refactor(runtimed): remove dead broadcast emissions#2065
Merged
Conversation
Remove 8 broadcast types that are fully superseded by RuntimeStateDoc CRDT sync: Already filtered at relay (never reached frontend): - KernelStatus, ExecutionStarted, ExecutionDone, QueueChanged, EnvSyncState Reached frontend but redundant with CellChangeset/CRDT materialization: - Output, DisplayUpdate, OutputsCleared The CRDT changeset pipeline (CellChangeset + scheduleMaterialize) is the source of truth for outputs and execution state. The broadcast path was a parallel channel that predated the CRDT and could cause double-renders or stale overwrites if messages arrived out of order. Also removes the peer relay filter that was dropping the already-filtered broadcasts, and removes broadcast_tx from KernelState (no broadcasts left to send). Remaining broadcasts (Comm, EnvProgress, PathChanged, NotebookAutosaved, KernelError) are either ephemeral events or not representable in the CRDT.
github-actions
Bot
added
the
daemon
This was referenced Apr 23, 2026
rgbkrk
added a commit
that referenced
this pull request
Apr 23, 2026
…broadcast (#2066) * fix(runtimed-py): poll RuntimeStateDoc for wait_for_ready instead of broadcast wait_for_ready was listening for KernelStatus broadcast which was removed in #2065. Switch to polling handle.get_runtime_state() for kernel.status == "idle", which is the CRDT source of truth. * fix(runtimed-py): wait for state transition, not stale idle snapshot Two-phase poll: first wait for status to leave "idle" (restart in progress), then wait for it to return to "idle" (new kernel ready). Prevents returning immediately against the pre-restart idle snapshot.
rgbkrk
added a commit
that referenced
this pull request
Apr 23, 2026
Reflect shipped state (PR 0 + RoomIdentity), drop the RoomDocState substruct (D1 rationale: doc stays top-level), update field counts against post-#2065 code, and reshape the migration path as three atomic per-substruct PRs instead of the original scaffolding-then- migrate split (Rust can't fake field access via methods). Target is now three PRs: RoomBroadcasts, RoomPersistence, RoomConnections. ~89 callsites remaining across them.
rgbkrk
added a commit
that referenced
this pull request
Apr 23, 2026
* docs(specs): streaming Arrow IPC for DataFrame repr (#1816) Design for dx emitting a Parquet head + a pull handle for incremental Arrow IPC continuation, so huge DataFrames render a first screenful immediately and grow in place. Key shape: - Head: 100-ish rows serialized as Parquet through the existing dx path. Sift's existing load hits immediately. - Continuation: new `nteract.dx.stream.<id>` comm. Runtime agent pulls Arrow IPC chunks outside the execution-message hot path and appends them as blob refs in a new manifest field on the same output id. - Transport is shared with #1815 (query backend). - No mutable blobs, no ContentRef shape change — chunks are a JSON list of existing blob refs inside the manifest. - Late joiners replay from the CRDT because chunks go through normal sync, not a side channel. * docs: update streaming Arrow IPC spec for runtime-doc crate changes - RuntimeStateDoc moved from notebook-doc to runtime-doc (#2056) - CRDT writes go through RuntimeStateHandle (#2059) - Pull task uses fork()/merge() for async blob work - Dead broadcasts removed (#2065) - manifest updates propagate via CRDT - Updated review pointers to current file paths * docs: fix reserved-comm-namespace pointers in streaming Arrow IPC spec The namespace rule moved out of CLAUDE.md and now lives in .claude/rules/architecture.md § "Reserved Comm Namespace: `nteract.dx.*`". Update the two spec references to point there. No change to the design itself. * feat(runtimed-wasm): install console_error_panic_hook on module init Rust panics inside WASM currently surface to the frontend as an opaque `__wbg___wbindgen_throw_6b64449b9b9ed33c` stack with wasm-function indices and no file/line. The error reaches the App ErrorBoundary and the "Something went wrong" fallback renders, but the cause is invisible in packaged / CI builds. This is exactly what's happening on UV Pyproject + UV Prewarmed E2E today (post-#2103): something in the runtime-doc read path panics when the daemon syncs a RuntimeState that walks through the full lifecycle starting → running, and we have no way to name it. Install `console_error_panic_hook::set_once()` from a `#[wasm_bindgen(start)]` function so it runs exactly once before any `NotebookHandle` is constructed. Panics now log with file, line, message, and a Rust backtrace. Combined with #2101 (ErrorBoundary → host logger), the next failing E2E run will emit both the React component stack and the Rust panic payload into `e2e-logs/app.log`. Rebuilds the WASM bundle to pick up the hook wiring. Verification: - `cargo xtask wasm runtimed` — succeeds - `deno test --allow-read crates/runtimed-wasm/tests/` — shape test still passes (51 filtered + 1 ok, the expected set) * feat(notebook-app): forward console.error to host logger The wasm panic hook from the previous commit calls `console.error`. In dev builds `attachConsole()` from tauri-plugin-log is DEV-only (see packages/notebook-host/src/tauri/index.ts:280), and the plugin only bridges Rust log output INTO the browser console — it doesn't forward browser console OUT to Rust. In packaged / CI builds the panic message goes to `console.error` and stops there. Install a small forwarder in main.tsx: wrap `console.error` to also call `logger.error` (host-log). WASM panics now land in notebook.log alongside everything else, visible in CI's `e2e-logs/app.log`. Preserves the original console.error behavior so devtools stays unchanged. The forwarding call is in a try/catch so a logger failure can't swallow the original error.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Remove 8 broadcast types fully superseded by RuntimeStateDoc CRDT sync. Net -285 lines.
Already filtered at relay (never reached frontend):
Reached frontend but redundant with CellChangeset/CRDT materialization:
Remaining broadcasts (kept):
Why
The CRDT changeset pipeline (
CellChangeset+scheduleMaterialize) is the source of truth for outputs, execution state, kernel status, and queue state. The broadcast path was a parallel channel that predated the CRDT. It could cause double-renders or stale overwrites if messages arrived out of order with CRDT sync.Also removes
broadcast_txfromKernelState(no broadcasts left to send) and the peer relay filter that was dropping the already-filtered broadcasts.Test plan
cargo test -p runtimed- 389 tests passcargo xtask clippy- cleancargo xtask lint- clean