Details
-
Bug
-
Status: Closed
-
Minor
-
Resolution: Fixed
-
None
-
None
-
Reviewed
Description
Issue:
=====
In StatsRulesProcFactory::evaluateCompator, when minValue is >= filtered value, it should return all rows. Currently, it returns numRows/3. This causes lesser number of reducers to be spun up in queries. E.g Q79 in TPC-DS.
E.g: TPC-DS store table stats:
=============================
hive --orcfiledump hdfs://nn:8020/apps/hive/warehouse/tpcds_bin_partitioned_orc_1000.db/store/000000_0 Stripe Statistics: Stripe 1: Column 0: count: 1002 hasNull: false Column 1: count: 1002 hasNull: false min: 1 max: 1002 sum: 502503 Column 2: count: 1002 hasNull: false min: AAAAAAAAAABAAAAA max: AAAAAAAAPPBAAAAA sum: 16032 Column 3: count: 1002 hasNull: false min: max: 2001-03-13 sum: 9950 Column 4: count: 1002 hasNull: false min: max: 2001-03-12 sum: 5010 Column 5: count: 273 hasNull: true min: 2450820 max: 2451313 sum: 669141525 Column 6: count: 1002 hasNull: false min: max: pri sum: 3916 Column 7: count: 994 hasNull: true min: 200 max: 300 sum: 249970 Column 8: count: 996 hasNull: true min: 5002549 max: 9997773 sum: 7382689071 Column 9: count: 1002 hasNull: false min: max: 8AM-8AM sum: 7088 select compute_stats(s_employee_count, 16) from store; {"columntype":"Long","min":200,"max":300,"countnulls":8,"numdistinctvalues":63,"ndvbitvector":"{0, 1, 2, 3, 4, 5, 11, 12}{0, 1, 2, 3, 6}{0, 1, 2, 3, 4, 5, 7, 11}{0, 1, 2, 3, 4, 5, 7}{0, 1, 2, 3, 4, 5, 6}{0, 1, 2, 3, 4, 5, 8}{0, 1, 2, 3, 4}{0, 1, 2, 3, 4, 5, 7, 9}{0, 1, 2, 3, 4}{0}{0, 1, 2, 3, 4, 5, 7}{0, 1, 2, 3, 4, 5, 6, 7}{0, 1, 2, 3, 4, 8, 9, 14}{0, 1, 2, 3, 5}{0, 1, 2, 3, 4, 5, 6, 7}{0, 1, 2, 3, 4, 5, 6, 8}"}
explain select count(s_store_sk) from store where s_number_employees > 200 and s_number_employees < 295;
Above query would first apply 1002/3 = 334 for s_number_employees > 200 and then 334 / 3 = 111 for s_number_employees < 295. Ideally it should return all 1002 rows for filter s_number_employees > 200.
In TPC-DS Q79, this causes too less reduce tasks to be spun up causing runtime delays.