ONEDB-10 GATE#1 step iii: flip venus-kpi-collect.sh to pg-write (telemetry_rw via metrics_write) — kills kpi.sqlite split-brain co-writer; nw lane (cross-host)

ONEDB-10 · one-db · child of #698

GATE#1 step iii: flip venus-kpi-collect.sh to pg-write (telemetry_rw via metrics_write) — kills kpi.sqlite split-brain co-writer; nw lane (cross-host)

Ref: ONEDB-10 (#742)
Project: one-db
Parent: done #698 No dual-write: cutover = flip reporter (netreport.sh) to pg + drop sqlite writes after parity. Per Elazar 2026-06-06.
Status: blocked
Priority: high
Type: task
Assigned: nw-whey-cc
Created by: wi-cli-whey
Blocked reason: Q#50: ONEDB-10 venus-kpi pg-flip is a fork decision, not a mechanical edit. venus-kpi-collect.sh is a stale BUGGY fork of kpi-enforce.sh (the canonical whey collector that already does keep-sqlite+pg-mirror correctly and fixed the FK/rowid + dead-DM bugs the venus script still has). Pick: (A) port just the mirror pattern into venus-kpi-collect.sh — minimal, but keeps a 2nd buggy script alive (its FK/dead-DM bugs need fixing anyway = scope creep); or (B) make kpi-enforce.sh host-aware (whey: memory_lint+backup_whey / venus: memory_lint+disk+systemd, gated on HOST), RETIRE venus-kpi-collect.sh, venus runs the canonical script. One collector, buggy fork dies, pg-mirror free. Touches whey's live collector. I recommend B. host col namespaces rows so no double-count either way.
Created: 2026-06-06T13:11:50.491Z
Updated: 2026-06-06T19:35:52.616Z

No sub-items.

2026-06-06T19:35:52.616Z · q#50 · asked by wi-cli-whey → elazar · open

ONEDB-10 venus-kpi pg-flip is a fork decision, not a mechanical edit. venus-kpi-collect.sh is a stale BUGGY fork of kpi-enforce.sh (the canonical whey collector that already does keep-sqlite+pg-mirror correctly and fixed the FK/rowid + dead-DM bugs the venus script still has). Pick: (A) port just the mirror pattern into venus-kpi-collect.sh — minimal, but keeps a 2nd buggy script alive (its FK/dead-DM bugs need fixing anyway = scope creep); or (B) make kpi-enforce.sh host-aware (whey: memory_lint+backup_whey / venus: memory_lint+disk+systemd, gated on HOST), RETIRE venus-kpi-collect.sh, venus runs the canonical script. One collector, buggy fork dies, pg-mirror free. Touches whey's live collector. I recommend B. host col namespaces rows so no double-count either way.

2026-06-06T13:11:50.491Z · created · wi-cli-whey
2026-06-06T13:11:55.259Z · parentChanged · wi-cli-whey

parent=#698
2026-06-06T19:35:52.616Z · questionAsked · wi-cli-whey

ONEDB-10 venus-kpi pg-flip is a fork decision, not a mechanical edit. venus-kpi-collect.sh is a stale BUGGY fork of kpi-enforce.sh (the canonical whey collector that already does keep-sqlite+pg-mirror correctly and fixed the FK/rowid + dead-DM bugs the venus script still has). Pick: (A) port just the mirror pattern into venus-kpi-collect.sh — minimal, but keeps a 2nd buggy script alive (its FK/dead-DM bugs need fixing anyway = scope creep); or (B) make kpi-enforce.sh host-aware (whey: memory_lint+backup_whey / venus: memory_lint+disk+systemd, gated on HOST), RETIRE venus-kpi-collect.sh, venus runs the canonical script. One collector, buggy fork dies, pg-mirror free. Touches whey's live collector. I recommend B. host col namespaces rows so no double-count either way.
2026-06-06T19:35:52.616Z · blocked · wi-cli-whey

blocked by question #50