You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@arrow.apache.org by Sanskar Modi <sa...@gmail.com> on 2023/06/07 13:12:17 UTC

InputSplit support for parquet reads using Arrow

Hi everyone,

We have a use case where we're writing a parquet file to a remote server
and we want to read this parquet file using arrow. But we want multiple
hosts to read splits of the parquet file based on parquet block size.

Ex: If the parquet file size is 10 GB, we want 5 hosts to read a 2 GB split
of the parquet file. This is possible if we read via native ParquetReader
but from documentation, it is not clear if arrow readers support this. Can
someone help with this?

Regards,
Sanskar Modi

Re: InputSplit support for parquet reads using Arrow

Posted by Weston Pace <we...@gmail.com>.
I'm not familiar with the "parquet block size".  However, you can use row
groups to accomplish this task.  You could write a single 10GB file with 5
row groups.  Then, when reading, the arrow readers allow you to specify
which row groups you would like to read.

On Wed, Jun 7, 2023 at 6:13 AM Sanskar Modi <sa...@gmail.com> wrote:

> Hi everyone,
>
> We have a use case where we're writing a parquet file to a remote server
> and we want to read this parquet file using arrow. But we want multiple
> hosts to read splits of the parquet file based on parquet block size.
>
> Ex: If the parquet file size is 10 GB, we want 5 hosts to read a 2 GB
> split of the parquet file. This is possible if we read via native
> ParquetReader but from documentation, it is not clear if arrow readers
> support this. Can someone help with this?
>
> Regards,
> Sanskar Modi
>