datafusion-ui

A desktop UI for inspecting Apache Parquet files and querying DataFusion / FlightSQL sources. Built in Rust with iced, arrow, and parquet.

Open one or more Parquet files, inspect their metadata, connect to FlightSQL servers, and explore everything through a SQL editor — all from a persistent object-explorer sidebar.

Screenshots

Overview	Row Groups

Explain Analyze	Settings

Features

Explorer sidebar

A persistent panel listing every source:

• Open files — each expands to its Overview and Row Groups views plus a collapsible schema tree. A ≫ button opens a SQL editor seeded with SELECT * FROM <table> LIMIT 100.
• FlightSQL connections — each lazily loads a catalog → schema → table → column tree; ≫ opens a query editor bound to that connection, ⓘ shows connection health.
• Local SQL Workspace — the shared session every open file is registered into.

SQL workspace

The primary way to look at data. A tabbed editor where each tab is bound to a source (the shared local session, or a FlightSQL connection):

• Syntax highlighting, autocomplete (tables, columns, keywords), and live syntax diagnostics, powered by the in-repo sql-ide crate.
• Undo/redo, Run (Ctrl/Cmd+Enter), and EXPLAIN / EXPLAIN ANALYZE with a formatted plan view (toggle to the raw grid).
• Export the full (uncapped) result to Parquet, CSV, or JSON, with format options (compression, CSV header/delimiter, NDJSON).
• A persistent query history panel — load or re-run any past query. History is itself queryable as a history table.

Results grid

The shared grid that renders every query result:

• A per-column stats row: distinct count, null share, min/max, a numeric histogram, and top values for low-cardinality columns — computed from the result set, so it works for both local files and FlightSQL.
• Pagination over the fetched rows, with first/prev/next/last controls.
• Nested-cell expansion — click a struct/list/map cell to open a tree view of its contents (copy as JSON).
• Resizable columns (double-click a handle to auto-fit); widths persist per schema.
• Click any scalar cell to copy its full value; horizontal + vertical scrolling for wide results.

Per-file inspection

• Overview — file size, total rows, row group count, columns, writer (created_by), version, compressed/uncompressed bytes with a compression ratio, and any key/value file metadata.
• Row Groups — one row per row group with sizes and column counts; expand a group for per-column-chunk stats (compression, encodings, value count, sizes, min/max/null counts).

Theming & settings

• Themes: a signature warm dark ("Instrument"), a light theme, and One Dark — or System, which follows the OS light/dark preference detected at startup.
• Adjustable UI scale.
• Tunable DataFusion session (target partitions, batch size, repartition flags, information_schema, statistics, default catalog/schema) and runtime (memory pool, disk/spill manager, result row cap).

Settings are persisted as TOML in the app directory (~/.datafusion-ui/config.toml by default).

Build

Requires a recent Rust toolchain (edition 2024 — Rust 1.95+).

git clone https://github.com/wyatt-herkamp/datafusion-ui
cd datafusion-ui
cargo build --release

The binary lands at target/release/datafusion-ui.

Run

# Empty window — use "Open Parquet…" or "Connect FlightSQL…"
cargo run --release

# Open a file directly
cargo run --release -- file /path/to/file.parquet

# Connect to a FlightSQL endpoint on launch
cargo run --release -- flight-sql http://localhost:50051

# Or after installing the binary:
datafusion-ui file /path/to/file.parquet

By default config and persistent state live in ~/.datafusion-ui; override with --app-dir <dir> or the DATAFUSION_UI_APP_DIR environment variable.

Install (Linux)

Install the binary somewhere on your PATH, then install the .desktop file so file managers list datafusion-ui as a handler for .parquet files.

# Binary
cargo build --release
install -Dm755 target/release/datafusion-ui ~/.local/bin/datafusion-ui

# Desktop entry
install -Dm644 packaging/datafusion-ui.desktop \
    ~/.local/share/applications/datafusion-ui.desktop

# Refresh the desktop database (some environments)
update-desktop-database ~/.local/share/applications 2>/dev/null || true

After this, right-clicking a .parquet file in your file manager should offer Open with → DataFusion UI, and you can make it the default with Properties → Open With.

To install system-wide instead, replace ~/.local/bin with /usr/local/bin and ~/.local/share/applications with /usr/share/applications (needs root).

How it works

Opening a file reads its footer metadata up front (for Overview / Row Groups) and registers it as a named table in a shared DataFusion SessionContext. From then on, browsing the data is just SQL: the workspace runs your query against the shared session (or a FlightSQL connection) and renders the result in the grid. Results are capped at a configurable row limit and paginated client-side. Disk and network I/O run off the UI thread so the app stays responsive on large files.

Workspace

This repo is a Cargo workspace. The GUI crate lives at the root; crates/sql-ide is a standalone, GUI-agnostic SQL editing-support library (lexer, completion, diagnostics).

License

Dual-licensed under MIT or Apache-2.0, at your option.