At least with Cassandra, there are cell-level timestamps which are very useful for doing data migrations while active writes are still incoming. You can simply

At least with Cassandra, there are cell-level timestamps which are very useful for doing data migrations while active writes are still incoming. You can simply mirror the writes to both systems, and then migrate the old data underneath. As long as the data transfer preserves the cell level timestamps, the read path resolves any differences and compaction will eventually clean up any duplicates. (and sstable loads will have the timestamps) Dynamodb does NOT have cell level timestamps, I believe they have row level timestamps. How it is doing globally replicated data and mutation merges: I have no idea. It seemed like a handwave when they were announcing it about two or three years ago.