laketune
open source · Apache Iceberg first

Your lakehouse tables are slower and costlier than they should be.

LakeTune reads your table metadata and tells you exactly what to fix — compaction, sort and Z-order, snapshot expiry, partitioning — with estimated impact. Read-only. No data scanned.

Star on GitHubJoin waitlist
$pip install laketune
coming soon
example output
❯

// the problem

Tables quietly degrade over time.

Nobody notices until dashboards slow down and the storage bill grows. These three issues cause most of it.

10k+ files / partition

Small files

Streaming writes and frequent commits leave thousands of tiny files. Every query pays for opening them, and planning slows to a crawl.

TBs of unreferenced data

Snapshot bloat

Without regular expiry, old snapshots pin deleted data and grow metadata. You keep paying for storage nobody can query.

full scans on filtered queries

Wrong layout

Partitions and sort order that don't match how the table is queried mean min/max pruning can't skip anything.

// how it works

Three steps. Zero risk to your data.

LakeTune only observes. It recommends; you decide what to run.

metadata-only read-only
  1. 01

    Point it at a table

    Pass a catalog table or a storage path. LakeTune uses your existing cloud credentials.

    laketune inspect s3://warehouse/events
  2. 02

    It reads metadata only

    Manifests, snapshots, file stats and partition specs. Never your data files, never a write.

    metadata.json · manifest lists · manifests
  3. 03

    Get ranked recommendations

    Each finding comes with the exact maintenance command and an estimated impact, so you choose what to run.

    → rewrite_data_files target=512MB

// table formats

Iceberg first. The rest of the lakehouse next.

One set of checks, adapted to each format's metadata model.

  • Apache Icebergavailable first
  • Delta Lakeplanned
  • Apache Hudiplanned
  • Apache Paimonplanned

// surfaces

Use it the way you work.

Same engine, three ways in.

CLI

Scriptable, CI-friendly. JSON output for pipelines and pull-request checks.

$ laketune inspect <table> --json

Terminal UI

Browse a catalog, drill into partitions and findings, all from the keyboard.

$ laketune tui

Browser app

Runs entirely on your machine with DuckDB-WASM. Metadata never leaves your browser.

$ localhost · DuckDB-WASM

// roadmap

Where it's headed.

Built in the open. Priorities shift with feedback from people running real tables.

  1. v0.1

    Iceberg inspector CLI

    Small-file, snapshot and layout checks with suggested maintenance commands.

    in progress
  2. v0.2

    Impact estimates

    Estimated scan reduction and storage savings per recommendation.

    next
  3. v0.3

    Terminal UI

    Interactive catalog browser with drill-down into partitions and files.

    planned
  4. v0.4

    Browser app

    Local-only web UI powered by DuckDB-WASM.

    planned
  5. v0.5

    Delta Lake support

    Same checks against Delta transaction logs.

    planned
  6. later

    Hudi & Paimon

    Extend coverage across the open table formats.

    planned

// open source

Free, open, and yours to run.

LakeTune is licensed under Apache-2.0. Read the code, open issues, and help shape which checks come next.

Apache-2.0
laketune-io/laketune