Skip to content
Opensource internals
  • Posts

Duckdb

The Validity Mask Is Optional: A Null Pointer Means No NULLs

Oct 22, 2026 14 min read Jay Hubbert

A null validity_mask pointer in DuckDB means ‘all rows valid’. That single bit of absence removes the NULL check from every kernel and lets the aggregate hash table swap in plain …

DuckdbVectorized executionInternalsMemoryNull handling

Read more →

The Auxiliary Buffer: How a DuckDB Vector Owns Memory It Never Allocated

Oct 22, 2026 13 min read Jay Hubbert

DuckDB’s AuxiliaryDataHolder lets a vector buffer keep other allocations alive without owning them, turning a 16 KiB copy into an O(1) pointer bump.

DuckdbVectorized executionInternalsMemoryC plus plus

Read more →

Orrify: How DuckDB Collapses Six Vector Types Into Three Words

Oct 22, 2026 13 min read Jay Hubbert

DuckDB’s UnifiedVectorFormat turns six physical vector layouts into three machine words, so a kernel written once runs on every vector type.

DuckdbVectorized executionInternalsExecution engine

Read more →

A Tour of DuckDB's Internals: The Forty-Three Tricks

Oct 22, 2026 21 min read Jay Hubbert

Forty-three deep dives into DuckDB internals, grouped by theme, each one a specific readable trick in the source, from vector layouts to Parquet pruning.

DuckdbInternalsPerformanceIndexParquet

Read more →

© 2026 Jay Hubbert · Built with Hugo 0.168.0

RSS ·