Design a Database Migration Service with Replication and Bottleneck Analysis

Design a database migration service. Explain how it copies existing data, keeps up with ongoing changes if required, scales the work, and identifies bottlenecks before and during cutover. ### Part 1 — Replication and Correctness Define the migration consistency goal and describe the relationship between the initial copy, incremental replication, validation, and switching clients to the destination. #### What This Part Should Cover A consistent starting point, change capture or another declared strategy, ordering, retries, schema compatibility, and cutover criteria. ### Part 2 — Scaling and Bottlenecks Explain how to divide work safely and determine whether the source, network, destination, or transformation stage limits progress. #### What This Part Should Cover Partitioning, backpressure, replication lag, skew, capacity measurements, and correctness-preserving tuning. ### Constraints Database engines, data volume, schema differences, downtime tolerance, and whether writes continue during migration are unspecified. State assumptions; do not promise zero downtime or universal cross-engine replication without supporting mechanisms. ### Clarifying Questions - Can the source provide a consistent snapshot and a durable change log? - Are schema and data types compatible, and how are schema changes handled during migration? - What downtime, lag, and validation thresholds are acceptable for cutover? ```hint Connect the snapshot to a change-log position The initial copy and later changes must meet without leaving a gap or applying an operation twice incorrectly. ``` ### What a Strong Answer Covers - A complete migration lifecycle with explicit consistency guarantees. - Replication choices, resumable work, and validation before cutover. - Scaling decisions based on measured bottlenecks and resource protection. ### Follow-up Questions - What happens if the destination falls behind faster than it can catch up? - How would you restart a failed copy task without losing or duplicating changes?