[jira] [Commented] (HIVE-17874) Parquet vectorization fails on tables with complex columns when there are no projected columns

["Ferdinand Xu (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-17874?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16214512#comment-16214512 ] 

Ferdinand Xu commented on HIVE-17874:
-------------------------------------

Thank you for the patch. Just a few minor comments.

Is the last line of comments not needed or half done?
{code:java}
+      //if there are colsToInclude initialize each columnReader
{code}

I see the following is moving from constructor to the initial method. Is it just for clean up code? If so, not sure whether we can move rbCtx = Utilities.getVectorizedRowBatchCtx(conf); as well.
{code:java}
colsToInclude = ColumnProjectionUtils.getReadColumnIDs(conf);
{code}

Unnecessary change for the following line.
{code:java}
+  private VectorizedColumnReader  buildVectorizedParquetReader(
{code}


> Parquet vectorization fails on tables with complex columns when there are no projected columns
> ----------------------------------------------------------------------------------------------
>
>                 Key: HIVE-17874
>                 URL: https://issues.apache.org/jira/browse/HIVE-17874
>             Project: Hive
>          Issue Type: Sub-task
>    Affects Versions: 2.2.0
>            Reporter: Vihang Karajgaonkar
>            Assignee: Vihang Karajgaonkar
>         Attachments: HIVE-17874.01-branch-2.patch, HIVE-17874.01.patch
>
>
> When a parquet table contains an unsupported type like {{Map}}, {{LIST}} or {{UNION}} simple queries like {{select count(*) from table}} fails with {{unsupported type exception}} even though vectorized reader doesn't really need read the complex type into batches.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)