perf(api): the sync freshness poll rides along the request's D1 batch

[remcovanmook]

Why

Every 15 s per isolate, the API worker polls _sync_meta to invalidate L1. The poll ran at the top of the request as a separate D1 query, alongside the request's own. If the D1 session or binding keeps only one query in flight, the request waited out a full extra round trip (~20 ms) whenever the poll fired.

Change

• The poll travels with the request's first D1 call. The router claims it right before dispatch (claimSyncPoll) and hands it to the per-request session wrapper (db.carry). The wrapper sends it in the same batch() as the request's first D1 call, whether that's all(), first() or batch(). It costs no round trip and never competes with the request's query.
• Its results stay separate. Its rows go to the poll's own apply step, and its D1 metadata doesn't count toward the request's Server-Timing d1.
• No D1 call, no wait: a request that never reaches D1 (L1/L2 hit) runs the claimed poll after the response, via ctx.waitUntil on the bare session.
• core/sync_state.js: refresh() is split into the poll statement, applyPoll(), and claimPoll()/runPoll().
• REST keeps ensureSyncFreshness unchanged.

Tests

• tests/unit/core/d1stats.test.js (riders):
  • first()/all() send the rider in the same batch, its rows reach apply and not the caller, and its metadata isn't counted;
  • only the first call carries it;
  • batch() appends it and strips its result;
  • takeRider() hands back an unsent rider;
  • a failing apply doesn't fail the request.
• tests/unit/api/sync_poll_rider_sqlite.test.js (end to end, mocked clock):
  • poll due and an L1 miss: one batch carries both, no separate poll query;
  • within 15 s: no poll;
  • poll due and an L1 hit: the poll runs separately after the response.
• The SQLite shim counts batched reads like all()/first().
• Workers 1496/1496, integration 24/24, upstream patterns 60/60, typecheck clean, pre-push hook passed.

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
4 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud