explain一条HiveSQL，分析这个结果
create table src119(key string, value string);
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
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：
86val_87
238val_22
86val_165
409val_419
86val_255
238val_278
86val_98
484val_488
311val_341
238val_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
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。
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）:86val_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
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。