, not an ordinary-PC synthetic gate. Ordinary peers stay STUN-only; initial ingress, peer-selected, and healthy-reselection browser gates remain open. ADR-0004 still needs its resource/quality matrix; the 20-viewer gate precedes a default change. Mobile uses the same capacity and is a compatibility observation. Screen audio defaults to 128 kbps; the source-complete 64/128/256 kbps presets propagate through P2P, browser relay, and SFU and lock for the active share. A/V/jitter-buffer/concealment details are local-only in this observability branch. Verify Host settings, negotiated/observed bitrate, audible quality, synchronization evidence, and route switches.

Blockers And Decisions

ADR-0004/0005 remain No-Go for broad rollout. DNS/TLS, independent secrets, host/provider UDP 7882 and bounded services are deployed, but real external media/device evidence is absent. Shared IP remains smoke-only and cannot approve clean-port migration; the full gate still needs an isolated VM/IP.

  • Initial deployment region and network cohort.
  • Project license and distribution model.