fix(rivetkit): keep actor event loop alive through shutdown state serialization

[abcxff]

No description provided.

[abcxff]

Stack for rivet-dev/rivet

• #5548 feat(container-runner): log version and git sha at actor sta …
• #5549 feat(container-runner): setup instance memory and cpu usage …
• #5546 fix(rivetkit): keep actor event loop alive through shutdown … 👈
• #5539 feat(container-runner): report actors as crashed on unexpect …
• #5536 feat(container-runner): log unexpected platform SIGTERM as a …
• #5530 fix(rivetkit-core): cancel driver alarm before sqlite teardo …
• #5531 feat(container-runner): keep instance warm instead of self-e …
• #5533 chore(container-runner): default stop grace to 10s
• #5528 fix(rivetkit): log and surface actor startup failures
• #5529 fix(rivetkit): treat empty actor input bytes as absent
• main

Get stack: forklift get 5546
Push local edits: forklift submit
Merge when ready: forklift merge 5546

_{change losmyvnt}

[claude]

Review

Re-verified against rivetkit-core's shutdown sequence (ActorTask::run_shutdown in rivetkit-rust/packages/rivetkit-core/src/actor/task.rs); the fix is correct and this comment supersedes my prior pass.

Root cause confirmed: run_shutdown calls save_final_state() (which sends ActorEvent::SerializeState) before close_actor_event_channel(). Previously, handle_actor_event returned Ok(true) right after handling RunGracefulCleanup, breaking the run_actor loop and dropping the Events<A> receiver. When core then tried to send the follow-up SerializeState event, the enqueue failed (send_actor_event -> not_ready), and core fell back to self.ctx.save_state(Vec::new()).await, discarding any state written by on_sleep/on_destroy. The new code correctly lets the loop keep running until core itself drops the sender via close_actor_event_channel() (task.rs:1494-1497), matching the pattern already used by the NAPI runtime (napi_actor_events.rs, which never terminates its dispatch loop on RunGracefulCleanup either).

I also checked whether this can deadlock if run_shutdown's inner block returns early (e.g. save_state fails before reaching close_actor_event_channel()): it cannot, because ActorTask::run takes self by value and returns right after, so actor_event_tx is dropped when the task struct itself goes out of scope regardless of the early return. No hang risk.

Test coverage: run_actor_serializes_state_after_cleanup is a solid regression test. It would fail under the old code since tx.send(SerializeState) happens after the loop already exited and dropped the receiver. Good that it asserts on the actual on_sleep-written log entry rather than just "no error." The mechanical drop(tx) additions across the other lifecycle tests are consistent and necessary now that the loop no longer self-terminates on RunGracefulCleanup; every request_sleep/request_destroy call site got the corresponding drop(tx).

Minor (non-blocking) nit: handle_actor_event's Result<bool> return and the should_stop/break in run_actor (start.rs:249-252) are now dead code. No match arm returns Ok(true) anymore, so should_stop is always false and the loop only ever ends via events.recv_raw() returning None. Worth collapsing to Result<()> (or a one-line comment noting it is kept for a future early-stop case) so the dead branch does not linger.

Scope is minimal and confined to rivetkit-rust/packages/rivetkit/src/start.rs, consistent with this crate's role as a thin wrapper. No bugs, security, or performance concerns found. Nothing blocking.