fix(rivetkit-core): decrement active actor metrics

[NathanFlurry]

Stack Context

This stack is improving RivetKit runtime metrics and serverless observability.

What?

• Marks actor metrics stopped when ActorTask terminates, with Drop as an idempotent fallback.
• Converts self-referential actor context callbacks to weak captures so callbacks do not retain actor contexts.
• Adds a regression test that verifies ActorTask callback setup does not keep the actor context alive.

Why?

The active actor gauge could appear to only increase because metric decrement was tied to ActorContext drop, and context-owned callbacks captured the same context strongly. Stopping the metric explicitly on task termination and removing the self-retaining callback captures makes rivetkit_actor_active_count reflect task lifetime instead of leaked context lifetime.

Validation

• cargo check -p rivetkit-core
• cargo test -p rivetkit-core request_save_hook_does_not_retain_actor_context
• cargo test -p rivetkit-core actor_active_count_tracks_metric_lifetime

[NathanFlurry]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• TODO: ktichen sink counter latency #5051
• fix(rivetkit): use engine actor stop threshold for shutdown #5049
• DO NOT MERGE: serverless restart race condition #4997
• fix(rivetkit): exit pid1 after signal shutdown #5045
• fix(rivetkit-core): decrement active actor metrics #5044 👈 (View in Graphite)
• fix(envoy-client): dont abort drain after 20s #5043
• temp: patch wasm-pack to point to latest binary #5037
• fix(rivetkit): require engine ping for health #5022
• feat(rivetkit): add current actor metrics #5020
• feat(rivetkit): expose low-cardinality metrics routes #5019
• docs(serverless): document timeout tuning #5018
• feat(kitchen-sink): add load testing harness #5017
• fix(rivetkit): expose client actor metadata #5016
• fix(runtime): avoid bounded dispatch channel stalls #5015
• fix(sqlite): fence remote requests by generation #5014
• feat(rivetkit-core): expose metrics endpoint #5013
• WIP: counter latency monitor #5011
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[claude]

Code Review fix rivetkit-core decrement active actor metrics. Good fix for a real leak. The root cause strong self-captures in callbacks preventing ActorContext from ever dropping was subtle, and the three-pronged approach explicit stop at task exit, idempotent AtomicBool guard, weak captures is solid. Overview: The PR addresses two related issues. 1. ActorMetrics was only decremented in Drop, but self-referential strong captures in callbacks kept ActorContext alive indefinitely, so Drop never fired. 2. registry/mod.rs had get_protocol_metadata missing () referenced as a field, not called as a method and was using a hardcoded SHUTDOWN_DRAIN_TIMEOUT instead of the protocol-negotiated stop threshold. Bugs: registry/mod.rs fix is significant - get_protocol_metadata without () means the method was never actually called; stop_threshold would have fallen back to the hardcoded default on every shutdown. The internal_keep_awake weak-upgrade timing: The weak reference is resolved synchronously when the callback fires outside the async move block, so the captured Option inside the pinned future holds a strong reference for the full duration of internal_keep_awake_task. Dual call sites in run(): record_actor_stopped is called in both the early-exit branch and the shutdown-complete branch. The AtomicBool guard makes the subsequent Drop call a no-op. Correct. Design: Silent no-ops in metric callbacks - Five of the six callbacks silently do nothing if the context weak ref is dead. These are metrics/observability callbacks, so the silent no-op is defensible but a brief comment would help readers. ActorRuntime::NotConfigured in internal_keep_awake is the one callback that returns an explicit Err, consistent with fail-by-default guidance. Good. Ordering::AcqRel on swap is correct. Test Coverage: Missing actor_active_count_tracks_metric_lifetime test listed in validation steps but absent from diff. request_save_hook_does_not_retain_actor_context is present. Gauge-level coverage gap: the test does not verify record_actor_stopped decrements rivetkit_actor_active_count. Summary: Core approach is correct. Main items: 1. Add or confirm actor_active_count_tracks_metric_lifetime test. 2. Optional brief comments on silent no-op branches. 3. registry/mod.rs fixes are good - note them in the description.