[HIVE-13109] NPE: Fix NDV estimator use within the StatsOptimizer

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Done
• Affects Version/s: 2.1.0
• Component/s: Statistics
• Assignee: Gopal Vijayaraghavan
• Reporter: Gopal Vijayaraghavan
• Created: 20/Feb/16 04:39
• Updated: 02/Jun/16 10:11
• Resolved: 20/Feb/16 05:16

Description

compute_stats() UDF's size estimates only work within the Execution engine, not during optimizer pass.

java.lang.NullPointerException
        at org.apache.hadoop.hive.ql.udf.generic.GenericUDAFComputeStats$GenericUDAFNumericStatsEvaluator$NumericStatsAgg.estimate(GenericUDAFComputeStats.java:434)
        at org.apache.hadoop.hive.ql.udf.generic.GenericUDAFComputeStats$GenericUDAFDoubleStatsEvaluator$DoubleStatsAgg.estimate(GenericUDAFComputeStats.java:643)
        at org.apache.hadoop.hive.ql.optimizer.stats.annotation.StatsRulesProcFactory$GroupByStatsRule.checkMapSideAggregation(StatsRulesProcFactory.java:996)

Issue Links

• is related to: HIVE-13021 GenericUDAFEvaluator.isEstimable(agg) always returns false
• is superceded by: HIVE-13021 GenericUDAFEvaluator.isEstimable(agg) always returns false