166 points by kaushiksrini 5 days ago
I built a Rust-based CLI/terminal UI for inspecting Parquet files—data, metadata, and row-group-level structure—right from the terminal. If someone sent me a Parquet file, I used to open DuckDB or Polars just to see what was inside. Now I can do it with one command.
Nice work—this hits a real pain point with Parquet. My main use case is debugging partitioned datasets on S3 with schema drift and skew, where I care about: which files/partitions have schema mismatches, weird row-group stats (all-null, out-of-range, huge skew), and doing that via metadata only.
Right now parqeye looks mainly single-file focused. Do you have plans for a “dataset mode” that takes a dir/S3 prefix and surfaces per-file/row-group summaries (row counts, min/max, null %, schema diffs vs a reference file) using just Parquet stats so it scales to tens of GB? Or do you see parqeye intentionally staying a single-file inspector?
I found a similar tool called nail-parquet[1] which has some nice query functions. I packaged[2] it up for nixpkgs but it’s stuck in merge limbo…
[1] https://github.com/Vitruves/nail-parquet [2] https://github.com/NixOS/nixpkgs/pull/449066
Yours looks much better for your use case, but fwiw you can do it in a single command with duckdb too (but not interactive etc.):
duckdb -c "from 'foo.parquet'"
I use a little shell alias that drops me into duckdb with the file loaded into a table for interactive querying:
https://github.com/llimllib/personal_code/blob/c1a74b1b9527f...
Looks great!
Another seemingly extremely similar project released in the last few days: https://github.com/raulcd/datanomy
Great! I worked a lot with parquet like 5 years ago. The frustration and tilt working with the tooling was immense. Thank you for building this, it feels like resolving some old knot in my soul.
Some kind soul made this repository then, and I found it on like the 13th page of Google while in the depths of despair. It is my most treasured GitHub star, a the shining beacon that saved me. I see it has saved 17 other people too.
https://github.com/casidiablo/parquet-tools-for-dumb-people-...
Similar tool for JSONL files: I built JSONL Viewer Pro after repeatedly crashing VS Code trying to inspect multi-GB training datasets and IoT device logs with nested objects.
Native Mac/Windows app with multi-threaded parsing (simdjson), automatic nested object flattening, and handles 10M+ rows instantly.
For HN: Use code HN100 for free access
https://iotdatasystems.gumroad.com/
Built with C++ for native performance (~6MB app, not Electron).
Would love feedback from folks working with large JSONL files.
Super quick feedback - opening that link on my phone shows me two options next to each other, seemingly with the same name / description (followed by …) and same pricetag. I had to turn my phone sideways to see that there is a windows and a Mac version.
I think you can afford the extra characters to show the whole page in portrait mode. (iPhone 16 pro Safari)
This looks very handy, thank you for working on this and making it open source.
I did submit a feature request for vi keybindings; though I could look into contributing this myself if I find a bit of spare time.
The other thing that surprised me was the size of the binaries: 90MB for a TUI tool (x64 Linux)? I wonder what the bulk of that is? Is there an issue with LTO? An other commenter noticed as well.
It also looks like you are building against a relatively recent glibc (2.34), which limits compatibility with older systems. Building against an older glibc can be hard to do, so I am not faulting you here, and you do provide a musl fallback, which is appreciated (mandatory notice that the musl allocator can dramatically degrade the performance of rust programs, just in case you were not aware of this).
A few more ideas for improvements (you probably already have your own laundry list):
- Mouse support?
- Seeing that you do have graphs, it would be fun to see a scatter plot as well as a distribution plot under statistics in the "Row Groups" tab (though you probably pull these from the metadata, so that would require further processing, which may be out of scope).
It's unfortunate that Python and R don't really have any out-of-the-box means of opening data files from arguments, but if you do this kind of stuff on a daily basis it's something that you can set up. My not directly usable examples below.
Python (uv + dataiter, but easy to modify for pandas or polars): https://github.com/otsaloma/dataiter/blob/master/bin/di-open
R (as per comment, requires also ~/.Rprofile code, nanoparquet in this case): https://github.com/otsaloma/R-tools/blob/master/r-load
Beautiful, I'm currently deep into getting our data into iceberg from firehose and I'm really curious what metadata is written, are bloomfilters being written for the columns i want? Has my compaction and sort jobs helped min-max statistics on those columns?
Will take a look when i get to my laptop!
It’s crazy how long we’ve gone without a tool like this. This is huge. Thank you for finally building this!
It is really incredible how poor the parquet tooling has been for years. The cornerstone of data engineering, yet just inspecting a file is needlessly clunky.
What is really missing for parquet's wide adoption is support in Excel.
yep! it’s available as a homebrew tap — you can install it with: `brew install kaushiksrini/parqeye/parqeye`
awesome! i was just looking at a bucket full of parquet files from last year trying to recall some things about them.
i tried to install with brew, but it told me my cli tools were "too out of date". Never seen that before! and also just upgraded.
Will try again tomorrow
Very nice that it can show the metadata. If you rather focus on the data itself, a Swiss army knife in the terminal is VisiData [1] . It works with many formats from CSV to Parquet. You'd need to install Pyarrow I think to read Parquet files. VisiData is great to not only peek into the file but filter it, sort, compute simple metrics and even can plot a histogram or scatterplot for ex. I avoided a lot of Jupyter notebooks by using VisiData :)
[1] https://www.visidata.org/