perf(db): index ai_usage_events(status, created_at) for the daily-budget query

[JSONbored]

Investigation: are hot-path queries doing full table scans?

Audited the prod D1 index inventory against every query the code runs. Conclusion: the per-webhook / gate hot path is fully indexed — issues/pull_requests/pull_request_files/bounties by repo_full_name, official_miner_detections by login (PK), repositories/repository_settings by PK. The recent prod D1 overload is therefore capacity / DB-size related (~2.87 GB, blob-heavy tables), not a missing-index problem on the hot path.

A few things I checked and ruled out:

• webhook_events (39,780 rows) — the app only reads it by delivery_id (PK). The 40k-row scans I saw earlier were my own ad-hoc diagnostic queries, not the app. No index needed.
• audit_events (15,876 rows) — both app reads filter by event_type first, covered by audit_events_type_created_idx. No index needed.

The one real gap

sumAiEstimatedNeuronsSince (the daily AI-neuron budget check) runs:

SELECT sum(estimated_neurons) FROM ai_usage_events WHERE status = 'ok' AND created_at >= ?

EXPLAIN QUERY PLAN on prod confirms a full SCAN ai_usage_events — the existing indexes are (feature, created_at) and (actor, created_at), neither of which covers a status-only filter. This query runs on every AI review/summary (it's the budget gate), and ai_usage_events grows with each AI call, so the scan cost grows unbounded once AI is enabled — which #652 (AI review) is about to do. It also serves the existing ai-summaries budget query on main.

Fix

Add a covering index ai_usage_events_status_created_idx (status, created_at) (schema + migration 0028). A test asserts via EXPLAIN QUERY PLAN that the budget query now SEARCHes via the index instead of scanning. Full suite green.

Migration numbered 0028 to avoid colliding with 0026/0027 reserved by #652; the D1 runner applies un-applied migrations in order and tolerates the gap.

Out of scope (flagged separately)

The actual prod D1 overload looks like capacity/size, not query plans — worth watching D1 metrics / considering retention pruning on the large append-only + blob tables, but that's an ops decision, not a code fix.

[ghost]

gittensory · advisory review

Reviewed 3 changed file(s) — two independent AI reviewers.

Suggested action: ✅ Safe to merge — both reviewers found no blocking issues.

Reviewer A · gpt-oss-120b — recommends ✅ merge
This PR adds a new SQLite index to support the daily AI usage budget query, updates the schema to reflect the index, and includes a unit test verifying the index exists and is used by the query plan. The changes are straightforward, safe, and well‑documented.

Suggestions

• Confirm that the migration runner includes the new 0028 file in production deployments.
• Consider adding a benchmark test to ensure the index provides the expected performance improvement under realistic data volumes.

Worth double-checking

• The new index will add write overhead on inserts/updates to ai_usage_events; monitor impact on write latency.
• If the migration order changes, the index could be created after dependent code runs; ensure migrations are applied before the app starts.

Reviewer B · nemotron-3-120b-a12b — recommends ✅ merge
The PR adds a performance index on ai_usage_events(status, created_at) to optimize the daily-budget query (sumAiEstimatedNeuronsSince) used in AI reviews/summaries. It includes a migration (0028), schema synchronization, and a test verifying index usage via EXPLAIN QUERY PLAN. The change is safe, well-documented, and does not affect public/private boundaries or introduce forbidden terms.

Worth double-checking

• Write performance on ai_usage_events may slightly decrease due to index maintenance (acceptable trade-off for read-heavy workload).
• Confirm migration number 0028 does not conflict with existing migrations in main branch (author notes 0026/0027 reserved for PR feat(ai-review): dual-AI maintainer review + BYOK (Phase C) #652).

[ghost]

Note

Gittensory Gate skipped

PR closed before full evaluation. No late first comment was created.

Signal	Result	Evidence	Action
Gate result	⚠️ Skipped	#657 is no longer open.	No action.

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.