Hive任务分析
如果要深入学习hive,那数据倾斜的问题是必然要面临的,又称为hive调优。想要调优,就必须了解hive是如何把hql转换成MP程序的,转换的规则是什么,只有知道规则,才能分析它的过程,从而有针对性的优化。
Hive通常以hql的形式执行,它的本质是MP程序。不同的hql,会根据自己的规则,转换成不同的MP任务,比如比较简单的hql,会转成一个job(包括一个map和一个reduce),比较复杂的hql(如三表联合),会转成成3个job。
hive提供了explain命令,用于告诉用户hql会转成成哪些stage以及每个stage会干什么。这为深入学习hive提供了可能。
先来看一个简单的例子:
hive> explain select * from weather where city='beijin' limit 5;
OK
ABSTRACT SYNTAX TREE:
(TOK_QUERY (TOK_FROM (TOK_TABREF(TOK_TABNAME weather))) (TOK_INSERT (TOK_DESTINATION (TOK_DIR TOK_TMP_FILE))(TOK_SELECT (TOK_SELEXPR TOK_ALLCOLREF)) (TOK_WHERE (= (TOK_TABLE_OR_COL city)'beijin')) (TOK_LIMIT 5)))
STAGE DEPENDENCIES:
Stage-1 is a root stage
Stage-0 is a root stage
STAGE PLANS:
Stage: Stage-1
Map Reduce
Alias -> Map OperatorTree:
weather
TableScan
alias: weather
Filter Operator
predicate:
expr: (city ='beijin')
type: boolean
Select Operator
expressions:
expr: date
type: string
expr:country
type: string
expr: city
type: string
expr: weath
type: string
expr:mintemperat
type: int
expr:maxtemperat
type: int
expr:pmvalue
type: int
outputColumnNames:_col0, _col1, _col2, _col3, _col4, _col5, _col6
Limit
File OutputOperator
compressed: false
GlobalTableId:0
table:
inputformat: org.apache.hadoop.mapred.TextInputFormat
outputformat: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
Stage: Stage-0
Fetch Operator
limit: 5
参考文章:
http://blog.youkuaiyun.com/kntao/article/details/12837329
http://yugouai.iteye.com/blog/1850816