spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Filipe Gonzaga Miranda (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-26509) Parquet DELTA_BYTE_ARRAY is not supported in Spark 2.x's Vectorized Reader
Date Wed, 13 Feb 2019 10:09:00 GMT

    [ https://issues.apache.org/jira/browse/SPARK-26509?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16767016#comment-16767016
] 

Filipe Gonzaga Miranda commented on SPARK-26509:
------------------------------------------------

HI [~qiaojialin] - thanks for sharing the property. I think it is nice to have a workaround.
However, the point here is to explore the possibility of supporting the Vectorized Reader.
The description already mentions that if this property is set to false, things work.

As I understand the Vectorized Reader reads in batches which may be quite advantageous, and
that's why I opened the ticket.

> Parquet DELTA_BYTE_ARRAY is not supported in Spark 2.x's Vectorized Reader
> --------------------------------------------------------------------------
>
>                 Key: SPARK-26509
>                 URL: https://issues.apache.org/jira/browse/SPARK-26509
>             Project: Spark
>          Issue Type: Bug
>          Components: SQL
>    Affects Versions: 2.4.0
>            Reporter: Filipe Gonzaga Miranda
>            Priority: Major
>   Original Estimate: 40h
>  Remaining Estimate: 40h
>
> I get the exception below Spark 2.4 reading parquet files where some columns are DELTA_BYTE_ARRAY
encoded.
>  
> {code:java}
> java.lang.UnsupportedOperationException: Unsupported encoding: DELTA_BYTE_ARRAY
>  
> {code}
>  
> If the property:
> spark.sql.parquet.enableVectorizedReader is set to false that works
> The parquet files were written with Parquet V2, and as far as I understand the V2 is
the version used in Spark 2.x.
>  I did not find any property to change which Parquet Version Spark uses (V1, V2).
> Is there anyway to benefit from the Vectorized Reader? Or this is something like a new
implementation to support this version? I would propose so.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org


Mime
View raw message