fix(rivetkit-napi): key actor runtime state by generation

[abcxff]

No description provided.

[railway-app]

🚅 Deployed to the rivet-pr-5825 environment in rivet-frontend

Service	Status	Web	Updated
website	❌ Build Failed (View Logs)	Web	Oct 3, 2026 at 2:05 am UTC
kitchen-sink	😴 Sleeping (View Logs)	Web	Oct 2, 2026 at 10:19 pm UTC
frontend-cloud	😴 Sleeping (View Logs)	Web	Oct 2, 2026 at 10:19 pm UTC
frontend-inspector	😴 Sleeping (View Logs)	Web	Oct 2, 2026 at 10:19 pm UTC
ladle	✅ Success (View Logs)	Web	Oct 2, 2026 at 10:11 pm UTC
mcp-hub	✅ Success (View Logs)	Web	Oct 2, 2026 at 10:08 pm UTC

[abcxff]

Stack for rivet-dev/rivet

Current stack:

• #5843 fix(envoy-client): keep the ping silence deadline bounded ac …
• #5838 fix(envoy-client): ignore pings from earlier connections in …
• #5823 fix(envoy-client): declare actors lost before the engine can …
• #5824 fix(rivetkit): stop lost actor generations immediately
• #5821 fix(envoy-client): signal lost actor generations synchronous …
• #5822 fix(rivetkit): isolate workflow run state per actor generati …
• #5825 fix(rivetkit-napi): key actor runtime state by generation 👈

Dependencies:

• #5816 chore(release): update version to 2.3.23
• main

Get stack: forklift get 5825
Push local edits: forklift submit
Merge when ready: forklift merge 5825

_{change ruyuonlz}

[claude]

Review

The fix looks correct. Keying ACTOR_CONTEXT_SHARED by (actor_id, generation) stops an older generation's reset_runtime_state and clear from touching the newer generation's JS state. The Rust unit test covers the key behavior directly.

Suggestions

• Test helpers vs. CLAUDE.md: serverless-lost-generation.test.ts defines a custom waitFor that swallows errors and retries every 250ms. CLAUDE.md says not to paper over races with retry loops, and any polling needs a one-line justification. Use vi.waitFor with an adjacent // comment, or await events directly (for example promises resolved from the hooks instead of the hookEvents array). check:wait-for-comments only checks vi.waitFor, so this helper escapes that check.
• Prototype patching: observeRuntimeStateClears monkey-patches NapiCoreRuntime.prototype.actorClearRuntimeState. This is close to the module-level mocking CLAUDE.md prohibits. It is restored in finally, but it breaks if tests run concurrently in one process.
• None generation key: generation() returns None before an envoy is configured, so all such contexts for one actor id share one entry. That matches the old behavior, but a short comment on the ActorContextSharedKey alias would make it explicit.
• generation() naming: it is a thin alias for sleep_generation(). Consider reusing the existing name, or note that they are intentionally the same.
• Test weight: The driver test depends on a TCP reset timing a Lost generation, with a 90s timeout, so it could be flaky in CI. The unit test already covers the invariant. Check whether this test belongs in the fast-driver list, and gate or document it if it is slow.
• Process: The PR body is empty. Add a short bullet list of what changed.

No security or performance concerns. The tuple-key allocation per context construction is negligible.

[the-company-company]

🟠 1 medium · 🔵 1 low

_{Reviewed commit 976d08b.}

[the-company-company]

🟠 Medium · Format the shared-state declaration so Rustfmt passes

The required Rustfmt check fails on this declaration at this head. Apply the formatter’s single-line generic layout (with the initializer on the next line) so the Rust CI gate can pass.

[the-company-company]

🔵 Low · Release the held sleep during failure cleanup

If any wait or assertion fails after heldSleep is installed but before releaseOldSleep() runs, assigning undefined here does not resolve the promise already awaited by onSleep. That leaves the old generation’s shutdown task permanently suspended and can leak native/registry work into the rest of the shared-engine suite. Keep the resolver in teardown scope and invoke it in finally before clearing the fixture.