migration-planner

Use a dual write migration pattern: write to both old and new systems, backfill historical data in batches, validate with shadow reads, then gradually flip read traffic using feature flags before decommissioning the old system.

A dual write migration writes data to both the old and new systems simultaneously, treats the old system as primary, backfills historical data, then shifts reads to the new system once validation confirms data parity.

Validate with row count comparisons, random sampling of records (e.g., 1000 rows), and checksum comparisons between source and target. Run these checks continuously during dual write and backfill phases, alerting on drift above 0.1%.

Use feature flags when you need gradual, reversible rollouts. Enable the new system for 1% of traffic, ramp through 10%, 50%, and 100% while monitoring errors and latency, and flip the flag off instantly to roll back.

Blue-green deployment switches all traffic at once, which suits stateless services but not stateful data migrations where records change continuously. For databases, dual writes with backfills provide safer incremental validation.