Parquet
Open a Parquet file without Python
Somebody sent you a .parquet link, and every answer to "how do I open this"
starts with installing Python. Foretaste reads the schema straight from the link: the row
count, the row groups, and every column with its type.
01
What you see
- The row count, exactly, for the whole file.
- The row groups, how many there are and how many rows are in each.
- Every column, with its name and its physical type:
INT32,BYTE_ARRAY,DOUBLE,BOOLEANand the rest.
That is the shape of the dataset, which is usually what you needed to know: whether the columns are the ones you expected, and whether this is the export you were after. The per-column statistics also held in the footer, such as null counts, distinct values and minimums, are a Pro feature. The schema is not.
02
Why it costs kilobytes
Parquet writes its metadata at the end of the file: a Thrift-encoded footer holding the
schema, the row count and the layout of every row group, followed by four bytes saying how
long the footer is and then the magic word PAR1.
So one range request for the last few kilobytes brings back the entire structure. It costs the same for a 2 GB dataset as for a 20 KB one, because none of the data is in the part being read.
The same property as a ZIP's index, which is also at the end, for much the same reason.
You do not have to take that on trust. Under the file's meter, How this was read lists every request Foretaste made for it: what it asked for, what came back and how much. On a 50 MB dataset that is one request for the last 64 KB, and the line underneath reads row groups touched: none.
03
Who this is for
Analysts who are sent a Parquet file and want the column list before opening a notebook.
Engineers checking that an export has the columns and types it was meant to, before writing anything against it.
Anyone handed a .parquet link who has never met one before.
04
Where it stops
Schema only. No data page is ever read, so no values are shown. If you need the rows themselves, download the file and use pandas, DuckDB or anything else built for it.
Nothing can be queried. There is no filtering and no aggregation, because there is no data loaded to filter.
That boundary is the point rather than a shortfall: the job here is telling you whether this is the right file, before you spend ten minutes finding out the slow way.
05
Nothing is uploaded
The range request goes to the website hosting the dataset, the same request your browser would make if you clicked the link. Foretaste has no server for a file to pass through, and nothing about what you opened is recorded anywhere.
The questions people ask
How do I open a Parquet file without Python?
Install the Foretaste browser extension, right-click a .parquet link and choose Open with Foretaste. It reads the schema out of the file footer and shows the row count, the row groups and every column with its type, with no Python, pandas or command line involved.
Can I view a Parquet file in my browser?
Yes, the schema of one. Foretaste fetches the footer metadata at the end of the file, typically a few kilobytes, and shows the row count, row groups, column names and column types. No data page is ever read, so the values themselves are not shown.
What does Foretaste show for a Parquet file?
The total row count, the number of row groups, and every column with its name and physical type. All of it comes from the footer. Column statistics such as null counts and minimums are a Pro feature; the schema is free.