What is a merge-on-read vs copy-on-write architecture in modern data lakes?

Merge-on-read defers updates until read time, while copy-on-write rewrites data during updates.

In lakehouse systems like Hudi and Iceberg, copy-on-write rewrites entire files when updates occur, ensuring fast reads but slower writes. Merge-on-read stores updates separately and merges them during reads, improving write performance but increasing read complexity. These strategies balance ingestion speed, storage cost, and query performance.