You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@drill.apache.org by "Steven Phillips (JIRA)" <ji...@apache.org> on 2015/04/08 23:21:13 UTC

[jira] [Created] (DRILL-2723) Inaccurate row count estimate for text files results in BroadcastExchange

Steven Phillips created DRILL-2723:
--------------------------------------

             Summary: Inaccurate row count estimate for text files results in BroadcastExchange
                 Key: DRILL-2723
                 URL: https://issues.apache.org/jira/browse/DRILL-2723
             Project: Apache Drill
          Issue Type: Bug
          Components: Query Planning & Optimization
            Reporter: Steven Phillips
            Assignee: Jinfeng Ni


The current method for estimating row count in text files is to divide the size of the file in bytes by 1024. This row count estimate is used by optimizer to decide if BroadcastExchange should be used.

This results in massive memory consumption when using BroadCast exchange on a large table.

We need a better way of estimating row count, or use a different metric when deciding to use BroadcastExchange.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)