Enhancing Druid’s Analytics with Apache Arrow and Flight SQL

Currently Druid has two result formats: a JSON format and a protobuf-based format using Apache Calcite Avatica. Both of these formats are row-oriented while all of Druid’s internal data representations are column-oriented, as expected for high-performance analytics. This means that performance is being left on the table due to these transpositions.

This talk will describe a host of possible benefits for Apache Druid that would come by supporting Apache Arrow as an output format. Apache Arrow is an in-memory, columnar data format. It is used as the backing memory representation for utilities like pandas and polars, and allows for zero-copy data communication across many libraries. This session will cover various ways that we could improve the performance, interoperability and flexibility of Apache Druid, along with making it easier to integrate Druid with new data sources and analytics pipelines by leveraging Arrow, FlightSQL and ADBC (Arrow Database Connectivity).

Given the competitive landscape of data computation engines right now (Snowflake, BigQuery, Druid, Dremio, DuckDB), embracing support for Arrow will help Druid “keep up with the competition” and open doors for enhanced connectivity and performance!