Comparison
Delta LakevsHudi
Delta Lake
there is a _delta_log folder of JSON commits next to the Parquet, and reading the table means replaying that log.
An open table format whose metadata is an ordered transaction log of commits alongside the data files. It came out of the Spark ecosystem and is deeply integrated with it, and its log-based design makes change data feeds and time travel natural. The practical difference from other formats is mostly ecosystem and maintenance vocabulary rather than capability.
Full entry →Hudi
the table is built for frequent upserts by key, keeping a log of changes beside the base files and merging them on read.
An open table format designed first around record-level upserts and incremental pulls, with an indexing layer mapping keys to files. It is the format whose design centre is mutation rather than append, which makes it a natural landing point for change data capture. Its copy-on-write and merge-on-read table types are the clearest statement of the write-cost versus read-cost trade in the whole lakehouse space.
Full entry →