Skip to content

Acknowledgements

Mink stands on open source projects that solved hard problems well. This page names the ones whose design or code is part of what Mink is.

ProjectContribution
Apache FlussThe data model. Log and primary-key tables, buckets, changelogs, merge engines, lakehouse tiering and union read are Fluss concepts. Mink follows its design and its documentation is the reference for how the two compare, see Why not Apache Fluss
AutoMQThe stream engine design. S3Stream is AutoMQ's stream storage over object storage: WAL on S3, stream objects, the object layout and the caching and compaction that make a diskless broker work. Every log tablet in Mink is an S3Stream stream
PicoMQThe Rust port of that engine, s3stream, derived from AutoMQ S3Stream and used by Mink as a dependency. The SQL command log as a metadata plane, node leases and epoch fencing follow PicoMQ's host design. This website is built on the PicoMQ theme
SurrealKV by SurrealDBThe default KV store engine behind primary-key tables. Pure Rust, an LSM with a checkpoint layout of shared .sst files that makes tablet snapshots cheap to take and cheap to restore
Apache DataFusionThe SQL engine in mink-query. Parsing, planning, optimization and vectorized execution over Arrow. Mink supplies the catalog, the table provider and the physical scan, see SQL

Mink is also built with Apache Arrow and Arrow Flight, Apache Iceberg through iceberg-rust, and Tokio. All of the above are Apache 2.0 or MIT licensed. Mink is Apache 2.0.