Rusty Conover
Data infrastructure engineer and founder of Query.Farm. Building DuckDB extensions, scaling data platforms, and writing about what I learn along the way.
Recent Activity
View all →Rusty spent the day advancing batched window aggregation support in duckdb/duckdb. He opened PR #22143 introducing aggregate_window_batch_t, a new callback variant that optimizes window function evaluation by pre-collecting subframes for all output rows and invoking the batch callback once, rather than looping through individual rows. This reduces function call overhead for custom aggregate window operations and demonstrates a thoughtful approach to performance in the aggregation pipeline.
Rusty added a health endpoint to vgi-rpc-python, extending the RPC service with a new monitoring capability. This addition supports better operational observability for deployments relying on the library.
Rusty had a busy day maintaining and updating the DuckDB ecosystem, with particular focus on the community extensions and addressing build issues in the core DuckDB library.
GeoSilo vs GeoArrow + Byte Stream Split: Two Approaches to Geometry Compression
A detailed comparison of GeoSilo's delta-integer encoding against GeoArrow's columnar float64 layout with Parquet BYTE_STREAM_SPLIT, covering compression ratios, precision tradeoffs, and when each approach wins.
VGI Injector: A Tiny HTTPS Download-and-Execute Binary in Zig
I needed a self-contained binary that downloads a program over HTTPS and exec's it — small enough to run in a FROM scratch container with nothing else. Go and Rust couldn't get small enough. Zig could.
TIME Data Type Compatibility Across Databases
A survey of the TIME data type across 14 databases, comparing supported ranges, maximum values, and whether the special value 24:00:00 is accepted.
Telemetry for DuckDB Extensions Without the Pain
I open-sourced the telemetry client I use across Query.Farm's DuckDB extensions. It's two files, one function call, and it only collects platform and version info.
Releasing vgi-rpc: An RPC Framework Built on Apache Arrow
I built an RPC framework for Python that uses Apache Arrow IPC as the wire format and Python Protocol classes as the interface definition. No .proto files, no codegen — just type annotations.
Acronym-Aware Case Conversions in the DuckDB Inflector Extension
The Inflector extension for DuckDB now supports configurable acronyms, so case conversions preserve terms like HTML, API, and URL as fully uppercase — configured through a native DuckDB setting.
DuckDB Extension Development Workshop
DuckDB Developer Meeting #1 — Amsterdam, Netherlands
Building on Flight: Real-World Lessons from the DuckDB Airport Extension
Apache Arrow Summit 2025 — Paris, France
What I've Been Working On
Full activity log →I've been deep in ecosystem stewardship over the past few months, systematically modernizing the DuckDB extension landscape while contributing upstream enhancements to foundational projects like Apache Arrow. February saw a full vgi-rpc ecosystem launch across four languages with production-ready implementations, while January and December focused on dependency upgrades, core database improvements, and extensive classified operations that remain tastefully mysterious. Throughout this period, I've balanced high-level infrastructure work with low-level performance archaeology—from fixing silent data drops in Arrow's IPC layer to debugging aggregate function callbacks in DuckDB core.
- Crushed DuckDB 1.5 compatibility across 40+ extensions (airport, shellfs, webmacro, inflector, and 25+ others) while standardizing documentation and modernizing the entire Query-farm portfolio.
- Shipped Fair Weather Friend with PWA support, astronomical stargazing mode, and specialized weather scoring for cycling and dog walking—transforming a practical tool into something genuinely delightful.
- Contributed upstream Arrow enhancements across JavaScript, Swift, Rust, and Go implementations, adding custom metadata support and public dictionary serialization APIs that prevent production data loss.