Restore drill: first run (2026-08-11)

Record, kept as written. This was the Background section of Back up Hardcount, moved here on 2026-10-22.

The first restore drill, on 2026-08-11, failed and found four defects in the scripts. All four were fixed the same day, and the drill then passed.

Date	Result	Notes
2026-08-11	FAIL → 4 defects → PASS (7s total: backup 1s, restore 3s; 5 critical tables count-matched)	First-ever run. Found: (1) host pg_dump 18 vs server 16 → unrestorable archive; (2) host pg_restore 18 vs target 16 → client SET transaction_timeout rolled restore back; (3) fresh-cluster restore rolled back on missing hardcount_app role — the DR path had never worked; (4) restore.sh’s validation query referenced a nonexistent column (table_name vs relname) — broken since birth. All four fixed same day; drill re-run green.