[HIVE-10940] HiveInputFormat::pushFilters serializes PPD objects for each getRecordReader call

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.2.0
• Fix Version/s: 2.0.0
• Component/s: File Formats
• Assignee: Gunther Hagleitner
• Reporter: Gopal Vijayaraghavan
• Created: 05/Jun/15 00:42
• Updated: 16/Feb/16 23:51
• Resolved: 07/Jul/15 21:00

Description

    String filterText = filterExpr.getExprString();
    String filterExprSerialized = Utilities.serializeExpression(filterExpr);

the serializeExpression initializes Kryo and produces a new packed object for every split.

HiveInputFormat::getRecordReader -> pushProjectionAndFilters -> pushFilters.

And Kryo is very slow to do this for a large filter clause.

Issue Links

• relates to: HIVE-11068 Hive throws OOM in client side