arrow2 VS ndjson.github.io

One of the interesting components of Polars that I've been watching is the use of the Apache Arrow memory format, which is a standard layout for data in memory that enables processing (querying, iterating, calculating, etc) in a language agnostic way, in particular without having to copy/convert it into the local object format first. This enables cross-language data access by mmaping or transferring a single buffer, with zero [de]serialization overhead.

For some history, there's has been a bit of contention between the official arrow-rs implementation and the arrow2 implementation created by the polars team which includes some extra features that they find important. I think the current status is that everyone agrees that having two crates that implement the same standard is not ideal, and they are working to port any necessary features to the arrow-rs crate and plan on eventually switching to it and deprecating arrow2. But that's not easy.

https://github.com/apache/arrow-rs/issues/1176

https://github.com/jorgecarleitao/arrow2/pull/1476

https://github.com/jorgecarleitao/arrow2 https://github.com/apache/arrow-datafusion https://github.com/apache/arrow-ballista https://github.com/pola-rs/polars https://github.com/duckdb/duckdb

Currently datafusion and polars aren't directly operable iirc because they use different underlying arrows implementations, but there seems to be work being done on that here https://github.com/jorgecarleitao/arrow2/issues/1429

[arrow2](https://github.com/jorgecarleitao/arrow2) and [parquet2](https://github.com/jorgecarleitao/parquet2) are great foundational libraries for and DataFrame libs in Rust.

[1] https://github.com/jorgecarleitao/arrow2

In lockstep with arrow2's 0.13 release, we have published polars 0.23.0.

This is so freaking life-saving! - we have been using StreamingIterator and FallibleStreamingIterator in libraries (arrow2 and parquet2) and the existing landscape is quite confusing for new users!

arrow2 has support for mssql via ODBC (which microsoft has first class support to). Here are the integration tests we have (both read and write) against mssql specifically.

However, since every JSON document can be represented in a single line, something like newline-delimited JSON / JSON Lines feels like it would've been more suitable for that kind of data.

No one uses that format for streamed json, see ndson and jsonl

http://ndjson.org/

The size complaint is overblown, as repeated fields are compressed away.

As other folks rightfully commented, csv is a mine field. One should assume every CSV file is broken in some way. They also don't enumerate any of the downsides of CSV.

What people should consider is using formats like Avro or Parquet that carry their schema with them so the data can be loaded and analyzed without have to manually deal with column meaning.

What is ndjson? Newline delimited JSON is the format the Algolia CLI reads from and writes to files. This means that any command that passes ndjson formatted data as output or accepts it as input can be piped together with an Algolia CLI command! We’ll see more of this in the next example

You might consider using json-lines format (also known as newline-delimited JSON), in which each line is a separate JSON document so they can be loaded individually.

The format you're getting is known as Newline-Delimited JSON. Instead of trying to parse the whole input and pass that to the JSON Decoder, you can use something like bufio.Scanner to get and parse it line by line.

This is in oposition to NDJSON, which allows to split records without deserializing JSON itself, via e.g. read_lines. fwiw CSV suffers from the same problem as JSON - generally not possible to break into records without deserializing. It is worse than NDJSON because the character \n may appear at any position within an item, thus forbidding read_lines.

I've always seen it referred to as ndjson

I think this would be fine, as long as the CSV layer was still parsable using the RFC 4180, then you could still use a normal CSV parser to parse the CSV layer and a normal JSON parser to parse the JSON layer. My worry with your example is that it is nether format, so it will need custom serialisation and deserialisation logic as it is essentially a bran new format.

https://datatracker.ietf.org/doc/html/rfc4180

If you’re looking for line-oriented JSON, another option would be ndjson: http://ndjson.org/