- Shipped
- August 21, 2026 at 12:20 AM UTC
- Author
- Kamo
- Commit
- 00d2922
Three findings from an endpoint-by-endpoint audit, each measured against production rather than inferred. member_rights_applied declared no indexes at all while every finder on it filters by MEMBER_ID. At 10,524 rows a single member's rights lookup was a full sequential scan: Seq Scan on member_rights_applied (actual rows=192) Execution Time: 368.278 ms 368 ms to read one member's rights, on a table that grows with members x rights. One composite (member_id, the_right) serves the member-only finders and lets findByMemberAndRight resolve from the index. Deleting a lead-intake endpoint loaded every stored payload as an entity — raw request bodies included — and then deleted them one at a time. The busiest endpoint in production holds 6,842, so that was ~6 MB hydrated and 6,842 distributed writes inside one request. The line above it already deleted mappings in bulk; payloads now match. The intake job's claim was unbounded. It holds each payload's raw and flattened text for the whole run, so its memory was a function of how far behind it had fallen — exactly when it can least afford it, and the farm import put 14,520 payloads in that table. It now takes the oldest 500 per tick and drains a backlog across runs. Also adds grouped counts for the changelog, so /changelog/public/stats can stop hydrating 16,571 entities to build two small maps.