- Shipped
- November 6, 2026 at 6:31 PM UTC
- Author
- Kamo
- Commit
- 5b273d8
Call Floor outcome press, profiled with JFR on a SecurityService pod (lab org, 40 presses): the /stage write took 160-360 ms, median ~220 ms. About 90 ms of it was the after-commit follow-ups nobody waits for: the contact tracker (status label read + 2-3 upserts, each its own YugabyteDB commit) and the pipeline follower (a second full Lead load). A service that declares the new LeadStatusFollowUps executor now answers once the status has committed and runs those two on it; without one they run inline as before. Stop All Outreach and the due-callback close stay synchronous. YugabyteDB's planner is heuristic: "org_id = ? AND <key>" walks the org's index. Measured on the largest org (EXPLAIN ANALYZE, read-only): - lead history by lead: 17.9 s -> 35 ms. The per-lead reads go by lead_uid, org checked by key. - DialLedger first-text lead read: 7.15 s -> ~1.5 ms (uid only, org checked in Java). - **************** by uid only, org checked in Java (recording access, labels). - /leads grid with named leads (Selected tab, stream leads): org check written org_id + 0 = ?, 7.06 s -> 4.6 ms; same filter, no index can serve it, so the pk is used.