Why it matters

ORC choice affects query perf + storage. Understanding shapes format decisions.

Advertisement

The architecture

Stripes: chunks of rows. Column-oriented within.

Indices: min/max/count per column per stripe.

ORC internalsStripeschunks of rowsColumn datacompressedIndices + statsfor pushdownPredicate pushdown skips stripes based on stats
ORC structure.
Advertisement

How it works end to end

Compression: Snappy, ZLIB, LZO, ZSTD.

Predicate pushdown: skip stripes based on stats.

Bloom filters: optional per column.

Stripe size: default 64MB.