fix: indexer getting stuck when syncing from genesis - #3647

Pull request overview

Fixes #3623, where the MPC indexer wedges at genesis on a fresh state-synced node. The root cause was creating the block streamer (indexer.streamer()) before neard had finished syncing — at that point neard is still at genesis, so the streamer's LatestSynced cursor was pinned to genesis, below the node's block tail. The fix defers stream creation until wait_for_full_sync() resolves, and extracts the wait + listen sequence into a new stream_blocks_after_sync helper that races against shutdown_token.

Changes:

• New stream_blocks_after_sync helper that lazily creates the stream only after sync completes, racing the whole sequence against shutdown.
• Replaces the two cfg-duplicated select! blocks with a single call into the helper, using #[cfg(...)] on the conditional process_blocks_receiver argument to listen_blocks.
• Adds three unit tests covering: stream-started-after-sync, no-stream-when-shutdown-precedes-sync, and listen-error-propagation.

Reviewed changes

Findings

Blocking (must fix before merge):

• None.

Non-blocking (nits, follow-ups, suggestions):

• crates/node/src/indexer/real.rs:288 — the doc comment on stream_blocks_after_sync is truncated mid-sentence: "...starting it early pins its cursor at genesis" (no terminating punctuation, no follow-up clause). Suggest closing it, e.g. "...pins its cursor at genesis, below the node's block tail." to match the inline comment at line 212–216.
• crates/node/src/indexer/real.rs:298-307 — the test suite covers (a) sync-then-stream, (b) shutdown-before-sync, and (c) listen returns Err. It does not cover shutdown arriving while listen is running (cancelling the listen future). That's the dominant production path (SIGTERM on a healthy node), and only adding ~10 lines using a pending::<anyhow::Result<()>>() listen plus a tokio::spawn of shutdown.cancel(). Worth adding for symmetry.
• crates/node/src/indexer/real.rs:212-227 — the comment block now bundles two distinct rationales (why we defer stream creation, and how shutdown cleanup works). It reads fine but could be split into two paragraphs with a blank // line between the "defer until synced" rationale and the "SIGTERM cleanup" rationale to make scanning easier. Stylistic only.
• crates/node/src/indexer/real.rs:232 — wait_for_full_sync() was already implicitly awaited via monitor_contract_state(...).await at line 184 (it loops on wait_for_full_sync before returning the initial state). So in practice this second wait resolves immediately. Not a bug — the explicit wait is a useful local invariant and survives any future refactor that decouples those orderings — but worth noting in the PR description so reviewers don't wonder why both exist.

✅ Approved

🤖 Generated by Claude Code