Parquet Viewer Editor
Open .parquet files in VS Code as a fast, searchable table. Files of any size open immediately, every value is shown exactly as it is stored, and you can edit values, delete rows and save the file back.
Also available for PyCharm and other JetBrains IDEs (2024.3 and later), with the same features: see JetBrains IDEs below.
Features
- Any compression: uncompressed, Snappy, Gzip, Brotli, Zstd, LZ4 and LZ4_RAW.
- Large files open at once: opening reads only the footer at the end of the file. Rows are read in blocks as you scroll, in a background thread, so VS Code never freezes.
- Exact values: DECIMAL with every digit, INT64 and UINT64 without rounding, timestamps down to the nanosecond (including INT96 timestamps from Spark and older pandas), dates, times, NaN and null shown separately.
- Nested data: LIST, MAP and STRUCT columns, nested as deep as needed. Select a cell to see its full value, pretty-printed, below the table.
- Search, filter and sort the whole file, not only the rows on screen:
- type in Search all rows and press Enter;
- + Filter adds a condition (equals, greater than, contains, is empty, …); click a filter to change it;
- click a column header to sort, or use its ⋯ menu.
- Copy a cell, or the selected rows as CSV, TSV (for Excel), JSON or a Markdown table.
- Edit values: double-click a cell (or select it and press Enter, F2 or just start typing), type the new value and press Enter. Values are checked against the column type: a DECIMAL keeps its exact digits, a timestamp must fit the column's precision, and so on. Set to null empties a value. Edited cells are highlighted until you save.
- Delete rows: select rows (click the row numbers; Shift or Ctrl to select more) and press Delete.
- Undo with Ctrl+Z and save with Ctrl+S. The file keeps its schema, compression and metadata. Search, filter, sort, copy, export and the profile all use your edits before you save.
- Export the rows in the current view (with filter and sort applied) to CSV or JSON Lines.
- Profile tab: one pass over the file gives, per column, the nulls, distinct values, min and max, mean, standard deviation and the 5 most common values. Open it as a Markdown report or save it as CSV.
- Compare two files: select two
.parquet files in the Explorer, right-click, Compare Parquet Files…. Shows added, removed and retyped columns, the row counts, and the rows that changed, matched by a key column (such as an id) or by position, with the number of changed values per column.
- Schema tab: the columns with their types, nulls and size, plus Copy schema as JSON.
- Clear messages for files that are empty, cut off, damaged, encrypted, Git LFS pointers or not Parquet at all (for example a renamed
.mat or .csv file), instead of a blank tab.
- Follows your dark, light and high-contrast theme.
Keyboard
| Key |
Action |
| Arrow keys, Page Up/Down, Ctrl+Home/End |
Move between cells |
| Click a row number, Shift/Ctrl+click |
Select rows |
| Ctrl+A |
Select all rows |
| Ctrl+C |
Copy the selected cell, or the selected rows as TSV |
| Enter, F2, double-click, or typing |
Edit the selected cell |
| Enter / Tab (while editing) |
Save the value and move down / right |
| Esc (while editing) |
Cancel the edit |
| Delete |
Delete the selected rows |
| Ctrl+Z / Ctrl+Y |
Undo / redo |
| Ctrl+S |
Save |
| Ctrl+F |
Search |
| Esc |
Clear the selection, or the search when in the search box |
Right-click a cell or row for more copy formats.
Saving
Saving writes a new file next to the original and then replaces it, so a failed save never leaves a half-written file. The schema, the key-value metadata (including pandas and Arrow metadata), the compression of each column and the row-group sizes are kept. Two things can change:
- INT96 timestamp columns are saved as
TIMESTAMP(ns), the modern equivalent. pandas, pyarrow and Spark read the values the same way.
- LZ4 compression cannot be written yet; those columns are saved with Zstd. You are told when this happens.
Nested values (lists, maps and structs) and raw binary values cannot be edited yet; everything else can.
Notes
- A file that changes on disk while open (for example re-written by a script) shows a Reload banner.
- Files on remote or virtual file systems that VS Code cannot read directly are loaded fully into memory.
- Sorting a very large file needs one pass over the sort column, and the first screen of a sorted view reads the rows it shows from across the whole file. Both show progress and can be cancelled by changing the search or filter.
JetBrains IDEs (PyCharm, IntelliJ IDEA, …)
The JetBrains plugin shows the same table in an editor tab, with the same features. The differences are small:
- Save with the IDE's Save shortcut (Ctrl+S, or ⌘S on macOS) or File → Parquet → Save Parquet File. JetBrains IDEs normally save automatically, but Parquet files are only written when you ask. When you close a tab or the project with unsaved changes, you are asked first.
- Undo and Redo are the IDE's own (Edit menu, Ctrl+Z, Ctrl+Shift+Z), and the Edit menu says what will be undone ("Undo Delete 3 rows").
- More actions are in File → Parquet and on the editor tab's right-click menu: Save As, Add Filter, Export Rows, Profile Columns, Copy Schema as JSON, Reload File.
- Compare two files: select one or two
.parquet files in the Project view, right-click, Compare Parquet Files… (also in the Tools menu). The comparison opens in its own tab; Open Full Report shows the Markdown report.
- Reports (profile, compare) open as Markdown files in the IDE.
- The plugin needs the IDE's built-in browser (JCEF), which every standard JetBrains IDE download includes.
| |