[Rust-KVM]

Requested by 4c · project thread

Summary

Before: the pipeline encoded 4K60 continuously even with nobody connected; idle, rustkvm_app used ~56% of a core, ~46% of it in the encoder's output thread, for frames that were thrown away.

After: encoding pauses while no session is watching and resumes, with a forced keyframe, when a viewer's video track is attached.

How

• pipeline.rs: a valve in front of the encoder, starting closed, in forward-sticky-events mode so caps still reach the encoder and video state, dynamic GOP and snapshots keep working. The screenshot probe moves to the valve's input so snapshots work while paused.
• video.rs / webrtc.rs: attaching a session's video track opens the valve; detaching the last one closes it; a restarted pipeline picks up the current state.
• docs/api.md: rustkvm_video_frames_total now counts only frames encoded for a viewer.

Testing

• cargo +nightly fmt --all (CI)
• cargo +nightly clippy --workspace --release --all-targets -- -D warnings (CI)
• cargo +nightly test --workspace (CI)
• Frontend touched: not applicable
• Tested on an RK3588 device: a build with all three video branches (fast first frame, idle without viewer, H.264 High profile) was deployed to the NanoPC-T6 (health ok, 4K60, no WARN/ERROR at startup). First frame, timed from PeerConnection creation to the first keyframe: cold connects 131 and 101 ms; reconnect after a takeover ×6: 162, 102, 132, 105, 152, 113 ms.

Checklist

• No #[allow(...)] suppressions and no new unwrap() / expect() in production paths
• Hot paths stay zero-copy; periodic loops use MissedTickBehavior::Delay
• Shared protocol types live in rkvm-proto
• RPC method set changed → not applicable
• Module tree or build invariants changed → not applicable

Part of a set of three video PRs (#60 and the H.264 High profile PR); merges onto dev together with them without conflicts.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PbdzY6PfT28goR4BhqdsXR

---

Generated by Claude Code