PREHOOK: query: -- small 1 part, 4 bucket & big 2 part, 2 bucket CREATE TABLE bucket_small (key string, value string) partitioned by (ds string) CLUSTERED BY (key) SORTED BY (key) INTO 4 BUCKETS STORED AS TEXTFILE PREHOOK: type: CREATETABLE POSTHOOK: query: -- small 1 part, 4 bucket & big 2 part, 2 bucket CREATE TABLE bucket_small (key string, value string) partitioned by (ds string) CLUSTERED BY (key) SORTED BY (key) INTO 4 BUCKETS STORED AS TEXTFILE POSTHOOK: type: CREATETABLE POSTHOOK: Output: default@bucket_small PREHOOK: query: load data local inpath '../data/files/srcsortbucket1outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_small POSTHOOK: query: load data local inpath '../data/files/srcsortbucket1outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_small POSTHOOK: Output: default@bucket_small@ds=2008-04-08 PREHOOK: query: load data local inpath '../data/files/srcsortbucket2outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_small@ds=2008-04-08 POSTHOOK: query: load data local inpath '../data/files/srcsortbucket2outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_small@ds=2008-04-08 PREHOOK: query: load data local inpath '../data/files/srcsortbucket3outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_small@ds=2008-04-08 POSTHOOK: query: load data local inpath '../data/files/srcsortbucket3outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_small@ds=2008-04-08 PREHOOK: query: load data local inpath '../data/files/srcsortbucket4outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_small@ds=2008-04-08 POSTHOOK: query: load data local inpath '../data/files/srcsortbucket4outof4.txt' INTO TABLE bucket_small partition(ds='2008-04-08') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_small@ds=2008-04-08 PREHOOK: query: CREATE TABLE bucket_big (key string, value string) partitioned by (ds string) CLUSTERED BY (key) SORTED BY (key) INTO 2 BUCKETS STORED AS TEXTFILE PREHOOK: type: CREATETABLE POSTHOOK: query: CREATE TABLE bucket_big (key string, value string) partitioned by (ds string) CLUSTERED BY (key) SORTED BY (key) INTO 2 BUCKETS STORED AS TEXTFILE POSTHOOK: type: CREATETABLE POSTHOOK: Output: default@bucket_big PREHOOK: query: load data local inpath '../data/files/srcsortbucket1outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-08') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_big POSTHOOK: query: load data local inpath '../data/files/srcsortbucket1outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-08') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_big POSTHOOK: Output: default@bucket_big@ds=2008-04-08 PREHOOK: query: load data local inpath '../data/files/srcsortbucket2outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-08') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_big@ds=2008-04-08 POSTHOOK: query: load data local inpath '../data/files/srcsortbucket2outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-08') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_big@ds=2008-04-08 PREHOOK: query: load data local inpath '../data/files/srcsortbucket1outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-09') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_big POSTHOOK: query: load data local inpath '../data/files/srcsortbucket1outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-09') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_big POSTHOOK: Output: default@bucket_big@ds=2008-04-09 PREHOOK: query: load data local inpath '../data/files/srcsortbucket2outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-09') PREHOOK: type: LOAD PREHOOK: Output: default@bucket_big@ds=2008-04-09 POSTHOOK: query: load data local inpath '../data/files/srcsortbucket2outof4.txt' INTO TABLE bucket_big partition(ds='2008-04-09') POSTHOOK: type: LOAD POSTHOOK: Output: default@bucket_big@ds=2008-04-09 PREHOOK: query: explain extended select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key PREHOOK: type: QUERY POSTHOOK: query: explain extended select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key POSTHOOK: type: QUERY ABSTRACT SYNTAX TREE: (TOK_QUERY (TOK_FROM (TOK_JOIN (TOK_TABREF (TOK_TABNAME bucket_small) a) (TOK_TABREF (TOK_TABNAME bucket_big) b) (= (. (TOK_TABLE_OR_COL a) key) (. (TOK_TABLE_OR_COL b) key)))) (TOK_INSERT (TOK_DESTINATION (TOK_DIR TOK_TMP_FILE)) (TOK_SELECT (TOK_HINTLIST (TOK_HINT TOK_MAPJOIN (TOK_HINTARGLIST a))) (TOK_SELEXPR (TOK_FUNCTIONSTAR count))))) STAGE DEPENDENCIES: Stage-3 is a root stage Stage-1 depends on stages: Stage-3 Stage-0 is a root stage STAGE PLANS: Stage: Stage-3 Map Reduce Local Work Alias -> Map Local Tables: a Fetch Operator limit: -1 Partition Description: Partition input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat partition values: ds 2008-04-08 properties: bucket_count 4 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_small numFiles 4 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_small { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 5812 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat properties: SORTBUCKETCOLSPREFIX TRUE bucket_count 4 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_small numFiles 4 numPartitions 1 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_small { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 5812 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe name: default.bucket_small name: default.bucket_small Alias -> Map Local Operator Tree: a TableScan alias: a GatherStats: false HashTable Sink Operator condition expressions: 0 1 handleSkewJoin: false keys: 0 [Column[key]] 1 [Column[key]] Position of Big Table: 1 Bucket Mapjoin Context: Alias Bucket Base File Name Mapping: a {ds=2008-04-08/srcsortbucket1outof4.txt=[ds=2008-04-08/srcsortbucket1outof4.txt, ds=2008-04-08/srcsortbucket3outof4.txt], ds=2008-04-08/srcsortbucket2outof4.txt=[ds=2008-04-08/srcsortbucket2outof4.txt, ds=2008-04-08/srcsortbucket4outof4.txt], ds=2008-04-09/srcsortbucket1outof4.txt=[ds=2008-04-08/srcsortbucket1outof4.txt, ds=2008-04-08/srcsortbucket3outof4.txt], ds=2008-04-09/srcsortbucket2outof4.txt=[ds=2008-04-08/srcsortbucket2outof4.txt, ds=2008-04-08/srcsortbucket4outof4.txt]} Alias Bucket File Name Mapping: #### A masked pattern was here #### Alias Bucket Output File Name Mapping: #### A masked pattern was here #### Stage: Stage-1 Map Reduce Alias -> Map Operator Tree: b TableScan alias: b GatherStats: false Map Join Operator condition map: Inner Join 0 to 1 condition expressions: 0 1 handleSkewJoin: false keys: 0 [Column[key]] 1 [Column[key]] Position of Big Table: 1 BucketMapJoin: true Select Operator Group By Operator aggregations: expr: count() bucketGroup: false mode: hash outputColumnNames: _col0 Reduce Output Operator sort order: tag: -1 value expressions: expr: _col0 type: bigint Local Work: Map Reduce Local Work Path -> Alias: #### A masked pattern was here #### Path -> Partition: #### A masked pattern was here #### Partition base file name: ds=2008-04-08 input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat partition values: ds 2008-04-08 properties: bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 2750 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat properties: SORTBUCKETCOLSPREFIX TRUE bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 4 numPartitions 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 5500 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe name: default.bucket_big name: default.bucket_big #### A masked pattern was here #### Partition base file name: ds=2008-04-09 input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat partition values: ds 2008-04-09 properties: bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 2750 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat properties: SORTBUCKETCOLSPREFIX TRUE bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 4 numPartitions 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 5500 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe name: default.bucket_big name: default.bucket_big Truncated Path -> Alias: /bucket_big/ds=2008-04-08 [b] /bucket_big/ds=2008-04-09 [b] Needs Tagging: false Reduce Operator Tree: Group By Operator aggregations: expr: count(VALUE._col0) bucketGroup: false mode: mergepartial outputColumnNames: _col0 Select Operator expressions: expr: _col0 type: bigint outputColumnNames: _col0 File Output Operator compressed: false GlobalTableId: 0 #### A masked pattern was here #### NumFilesPerFileSink: 1 #### A masked pattern was here #### table: input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat properties: columns _col0 columns.types bigint escape.delim \ hive.serialization.extend.nesting.levels true serialization.format 1 TotalFiles: 1 GatherStats: false MultiFileSpray: false Stage: Stage-0 Fetch Operator limit: -1 PREHOOK: query: select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key PREHOOK: type: QUERY PREHOOK: Input: default@bucket_big PREHOOK: Input: default@bucket_big@ds=2008-04-08 PREHOOK: Input: default@bucket_big@ds=2008-04-09 PREHOOK: Input: default@bucket_small PREHOOK: Input: default@bucket_small@ds=2008-04-08 #### A masked pattern was here #### POSTHOOK: query: select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key POSTHOOK: type: QUERY POSTHOOK: Input: default@bucket_big POSTHOOK: Input: default@bucket_big@ds=2008-04-08 POSTHOOK: Input: default@bucket_big@ds=2008-04-09 POSTHOOK: Input: default@bucket_small POSTHOOK: Input: default@bucket_small@ds=2008-04-08 #### A masked pattern was here #### 928 PREHOOK: query: explain extended select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key PREHOOK: type: QUERY POSTHOOK: query: explain extended select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key POSTHOOK: type: QUERY ABSTRACT SYNTAX TREE: (TOK_QUERY (TOK_FROM (TOK_JOIN (TOK_TABREF (TOK_TABNAME bucket_small) a) (TOK_TABREF (TOK_TABNAME bucket_big) b) (= (. (TOK_TABLE_OR_COL a) key) (. (TOK_TABLE_OR_COL b) key)))) (TOK_INSERT (TOK_DESTINATION (TOK_DIR TOK_TMP_FILE)) (TOK_SELECT (TOK_HINTLIST (TOK_HINT TOK_MAPJOIN (TOK_HINTARGLIST a))) (TOK_SELEXPR (TOK_FUNCTIONSTAR count))))) STAGE DEPENDENCIES: Stage-1 is a root stage Stage-0 is a root stage STAGE PLANS: Stage: Stage-1 Map Reduce Alias -> Map Operator Tree: b TableScan alias: b GatherStats: false Sorted Merge Bucket Map Join Operator condition map: Inner Join 0 to 1 condition expressions: 0 1 handleSkewJoin: false keys: 0 [Column[key]] 1 [Column[key]] Position of Big Table: 1 Select Operator Group By Operator aggregations: expr: count() bucketGroup: false mode: hash outputColumnNames: _col0 Reduce Output Operator sort order: tag: -1 value expressions: expr: _col0 type: bigint Path -> Alias: #### A masked pattern was here #### Path -> Partition: #### A masked pattern was here #### Partition base file name: ds=2008-04-08 input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat partition values: ds 2008-04-08 properties: bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 2750 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat properties: SORTBUCKETCOLSPREFIX TRUE bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 4 numPartitions 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 5500 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe name: default.bucket_big name: default.bucket_big #### A masked pattern was here #### Partition base file name: ds=2008-04-09 input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat partition values: ds 2008-04-09 properties: bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 2750 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat properties: SORTBUCKETCOLSPREFIX TRUE bucket_count 2 bucket_field_name key columns key,value columns.types string:string #### A masked pattern was here #### name default.bucket_big numFiles 4 numPartitions 2 numRows 0 partition_columns ds rawDataSize 0 serialization.ddl struct bucket_big { string key, string value} serialization.format 1 serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe totalSize 5500 #### A masked pattern was here #### serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe name: default.bucket_big name: default.bucket_big Truncated Path -> Alias: /bucket_big/ds=2008-04-08 [b] /bucket_big/ds=2008-04-09 [b] Needs Tagging: false Reduce Operator Tree: Group By Operator aggregations: expr: count(VALUE._col0) bucketGroup: false mode: mergepartial outputColumnNames: _col0 Select Operator expressions: expr: _col0 type: bigint outputColumnNames: _col0 File Output Operator compressed: false GlobalTableId: 0 #### A masked pattern was here #### NumFilesPerFileSink: 1 #### A masked pattern was here #### table: input format: org.apache.hadoop.mapred.TextInputFormat output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat properties: columns _col0 columns.types bigint escape.delim \ hive.serialization.extend.nesting.levels true serialization.format 1 TotalFiles: 1 GatherStats: false MultiFileSpray: false Stage: Stage-0 Fetch Operator limit: -1 PREHOOK: query: select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key PREHOOK: type: QUERY PREHOOK: Input: default@bucket_big PREHOOK: Input: default@bucket_big@ds=2008-04-08 PREHOOK: Input: default@bucket_big@ds=2008-04-09 PREHOOK: Input: default@bucket_small PREHOOK: Input: default@bucket_small@ds=2008-04-08 #### A masked pattern was here #### POSTHOOK: query: select /* + MAPJOIN(a) */ count(*) FROM bucket_small a JOIN bucket_big b ON a.key = b.key POSTHOOK: type: QUERY POSTHOOK: Input: default@bucket_big POSTHOOK: Input: default@bucket_big@ds=2008-04-08 POSTHOOK: Input: default@bucket_big@ds=2008-04-09 POSTHOOK: Input: default@bucket_small POSTHOOK: Input: default@bucket_small@ds=2008-04-08 #### A masked pattern was here #### 928