A dashboard for looking inside object storage: point it at a bucket and it finds what's there (Iceberg, Delta Lake, Hudi, Paimon, Lance, and other tables, or plain files) and lets you browse tables, metadata files, and data files, query them with the engine of your choice, see how a query reads a table, and find what could be stored better.
Status: design. Nothing is built yet. The design is in spec/, one spec per part.
- Connect to S3, GCS, Azure Blob Storage, S3-compatible stores (R2, Tigris, MinIO), or a local directory
- Detect table formats and file formats from listings and file signatures, or browse the store as plain prefixes
- Table views: snapshots and commits over time, schema history, partitions, metadata files, data files, and deletes
- File views: the byte layout of Parquet, Arrow IPC, ORC, Avro, Vortex, Lance, Puffin, and SSTable files, with their encodings, statistics, and indexes
- SQL through pluggable engines (DataFusion first, DuckDB next), each in its own crate
- A trace of each query: which metadata it read, which files and row groups it skipped and why, and the requests and bytes it cost
- Checks that point out small files, missing statistics, unsorted data, piled-up deletes, old snapshots, and more
- Read-only: lakeglass never writes to your store