Replace ractor_supervisor with custom SessionActor for per-actor-type supervision

[devin-ai-integration]

Replace ractor_supervisor with custom SessionActor supervision

Summary

Replaces the ractor_supervisor::Supervisor (RestForOne strategy) with a custom SessionActor that implements ractor::Actor directly, enabling per-actor-type supervision. This is the core fix for the bug where recording stops when STT (transcription) fails.

Key behavioral change: When ListenerActor dies (STT failure), the session enters "degraded mode" — recording and audio capture continue uninterrupted. Previously, RestForOne would kill RecorderActor too.

What changed:

• session/mod.rs: Replaced Supervisor + ChildSpec pattern with SessionActor implementing Actor trait. handle_supervisor_evt routes events by child type: Listener death → degraded mode; Source/Recorder death → restart with limits (3 in 15s), then meltdown. Graceful shutdown is handled via SessionMsg::Shutdown which stops children in order (Recorder first, with wait for WAV finalization, then Source/Listener) and sets shutting_down flag to suppress supervision restarts during teardown.
• events.rs: SessionLifecycleEvent::Active now carries an optional DegradedError to notify the frontend of degraded state.
• root.rs: Updated Active event emission to pass error: None. stop_session_impl now sends SessionMsg::Shutdown to the SessionActor instead of directly stopping children by name. ActorTerminated handler now parses DegradedError from the session stop reason and propagates it to emit_session_ended.
• Cargo.toml: Removed ractor-supervisor dependency.

Updates since last revision

Addressed review findings:

1. Restart limit checked before spawn: Moved record_restart() to before Actor::spawn_linked() in try_restart_source and try_restart_recorder. Previously the actor was spawned first, and if the limit was exceeded the new actor was already running but never stored — an orphaned actor leak.
2. 50ms backoff delay before respawn: Added tokio::time::sleep(50ms) before each spawn_linked in the restart paths. Without this, rapid respawns hit name collisions in ractor's actor registry (the old supervisor had a 500ms backoff).
3. Root propagates DegradedError on session end: root.rs ActorTerminated handler now parses DegradedError JSON from the session actor's stop reason and passes it to emit_session_ended, instead of always passing None.
4. #[derive(Debug)] on SessionMsg: Added missing derive required by ractor's Message trait.

Review & Testing Checklist for Human

• 50ms backoff may be insufficient: The old code used 500ms. Verify 50ms is enough for ractor's registry to deregister actor names, especially under load. If restarts still fail with name collisions, increase the delay.

• identify_child uses cell-ID matching: When a child terminates, its cell is cleared from state, then restart is attempted. If the new actor dies quickly and the old cell's termination event arrives late, identify_child won't match it (returns None → silently ignored). Name-based identification would be more robust but is lower priority.

• ActorRef<SessionMsg> cast from ActorCell: In stop_session_impl, let session_ref: ActorRef<SessionMsg> = supervisor.clone().into() performs an unchecked type cast. This relies on state.supervisor always being the SessionActor cell. Confirm ractor's into() behavior — if the cell is somehow not a SessionActor, the cast(Shutdown) would silently fail or panic.

• Test manually: Start a recording session, then simulate STT failure (e.g., disconnect network, use invalid API key). Verify:

  • Recording continues (WAV file keeps growing)
  • Frontend receives Active event with error field populated
  • Session can still be stopped gracefully (no double WAV finalization, no spurious restarts logged)

Notes

• The listener_degraded field in SessionState is set but never read — it's placeholder state for future use.
• stop_actor_by_name_and_wait in lifecycle.rs is now unused (compiler warning). Can be removed in a follow-up.
• Frontend TypeScript types will auto-regenerate via specta, but frontend code doesn't yet display degraded mode info.
• cargo check -p tauri-plugin-listener passes.

Link to Devin run: https://app.devin.ai/sessions/ab6b741af4114e40ba11334c08eaf56f
Requested by: @yujonglee

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[netlify]

✅ Deploy Preview for hyprnote canceled.

Name	Link
🔨 Latest commit	93a3b73
🔍 Latest deploy log	https://app.netlify.com/projects/hyprnote/deploys/69903fe65de2630008fb991c

[netlify]

✅ Deploy Preview for hyprnote-storybook canceled.

Name	Link
🔨 Latest commit	93a3b73
🔍 Latest deploy log	https://app.netlify.com/projects/hyprnote-storybook/deploys/69903fe68d34880008aff1a6

[devin-ai-integration]

Devin Review found 1 potential issue.

View 6 additional findings in Devin Review.

[devin-ai-integration]

Devin Review found 1 new potential issue.

View 9 additional findings in Devin Review.

[devin-ai-integration]

Devin Review found 3 new potential issues.

View 12 additional findings in Devin Review.

[devin-ai-integration]

Devin Review found 1 potential issue.

View 5 additional findings in Devin Review.

[devin-ai-integration]

Devin Review found 2 new potential issues.

View 9 additional findings in Devin Review.