[HIVE-22735] TopNKey operator deduplication

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 4.0.0-alpha-1
• Component/s: Physical Optimizer
• Assignee: Krisztian Kasa
• Reporter: Krisztian Kasa
• Created: 16/Jan/20 07:16
• Updated: 27/Feb/24 22:23
• Resolved: 18/Jan/20 15:43

Description

In some cases more than one TNK operator has the same expressions in the same operator tree or the difference is only a constant column. Most of this cases only one TNK op. should remain.

+----------------------------------------------------+
|                      Explain                       |
+----------------------------------------------------+
| Plan not optimized by CBO.                         |
|                                                    |
| Vertex dependency in root stage                    |
| Map 1 <- Reducer 8 (BROADCAST_EDGE)                |
| Reducer 2 <- Map 1 (SIMPLE_EDGE), Map 5 (SIMPLE_EDGE), Map 6 (BROADCAST_EDGE), Map 7 (BROADCAST_EDGE), Map 9 (BROADCAST_EDGE) |
| Reducer 3 <- Reducer 2 (SIMPLE_EDGE)               |
| Reducer 4 <- Reducer 3 (SIMPLE_EDGE)               |
| Reducer 8 <- Map 7 (CUSTOM_SIMPLE_EDGE)            |
|                                                    |
| Stage-0                                            |
|   Fetch Operator                                   |
|     limit:50                                       |
|     Stage-1                                        |
|       Reducer 4 vectorized                         |
|       File Output Operator [FS_127]                |
|         Limit [LIM_126] (rows=50 width=538)        |
|           Number of rows:50                        |
|           Select Operator [SEL_125] (rows=190 width=538) |
|             Output:["_col0","_col1","_col2","_col3","_col4","_col5","_col6"] |
|           <-Reducer 3 [SIMPLE_EDGE]                |
|             SHUFFLE [RS_30]                        |
|               Select Operator [SEL_29] (rows=190 width=538) |
|                 Output:["_col0","_col1","_col2","_col3","_col4","_col5","_col6"] |
|                 Group By Operator [GBY_28] (rows=190 width=538) |
|                   Output:["_col0","_col1","_col2","_col3","_col4","_col5","_col6"],aggregations:["avg(VALUE._col0)","avg(VALUE._col1)","avg(VALUE._col2)","avg(VALUE._col3)"],keys:KEY._col0, KEY._col1, KEY._col2 |
|                 <-Reducer 2 [SIMPLE_EDGE]          |
|                   SHUFFLE [RS_27]                  |
|                     PartitionCols:_col0, _col1, _col2 |
|                     Group By Operator [GBY_26] (rows=190 width=1134) |
|                       Output:["_col0","_col1","_col2","_col3","_col4","_col5","_col6"],aggregations:["avg(_col9)","avg(_col11)","avg(_col18)","avg(_col12)"],keys:_col102, _col93, 0L |
|                       Top N Key Operator [TNK_60] (rows=127 width=234) |
|                         keys:_col102, _col93, 0L,top n:50 |
|                         Select Operator [SEL_25] (rows=127 width=234) |
|                           Output:["_col9","_col11","_col12","_col18","_col93","_col102"] |
|                           Top N Key Operator [TNK_58] (rows=127 width=234) |
|                             keys:_col102, _col93,top n:50 |
|                             Filter Operator [FIL_49] (rows=127 width=234) |
|                               predicate:((_col22 = _col38) and (_col1 = _col101) and (_col6 = _col69) and (_col3 = _col26)) |
|                               Map Join Operator [MAPJOIN_102] (rows=2044 width=232) |
|                                 Conds:MAPJOIN_101._col1=RS_123.i_item_sk(Inner),Output:["_col1","_col3","_col6","_col9","_col11","_col12","_col18","_col22","_col26","_col38","_col69","_col93","_col101","_col102"] |
|                               <-Map 9 [BROADCAST_EDGE] vectorized |
|                                 BROADCAST [RS_123] |
|                                   PartitionCols:i_item_sk |
|                                   Filter Operator [FIL_122] (rows=204000 width=108) |
|                                     predicate:i_item_sk is not null |
|                                     TableScan [TS_4] (rows=204000 width=108) |
|                                       tpcds_bin_partitioned_orc_100@item,item, ACID table,Tbl:COMPLETE,Col:COMPLETE,Output:["i_item_sk","i_item_id"] |
|                               <-Map Join Operator [MAPJOIN_101] (rows=2010 width=118) |
|                                   Conds:MAPJOIN_100._col6=RS_107.s_store_sk(Inner),Output:["_col1","_col3","_col6","_col9","_col11","_col12","_col18","_col22","_col26","_col38","_col69","_col93"] |
|                                 <-Map 7 [BROADCAST_EDGE] vectorized |
|                                   PARTITION_ONLY_SHUFFLE [RS_107] |
|                                     PartitionCols:s_store_sk |
|                                     Filter Operator [FIL_106] (rows=402 width=94) |
|                                       predicate:s_store_sk is not null |
|                                       TableScan [TS_3] (rows=402 width=94) |
|                                         tpcds_bin_partitioned_orc_100@store,store, ACID table,Tbl:COMPLETE,Col:COMPLETE,Output:["s_store_sk","s_state"] |
|                                 <-Map Join Operator [MAPJOIN_100] (rows=9604000 width=24) |
|                                     Conds:MERGEJOIN_99._col22=RS_118.d_date_sk(Inner),Output:["_col1","_col3","_col6","_col9","_col11","_col12","_col18","_col22","_col26","_col38"] |
|                                   <-Map 6 [BROADCAST_EDGE] vectorized |
|                                     BROADCAST [RS_118] |
|                                       PartitionCols:d_date_sk |
|                                       Filter Operator [FIL_117] (rows=73049 width=8) |
|                                         predicate:d_date_sk is not null |
|                                         TableScan [TS_2] (rows=73049 width=8) |
|                                           tpcds_bin_partitioned_orc_100@date_dim,date_dim, ACID table,Tbl:COMPLETE,Col:COMPLETE,Output:["d_date_sk"] |
|                                     Dynamic Partitioning Event Operator [EVENT_121] (rows=1 width=8) |
|                                       Group By Operator [GBY_120] (rows=1 width=8) |
|                                         Output:["_col0"],keys:_col0 |
|                                         Select Operator [SEL_119] (rows=73049 width=8) |
|                                           Output:["_col0"] |
|                                            Please refer to the previous Filter Operator [FIL_117] |
|                                   <-Merge Join Operator [MERGEJOIN_99] (rows=9604000 width=16) |
|                                       Conds:RS_114.ss_cdemo_sk=RS_116.cd_demo_sk(Inner),Output:["_col1","_col3","_col6","_col9","_col11","_col12","_col18","_col22","_col26"] |
|                                     <-Map 1 [SIMPLE_EDGE] vectorized |
|                                       SHUFFLE [RS_114] |
|                                         PartitionCols:ss_cdemo_sk |
|                                         Filter Operator [FIL_113] (rows=235814137 width=353) |
|                                           predicate:(ss_cdemo_sk is not null and ss_store_sk is not null and ss_item_sk is not null and ss_store_sk BETWEEN DynamicValue(RS_17_store_s_store_sk_min) AND DynamicValue(RS_17_store_s_store_sk_max) and in_bloom_filter(ss_store_sk, DynamicValue(RS_17_store_s_store_sk_bloom_filter))) |
|                                           TableScan [TS_0] (rows=275041999 width=723) |
|                                             tpcds_bin_partitioned_orc_100@store_sales,store_sales, ACID table,Tbl:COMPLETE,Col:PARTIAL,Output:["ss_item_sk","ss_cdemo_sk","ss_store_sk","ss_quantity","ss_list_price","ss_sales_price","ss_coupon_amt"] |
|                                           <-Reducer 8 [BROADCAST_EDGE] vectorized |
|                                             BROADCAST [RS_112] |
|                                               Group By Operator [GBY_111] (rows=1 width=24) |
|                                                 Output:["_col0","_col1","_col2"],aggregations:["min(VALUE._col0)","max(VALUE._col1)","bloom_filter(VALUE._col2, expectedEntries=1000000)"] |
|                                     <-Map 5 [SIMPLE_EDGE] vectorized |
|                                       SHUFFLE [RS_116] |
|                                         PartitionCols:cd_demo_sk |
|                                         Filter Operator [FIL_115] (rows=1920800 width=8) |
|                                           predicate:cd_demo_sk is not null |
|                                           TableScan [TS_1] (rows=1920800 width=8) |
|                                             tpcds_bin_partitioned_orc_100@customer_demographics,customer_demographics, ACID table,Tbl:COMPLETE,Col:COMPLETE,Output:["cd_demo_sk"] |
|                                                    |
+----------------------------------------------------+

Issue Links

• relates to: HIVE-17896 TopNKey: Create a standalone vectorizable TopNKey operator
• links to: Review Board