I'll architect a zero-downtime database migration with compatibility checks, staged rollout, and explicit rollback paths.

MIGRATION STRATEGY
PHASE 0: DUAL-WRITE SETUP (24 hours before cutover)
Deploy code that writes to BOTH old and new database simultaneously. Reads still come from old DB.
Verify data consistency: new DB receives same writes as old DB.

PHASE 1: MIGRATION RUN (low-traffic window, approximately 2 hours)
Backfill data: copy all old DB data to new DB (validate row counts, checksums match)
Run data integrity checks: spot-check 1000 random rows across tables for correctness

PHASE 2: READ SWITCHOVER (5-minute window)
Update app to READ from new DB (still writing to both)
Monitor query latency, error rates for 30 minutes
If anything breaks: ROLLBACK (switch reads back to old DB in 30 seconds; dual-write is still active)

PHASE 3: CLEANUP (hours 4-24 after cutover)
Verify old DB has received no writes for 6 hours
Decommission old database (after keeping backup for 7 days)

ROLLBACK PLAN
T+0 to T+6 hours: full rollback available (dual-write still active, revert to old DB, ~30 sec downtime)
T+6 to T+48 hours: rollback needs careful data reconciliation (dual-write off)
T+48 hours: rollback risky (data diverged, manual work needed)

RUNBOOK CHECKLIST
Backup old DB before any migration
Test migration on staging clone
Identify which queries change performance (index new DB appropriately first)
Notify ops team of maintenance window
Monitor CPU, disk, connection pools during migration
Plan communication: status page, Slack updates

DELIVERABLE
Step-by-step migration plan with checkpoints, rollback triggers, monitoring dashboard config, and runbook.