#read-write #parquet #async-read #thrift #async-write #hadoop #byte

parquet-format-async-temp

Temporary crate containing thrift library + parquet definitions compiled to support read+write async

5 unstable releases

0.3.1 Jun 14, 2022
0.3.0 Mar 15, 2022
0.2.0 Aug 28, 2021
0.1.1 Aug 10, 2021
0.1.0 Aug 8, 2021

#529 in Asynchronous

Download history 301/week @ 2024-05-20 302/week @ 2024-05-27 172/week @ 2024-06-03 141/week @ 2024-06-10 185/week @ 2024-06-17 210/week @ 2024-06-24 49/week @ 2024-07-01 128/week @ 2024-07-08 210/week @ 2024-07-15 162/week @ 2024-07-22 218/week @ 2024-07-29 258/week @ 2024-08-05 180/week @ 2024-08-12 146/week @ 2024-08-19 175/week @ 2024-08-26 134/week @ 2024-09-02

651 downloads per month

Apache-2.0

465KB
10K SLoC

parquet-format-async-temp

This is a temporary crate containing a subset of rust's thirft library and parquet to support native async parquet read and write.

Specifically, it:

  • supports async read API (via futures)
  • supports async write API (via futures)
  • the write API returns the number of written bytes

It must be used with the fork of thrift's compiler available at https://github.com/jorgecarleitao/thrift/tree/write_size .

Why

To read and write files with thrift (e.g. parquet) without commiting to a particular runtime (e.g. tokio, hyper, etc.), the protocol needs to support AsyncRead + AsyncSeek and AsyncWrite respectively.

To not require Seek and AsyncSeek on write, the protocol must return the number of written bytes on its write_* API.

This crate addresses these two concerns for parquet. It is essentially:

Dependencies

~1–1.8MB
~37K SLoC