如何避免在蜂巢中交叉加入？

Question

我有两张桌子。一个包含 100 万条记录，另一个包含 2000 万条记录。

表格1 价值 (1, 1) (2, 2) (3, 3) (4, 4) (5, 4) …… 表 2 价值 (55, 11) (33, 22) (44, 66) (22, 11) (11, 33) ……

我需要用表1中的值乘以表2中的值，得到结果的排名，得到排名前5。 他们的结果是：

表 1 中的值，表 1 中每个值的前 5 (1, 1), 1*44 + 1*66 = 110 (1, 1), 1*55 + 1*11 = 66 (1, 1), 1*33 + 1*22 = 55 (1, 1), 1*11 + 1*33 = 44 (1, 1), 1*22 + 1* 11 = 33 ......

我尝试在 hive 中使用交叉连接。但我总是因为表太大而失败。

Answer 1

首先从表 2 中选择前 5 个，然后与第一个表进行交叉连接。这将与交叉连接两个表并在交叉连接后取 top5 相同，但在第一种情况下连接的行数会少得多。带有 5 行小数据集的交叉连接将被转换为 map-join 并以与 table1 全扫描一样快的速度执行。

看下面的演示。交叉连接被转换为映射连接。请注意计划中的"Map Join Operator" 和此警告："Warning: Map Join MAPJOIN[19][bigTable=?] in task 'Map 1' is a cross product"：

hive> set hive.cbo.enable=true;
hive> set hive.compute.query.using.stats=true;
hive> set hive.execution.engine=tez;
hive> set hive.auto.convert.join.noconditionaltask=false;
hive> set hive.auto.convert.join=true;
hive> set hive.vectorized.execution.enabled=true;
hive> set hive.vectorized.execution.reduce.enabled=true;
hive> set hive.vectorized.execution.mapjoin.native.enabled=true;
hive> set hive.vectorized.execution.mapjoin.native.fast.hashtable.enabled=true;
hive>
    > explain
    > with table1 as (
    > select stack(5,1,2,3,4,5) as id
    > ),
    > table2 as
    > (select t2.id
    >    from (select t2.id, dense_rank() over(order by id desc) rnk
    >            from (select stack(11,55,33,44,22,11,1,2,3,4,5,6) as id) t2
    >         )t2
    >   where t2.rnk<6
    > )
    > select t1.id, t1.id*t2.id
    >   from table1 t1
    >        cross join table2 t2;
Warning: Map Join MAPJOIN[19][bigTable=?] in task 'Map 1' is a cross product
OK
Plan not optimized by CBO.

Vertex dependency in root stage
Map 1 <- Reducer 3 (BROADCAST_EDGE)
Reducer 3 <- Map 2 (SIMPLE_EDGE)

Stage-0
   Fetch Operator
      limit:-1
      Stage-1
         Map 1
         File Output Operator [FS_17]
            compressed:false
            Statistics:Num rows: 1 Data size: 26 Basic stats: COMPLETE Column stats: NONE
            table:{"serde:":"org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe","input format:":"org.apache.hadoop.mapred.TextInputFormat","output format:":"org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat"}
            Select Operator [SEL_16]
               outputColumnNames:["_col0","_col1"]
               Statistics:Num rows: 1 Data size: 26 Basic stats: COMPLETE Column stats: NONE
               Map Join Operator [MAPJOIN_19]
               |  condition map:[{"":"Inner Join 0 to 1"}]
               |  HybridGraceHashJoin:true
               |  keys:{}
               |  outputColumnNames:["_col0","_col1"]
               |  Statistics:Num rows: 1 Data size: 26 Basic stats: COMPLETE Column stats: NONE
               |<-Reducer 3 [BROADCAST_EDGE]
               |  Reduce Output Operator [RS_14]
               |     sort order:
               |     Statistics:Num rows: 1 Data size: 0 Basic stats: PARTIAL Column stats: COMPLETE
               |     value expressions:_col0 (type: int)
               |     Select Operator [SEL_9]
               |        outputColumnNames:["_col0"]
               |        Statistics:Num rows: 1 Data size: 0 Basic stats: PARTIAL Column stats: COMPLETE
               |        Filter Operator [FIL_18]
               |           predicate:(dense_rank_window_0 < 6) (type: boolean)
               |           Statistics:Num rows: 1 Data size: 0 Basic stats: PARTIAL Column stats: COMPLETE
               |           PTF Operator [PTF_8]
               |              Function definitions:[{"Input definition":{"type:":"WINDOWING"}},{"partition by:":"0","name:":"windowingtablefunction","order by:":"_col0(DESC)"}]
               |              Statistics:Num rows: 1 Data size: 0 Basic stats: PARTIAL Column stats: COMPLETE
               |              Select Operator [SEL_7]
               |              |  outputColumnNames:["_col0"]
               |              |  Statistics:Num rows: 1 Data size: 0 Basic stats: PARTIAL Column stats: COMPLETE
               |              |<-Map 2 [SIMPLE_EDGE]
               |                 Reduce Output Operator [RS_6]
               |                    key expressions:0 (type: int), col0 (type: int)
               |                    Map-reduce partition columns:0 (type: int)
               |                    sort order:+-
               |                    Statistics:Num rows: 1 Data size: 48 Basic stats: COMPLETE Column stats: COMPLETE
               |                    UDTF Operator [UDTF_5]
               |                       function name:stack
               |                       Statistics:Num rows: 1 Data size: 48 Basic stats: COMPLETE Column stats: COMPLETE
               |                       Select Operator [SEL_4]
               |                          outputColumnNames:["_col0","_col1","_col2","_col3","_col4","_col5","_col6","_col7","_col8","_col9","_col10","_col11"]
               |                          Statistics:Num rows: 1 Data size: 48 Basic stats: COMPLETE Column stats: COMPLETE
               |                          TableScan [TS_3]
               |                             alias:_dummy_table
               |                             Statistics:Num rows: 1 Data size: 1 Basic stats: COMPLETE Column stats: COMPLETE
               |<-UDTF Operator [UDTF_2]
                     function name:stack
                     Statistics:Num rows: 1 Data size: 24 Basic stats: COMPLETE Column stats: COMPLETE
                     Select Operator [SEL_1]
                        outputColumnNames:["_col0","_col1","_col2","_col3","_col4","_col5"]
                        Statistics:Num rows: 1 Data size: 24 Basic stats: COMPLETE Column stats: COMPLETE
                        TableScan [TS_0]
                           alias:_dummy_table
                           Statistics:Num rows: 1 Data size: 1 Basic stats: COMPLETE Column stats: COMPLETE

Time taken: 0.199 seconds, Fetched: 66 row(s)

只需将我演示中的堆栈替换为您的表格即可。