Print a Parquet File's Schema Without Installing Anything

The situation

Someone hands you a Parquet file and the first question is always the same: what is in it? Column names, types, row count. The traditional answers all start with an installation — and if you are on a locked-down laptop, in a hurry, or just do not want a Python environment for a ten-second question, that is friction you do not need.

The zero-install way

Drop the file into the Parquet viewer. The schema tab shows every column with its physical and logical type, plus row count, row groups, compression and the writer that produced the file (created_by). It is effectively parquet-tools schema as a drag & drop.

Two properties make this better than it sounds:

• It is instant regardless of size. Parquet keeps the schema in a footer at the end of the file, and the viewer reads only that — a multi-gigabyte file shows its schema as fast as a tiny one, because the data pages are never touched.
• Nothing is uploaded. The read happens locally in your browser via WebAssembly, so a confidential extract is as safe to inspect as it is sitting in your Downloads folder.

If you want the schema as a query result — say, to copy column lists — open the file in the SQL workbench instead:

DESCRIBE SELECT * FROM read_parquet('yourfile.parquet');

The traditional ways, for completeness

If you are already in an environment with tooling, the equivalents are:

import pyarrow.parquet as pq
print(pq.read_schema("yourfile.parquet"))

-- DuckDB CLI
DESCRIBE SELECT * FROM 'yourfile.parquet';

Both are fine — they simply assume an installed environment, which is exactly the assumption the browser route removes. (The once-standard parquet-tools jar is deprecated and increasingly awkward to obtain, so it is no longer the easy answer it used to be.)

Reading what the schema tells you

• Physical vs logical types. BYTE_ARRAY + STRING is a text column; INT64 + TIMESTAMP(MICROS) is a proper timestamp; a bare INT96 is a legacy timestamp with its own problems.
• DECIMAL precision and scale live in the logical type — worth checking before money columns flow into a system that silently casts them to float.
• Row groups hint at how the file was written: one giant row group and thousands of tiny ones both have performance implications for readers.

And when the real question is not "what is the schema" but "did the schema change" — yesterday's file versus today's — skip the manual comparison: diff the two files and added, removed and re-typed columns are listed for you, before a single row is read.