TODO: ktichen sink counter latency

[NathanFlurry]

Description

Please include a summary of the changes and the related issue. Please also include relevant motivation and context.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

How Has This Been Tested?

Please describe the tests that you ran to verify your changes.

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[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 👈 (View in Graphite)
• 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
• 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]

PR Review: counter-latency Rust Port + Kitchen-Sink Harness Improvements\n\n> Note: This is a DRAFT PR with an unfilled description template. The review covers the code as written.\n\n### Overview\n\nThis PR ports the TypeScript counter-latency benchmarking script to Rust, adds an echo handler to the counter actor for raw-WS RTT measurement, adds an on-sleep-remote-watch remote monitoring script, and makes several kitchen-sink Docker/build improvements. The Rust port adds meaningful capabilities: multi-threaded concurrency, colored histogram output, teed transcript logging, and separate rtt/concurrent/agent-concurrent modes.\n\n---\n\n### Bugs / Missing Files\n\n**tsup.server.config.ts is referenced but not present.\npackage.json adds pnpm exec tsup --config tsup.server.config.ts to the build script, and the Dockerfile CMD now runs dist-server/server.mjs. Neither the tsup config file nor the dist-server/ output appear in the diff, and the file does not exist on disk. The Docker image will produce a broken server entry point.\n\nRIVET_PROTOCOLS constant is declared but unused (ws.rs:11).\nThe gateway protocol header is hardcoded as "rivet, rivet_encoding.json" on line 18 instead of using the constant. The constant is dead code.\n\nactor_id is always None (rtt.rs:78, concurrent.rs throughout).\nBoth RTT and concurrent workers hardcode let actor_id: Option<String> = None; and never populate it. format_actor() always returns an empty string, so actor IDs are never visible in output. This looks like unfinished wire-up.\n\nDuplicate base36 implementation.\nrtt.rs:162 defines fn base36 and concurrent.rs defines an identical fn to_base36. These should be consolidated into a shared module.\n\n---\n\n### CLAUDE.md Violations\n\nMutex<HashMap<...>> is banned project-wide.\n- stats.rs:52: worker_health: Mutex<HashMap<u32, WorkerHealth>> uses std::sync::Mutex<HashMap>. CLAUDE.md explicitly bans both Mutex<HashMap> patterns and std::sync::Mutex in async Rust. Use scc::HashMap instead.\n- concurrent.rs (AgentWorkload): Arc<tokio::sync::Mutex<HashMap<String, Instant>>> for pending_inference_sends. Same policy applies - use scc::HashMap.\n\nstd::sync::Mutex in async context (stats.rs).\nCLAUDE.md says async Rust code defaults to tokio::sync::Mutex; std::sync::Mutex is reserved for forced-sync contexts (Drop, FFI, sync traits). State is used from async tasks. Switch to scc::HashMap to eliminate the mutex entirely, or parking_lot::Mutex if locking must remain sync.\n\n---\n\n### Design Issues\n\nTwo WorkloadCtx objects are created for concurrent mode.\nmain.rs::run() creates a WorkloadCtx and passes it to install_signal_handlers. Then run_concurrent_mode creates a second WorkloadCtx from the same values. They share the same Arc<State> so stopping propagates correctly, but the duplicate allocation is confusing. Pass the ctx into run_concurrent_mode instead of rebuilding it.\n\nConnect errors do not trigger reconnect (concurrent.rs:291-302).\nWhen open_raw_ws fails the worker returns immediately. A transient network blip during ramp-up permanently drops that concurrency slot with only a CONNECT-ERROR log line. Persistent load tests typically backoff and retry.\n\nACTOR_STOPPED_CLOSE_REASON = "hack_force_close" (args.rs:16).**\nThe name signals a workaround but gives no explanation. Rename to reflect the actual semantic or add a comment explaining what protocol behavior requires this exact string.\n\n---\n\n### Minor Issues\n\n- on-sleep-remote-watch.ts: createClient<any> bypasses type-safety. Prefer a typed client or at least a comment explaining why any is necessary.\n- counter.ts echo handler: websocket.readyState !== 1 is a magic number - prefer WebSocket.OPEN.\n- sigterm-sleep-probe.ts broadcast ordering: The new onSleepStarted broadcast to open websockets happens before the DB INSERT. A client receiving the event cannot immediately fetch proof rows. Swap the order or document the sequencing intent.\n- tee.rs:emit: Acquires stdout().lock() then the file mutex - nested lock ordering is worth a comment.\n- Comment style: File-level comments explain what each file does. CLAUDE.md asks comments to explain the why only when non-obvious.\n\n---\n\n### Positive Highlights\n\n- Dockerfile CMD change to node as PID 1 is correct - SIGTERM reaches the RivetKit handler without going through pnpm wrappers.\n- SKIP_WASM_BUILD=1 / SKIP_NAPI_BUILD=1 cleanly skips unnecessary native builds in Docker.\n- wait_for_echo in rtt.rs correctly skips Ping/Pong/Frame messages and surfaces Close frame code+reason in errors.\n- Per-run transcript in /tmp/ via tee.rs is useful for post-mortem analysis of long benchmark runs.\n- NAPI type fix (Promise<number | undefined | null> to Promise<number | null>) is a correct API surface tightening.