Why it matters

Parquet vs ORC choice affects cross-engine compat. Understanding shapes decisions.

Advertisement

The architecture

Row groups: chunks like stripes.

Columns compressed.

Metadata footer.

Parquet internalsRow groupschunksColumn datacompressedFootermetadataParquet is more portable across engines than ORC
Parquet structure.
Advertisement

How it works end to end

Row group size: default 128MB.

Compression: Snappy default, ZSTD modern.

Encodings: dictionary, RLE, delta.

Cross-engine: Spark, Impala, Presto, Trino all read.