[tonyalaribe]

Follow-up to #427/#430. Today's partition for 6297304f is 33 live files, all far below the 128 MiB packing target, so the converged-file backfill (#430) never applies; at a carry cap of 8 those 33 files carried 260 index entries (none stale), and a warm 7-day errors list spends ~2.8 s of its 3.8 s searching them.

MAX_CARRIED_ENTRIES 8 → 2. Hot-tail wave outputs are small, so the extra rebuilds are cheap (~1 core estimated at ~2.4 waves/min against 48 cores, 2-7 in use).

Test: carry_forward_declines_past_the_carried_entry_cap now 3 inputs decline / 2 carry; red with the cap at 3.

Local signoff: make ci-signoff fmt/test/pg-smoke/e2e/clippy all pass (attested).

🤖 Generated with Claude Code

[monoscope-bot]

Monoscope production impact · Worth checking

Project 87576849-4941-49d3-a15d-680fef88a1a8 · reviewed 93f7b39089943c5c24a307093b247bc1d771eb83 · advisory

Evidence window (UTC): 2026-10-18 18:37:20.411825172 UTC → 2026-10-25 18:37:20.411825172 UTC

src/tantivy/search.rs:234 (Head)

Production details are available in Monoscope.

telemetry evidence

Coverage: Open Monoscope for evidence and coverage details.

Review history and settings

[claude]

Review: small, focused change (const 8 → 2 plus a matching test update). Looks good to me.

• Correctness: The test boundary is exact: 3 inputs decline and 2 carry, so it fails at the old cap. The PR says it was checked red with the cap at 3.
• Tradeoff: A lower cap means more index rebuilds on the write and maintenance path. The ~1 core estimate is plausible against 48 cores. The PR doesn't name a timefusion_stats metric to watch after deploy, which the repo guidance asks for. I'd watch index-build or backfill counters, and maintenance wave latency, for a couple of hours once it's live.
• Rollback: This is a hardcoded const with no kill switch. The cap is cheap to revert, but an env knob would make a bad prod outcome a config change instead of a deploy. This is optional, and I wouldn't add it if you want to keep the change minimal.
• Doc comment: It is fairly long and mixes incident data (33 files, 260 indexes) into a code comment. A one-line rationale plus a pointer to the PR would match the repo's minimal-comment style.

No blocking issues.

🤖 Generated with Claude Code