An incremental / streaming csv parser that is 20 times faster than a pure Erlang-implemented parser.
The heavy lifting is done through a nif. See ecsv_parser:parse/2 on how to use it in a streaming context.
$ rebar3 compile
or
$ rebar compile
When building an analytical database, data loading speed is important. A highly-optimized 100% Erlang-based incremental CSV parser was only capable of parsing ~15mb/s of CSV data per actor.
This library solves this parsing bottleneck problem by leveraging a C-based NIF.
Some performance results (MBP from 2013, single actor, incremental CSV line-parsing on top of a single big binary):
| # rows | # fields | fieldsize | throughput | runtime | rows/s |
|---|---|---|---|---|---|
| 100k | 8 | 4 bytes | 110MiB/s | 34ms | 2892598 |
| 100k | 32 | 4 bytes | 153MiB/s | 49ms | 2008920 |
| 100k | 128 | 4 bytes | 202MiB/s | 301ms | 331476 |
| 100k | 8 | 16 bytes | 255MiB/s | 50ms | 1969822 |
| 100k | 32 | 16 bytes | 359MiB/s | 144ms | 692037 |
| 100k | 128 | 16 bytes | 360MiB/s | 576ms | 173490 |
