hive之explain命令

EXPLAIN 
FROM src119 SELECT key , count(distinct value) group by key 

ABSTRACT SYNTAX TREE: 
(TOK_QUERY (TOK_FROM (TOK_TABREF src119)) (TOK_INSERT (TOK_DESTINATION (TOK_DIR TOK_TMP_FILE)) (TOK_SELECT (TOK_SELEXPR (TOK_TABLE_OR_COL key)) (TOK_SELEXPR (TOK_FUNCTIONDI count (TOK_TABLE_OR_COL value)))) (TOK_GROUPBY (TOK_TABLE_OR_COL key)))) 

STAGE DEPENDENCIES: 
Stage-1 is a root stage 
Stage-2 depends on stages: Stage-1 
Stage-0 is a root stage 

STAGE PLANS: 
Stage: Stage-1 
Map Reduce 
Alias -> Map Operator Tree: 
src119 
TableScan 
alias: src119 
Select Operator 
expressions: 
expr: key 
type: string 
expr: value 
type: string 
outputColumnNames: key, value  dedecms.com  
Group By Operator 
aggregations: 
expr: count(DISTINCT value) 
bucketGroup: false 
keys: 
expr: key 
type: string 
expr: value 
type: string 
mode: hash 
outputColumnNames: _col0, _col1, _col2 
Reduce Output Operator 
key expressions: 
expr: _col0 
type: string 
expr: _col1 
type: string 
sort order: ++ 
Map-reduce partition columns: 
expr: _col0 
type: string 
tag: -1 
value expressions: 
expr: _col2 
type: bigint 
Reduce Operator Tree: 
Group By Operator 
aggregations: 
expr: count(DISTINCT KEY._col1:0._col0) 
bucketGroup: false 
keys: 
expr: KEY._col0 
type: string 
mode: partials 
outputColumnNames: _col0, _col1 
File Output Operator 
compressed: false 
GlobalTableId: 0 
table: 
input format: org.apache.hadoop.mapred.SequenceFileInputFormat 
output format: org.apache.hadoop.hive.ql.io.HiveSequenceFileOutputFormat 

Stage: Stage-2 
Map Reduce 
Alias -> Map Operator Tree: 
file:/tmp/tianzhao/hive_2011-06-11_05-50-09_095_6055107404619839036/-mr-10002 
Reduce Output Operator 
key expressions: 
expr: _col0 
type: string 
sort order: + 
Map-reduce partition columns: 
expr: _col0 
type: string 
tag: -1 
value expressions: 
expr: _col1 
type: bigint 
Reduce Operator Tree: 
Group By Operator 
aggregations: 
expr: count(VALUE._col0) 
bucketGroup: false 
keys: 
expr: KEY._col0 
type: string 
mode: final 
outputColumnNames: _col0, _col1 
Select Operator 
expressions: 
expr: _col0 
type: string 
expr: _col1 
type: bigint 
outputColumnNames: _col0, _col1 
File Output Operator 
compressed: false 
GlobalTableId: 0 
table: 
input format: org.apache.hadoop.mapred.TextInputFormat 
output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat 

Stage: Stage-0 
Fetch Operator 
limit: -1 


输入数据是： 
86
val_87 
238
val_22 
86
val_165 
409
val_419 
86
val_255 
238
val_278 
86
val_98 
484
val_488 
311
val_341 
238
val_278 


FROM src119 SELECT key , count(distinct value) group by key 
238 2 
311 1 
409 1 
484 1 
86 4 



(TOK_QUERY 
(TOK_FROM (TOK_TABREF src119)) 
(TOK_INSERT 
(TOK_DESTINATION (TOK_DIR TOK_TMP_FILE)) 
(TOK_SELECT 
(TOK_SELEXPR (TOK_TABLE_OR_COL key)) 
(TOK_SELEXPR (TOK_FUNCTIONDI count (TOK_TABLE_OR_COL value))) 
) 
(TOK_GROUPBY (TOK_TABLE_OR_COL key)) 
) 
) 


Stage: Stage-1 
Map Reduce 
Stage: Stage-2 
Map Reduce 
这里的Map Reduce说明这两个Stage都是MapReduce job。 
Stage-1 is a root stage 

Stage-2 depends on stages: Stage-1 // Stage-2依赖Stage-1，Stage-1先运行。

分析Stage-1： 
Alias -> Map Operator Tree: //Map阶段 
src119 
TableScan 
alias: src119 
Select Operator 
expressions: 
expr: key 
type: string 
expr: value 
type: string 
outputColumnNames: key, value 
Group By Operator 
aggregations: 
expr: count(DISTINCT value) 
bucketGroup: false 
keys: 
expr: key 
type: string 
expr: value 
type: string 
mode: hash 
outputColumnNames: _col0, _col1, _col2 
Reduce Output Operator 
key expressions: 
expr: _col0 
type: string 
expr: _col1 织梦好，好织梦 
type: string 
sort order: ++ 
Map-reduce partition columns: 
expr: _col0 
type: string 
tag: -1 
value expressions: 
expr: _col2 
type: bigint 
上面的 Map Operator Tree 说明这个是MapReduce job在Map阶段执行的操作。Map阶段执行4个Operator，按顺序分别是TableScanOperator、SelectOperator、GroupByOperator、ReduceSinkOperator。 


TableScan 
alias: src119 
上面这个是TableScanOperator，扫描表src119。 
Select Operator 
expressions: 
expr: key 
type: string 
expr: value 
type: string 
outputColumnNames: key, value 
上面这个是SelectOperator，需要的字段是key和value。 
Group By Operator 
aggregations: 
expr: count(DISTINCT value) 
bucketGroup: false 
keys: 
expr: key 
type: string 
expr: value 
type: string 
mode: hash 
outputColumnNames: _col0, _col1, _col2 
上面这个是GroupByOperator，执行聚合函数（aggregations）是count(DISTINCT value)。 
Reduce Output Operator 
key expressions: 
expr: _col0 
type: string 
expr: _col1 
type: string 
sort order: ++ 
Map-reduce partition columns: 
expr: _col0 
type: string 
tag: -1 
value expressions: 
expr: _col2 
type: bigint 
上面这个是ReduceSinkOperator。 织梦好，好织梦 

MapRunner读取一条条记录(record)，把一条条record传递给Mapper（ExecMapper）处理。 
对于一条记录（record）:86
val_87（处理后key=86，value=val_87,blog上面显示稍微有问题）， ExecMper.map 会依次调用TableScanOperator、SelectOperator、GroupByOperator、ReduceSinkOperator的processOp处理这条记录，最后在ReduceSinkOperator的processOp里面 out.collect(keyWritable, value); 收集到MapTask的环形缓冲区(circle buffer)里。（out是OutputCollector），OutputCollector.collect可以参考http://caibinbupt.iteye.com/blog/401374。 


Reduce Operator Tree: // reduce阶段 
Group By Operator 
aggregations: 
expr: count(DISTINCT KEY._col1:0._col0) 
bucketGroup: false 
keys: 
expr: KEY._col0 
type: string 
mode: partials 
outputColumnNames: _col0, _col1 
File Output Operator 
compressed: false 
GlobalTableId: 0 
table: 
input format: org.apache.hadoop.mapred.SequenceFileInputFormat 织梦好，好织梦 
output format: org.apache.hadoop.hive.ql.io.HiveSequenceFileOutputFormat 
上面的 Reduce Operator Tree: 说明这个是MapReduce job在Reduce阶段执行的操作。 

Reduce阶段执行Group By Operator（GroupByOperator）和File Output Operator（FileSinkOperator） 

Group By Operator 
aggregations: 
expr: count(DISTINCT KEY._col1:0._col0) 
bucketGroup: false 
keys: 
expr: KEY._col0 
type: string 
mode: partials 
outputColumnNames: _col0, _col1 
上面这个是GroupByOperator 

File Output Operator 
compressed: false 
GlobalTableId: 0 
table: 
input format: org.apache.hadoop.mapred.SequenceFileInputFormat 
output format: org.apache.hadoop.hive.ql.io.HiveSequenceFileOutputFormat

上面这个是FileSinkOperator 

ReduceTask.run里面 
while (values.more()) { 
reduceInputKeyCounter.increment(1); 
reducer.reduce(values.getKey(), values, collector, reporter);
if(incrProcCount) { 
reporter.incrCounter(SkipBadRecords.COUNTER_GROUP, 
SkipBadRecords.COUNTER_REDUCE_PROCESSED_GROUPS, 1); 
} 
values.nextKey(); 
values.informReduceProgress(); 
} 
reducer是ExecReducer。 

Reduce阶段执行Group By Operator（GroupByOperator）和File Output Operator（FileSinkOperator），Reduce对于每条record（key value对）执行一次GroupByOperator.processOp，当处理了一定的记录后（默认是1000），需要flush一次，flush是调用FileSinkOperator写入HDFS。最后Reduce.close的时候会顺序调用各个operator的close。顺序是因为这些opeator之间是父子关系。所以最后GroupByOperator中残留的数据会forward到FileSinkOperator，通过FileSinkOperator写入HDFS。