zacheryasc/swactor
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
swactor/CLAUDE/notes/progress.md
Developer 473999d1df feat: actor watching — local death notifications 
Add watch/unwatch API to the actor system so actors can monitor each
other's liveness. When a watched actor dies (panic or stop), watchers
receive an ActorExited notification via on_actor_exit().

- ExitReason enum (Stopped, Panicked, NodeDown) and ActorExited struct
- ContextInner::watch()/unwatch() + Ctx typed wrappers
- ActorInterface::on_actor_exit() default method (system message fallback)
- WatchRegistry in worker with bidirectional tracking
- Death notification dispatch as phase 5b in tick_once
- Runtime-level watch for external callers
- 10 behavioral tests in tests/watch_api.rs
- Design documents for OS features in docs/os-design/

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-13 07:25:27 +00:00

1.9 KiB
Raw Blame History

Progress

Completed

Feature 1: Actor Watching (local only) — docs/os-design/01-actor-watching.md

  • Added ExitReason enum (Stopped, Panicked, NodeDown) and ActorExited struct to src/actor.rs
  • Extended ContextInner trait with watch()/unwatch() methods
  • Added Ctx::watch(target) and Ctx::unwatch(target) typed API
  • Added on_actor_exit() default method to ActorInterface trait
  • Updated AnyActor::handle_any with system message fallback (tries ActorExited after Incoming)
  • Implemented WatchRegistry in src/worker.rs (bidirectional HashMap tracking)
  • Integrated death notification dispatch as phase 5b in tick_once
  • Implemented watch/unwatch on Runtime's ContextInner impl
  • Added watch_registry field to TickContext in src/delivery.rs
  • 10 behavioral tests in tests/watch_api.rs — all passing
  • All 53 tests pass (42 existing + 10 new + 1 doctest)
  • All 134 distribution crate tests pass

Next Steps

  1. Command Interface — docs/os-design/04-command-interface.md
    • Create crates/command/ crate with CommandHandler trait, CommandRouter, CommandRequest/CommandResponse types
    • Extract existing cmd_* functions from crates/runtime-dashboard/src/investigate.rs into CommandHandler impls
    • Add built-in read commands (help, overview, nodes, workers, actors)
    • Add built-in write commands (spawn, stop, send, drain, shutdown)
    • Write behavioral tests
  2. Cluster Registry — docs/os-design/02-cluster-registry.md
  3. Node Capabilities — docs/os-design/03-node-capabilities.md
  4. Remote Watching — extends 01-actor-watching with wire protocol
  5. Supervision — docs/os-design/05-supervision.md

Open Questions

  • Command Interface: Should custom actor commands (via Ctx::register_command()) be deferred to a later PR?
  • Should the dashboard migration happen in the same PR as the command crate creation?

Blockers

  • None