hive优化方式和使用技巧

部分内容出处:

http://www.atatech.org/article/detail/5617/0

http://www.atatech.org/article/detail/4392/515 

一.UDFS函数介绍

1. 基本UDF

(1)SHOWFUNCTIONS:这个用来熟悉未知函数。

     DESCRIBE FUNCTION<function_name>;

(2)A IS NULL

     A IS NOT NULL

(3)A LIKE B 普通sql匹配如 like “a%”

     A RLIKE B通过正则表达式匹配

     A REGEXP B 通过正则表达式匹配

(4)round(double a):四舍五入

(5)rand(),rand(int seed):返回在(0,1)平均分布的随机数

(6)COALESCE(pv, 0):将 pv 为 null 的行转为0,很实用

2. 日期函数

(1)datediff(string enddate, stringstartdate):

     返回enddate和startdate的天数的差,例如datediff('2009-03-01','2009-02-27') = 2

(2)date_add(stringstartdate, int days):

     加days天数到startdate:date_add('2008-12-31', 1) ='2009-01-01'

(3)date_sub(stringstartdate, int days):

     减days天数到startdate:date_sub('2008-12-31', 1) ='2008-12-30'

(4)date_format(date,date_pattern)

     CREATETEMPORARY FUNCTION date_format AS'com.taobao.hive.udf.UDFDateFormat';

     根据格式串format 格式化日期和时间值date,返回结果串。

     date_format('2010-10-10','yyyy-MM-dd','yyyyMMdd')

(5)str_to_date(str,format)

     将字符串转化为日期函数

CREATE TEMPORARY FUNCTIONstr_to_date AS 'com.taobao.hive.udf.UDFStrToDate';

      str_to_date('09/01/2009','MM/dd/yyyy')

3. 字符串函数

(1)length(stringA):返回字符串长度

(2)concat(stringA, string B...):

     合并字符串,例如concat('foo','bar')='foobar'。注意这一函数可以接受任意个数的参数

(3)substr(stringA, int start) substring(string A,int start):

     返回子串,例如substr('foobar',4)='bar'

(4)substring(string A, int start,int len):

     返回限定长度的子串,例如substr('foobar',4, 1)='b'

(5)split(stringstr, string pat):

     返回使用pat作为正则表达式分割str字符串的列表。例如,split('foobar','o')[2] = 'bar'。

(6)getkeyvalue(str,param):

     从字符串中获得指定 key 的 value 值 UDFKeyValue

     CREATE TEMPORARY FUNCTION getkeyvalue  AS 'com.taobao.hive.udf.UDFKeyValue';

4. 自定义函数

(1)row_number

CREATE TEMPORARY FUNCTION row_number  AS 'com.taobao.ad.data.search.udf.UDFrow_number'; 
select ip,uid,row_number(ip,uid) from ( 
	select ip,uid,logtime from atpanel 
	distribute by ip,uid 
	sort by ip,uid,logtime desc 
) a

(2)拆分key_value键值对

CREATE TEMPORARY FUNCTION ExplodeEX AS 'com.taobao.hive.udtf.UDTFExplodeEX'; 
select 
	split(kvs,'_')[0] as key,
	split(kvs,'_')[1] as key,
from ( select 'a-1|b-2' as kv from dual ) t
lateral view explode (split(kv,'\\|')) result as kvs

二. HIVE新特性

1. 支持多列的COUNT(*)和COUNT DISTINCT查询

   select count(distinct col1, col2) from table_name;select count(*) from table_name;

2. 提供以本地模式运行Hive的选项

   设置mapred.job.tracker=local可开启本地运行模式

3. 增强的列重命名语法

   增加 ALTERTABLE table_name CHANGE old_name new_name语法。

4. 支持UNIQUE JOIN HIVE-591

   select .. from JOINTABLES (A,B,C) WITH KEYS (A.key, B.key, C.key) where ....

5. 增加检测表和分区状态的语法HIVE-667

   使用show table_name语法,检查表和分区的状态,包括大小和创建、访问时间戳。6.      增加建表时支持STRUCT,结构体

7. 增加选择驱动表的提示

8. 增加/*+STREAMTABLE(tb_alias)*/ HINT,以在Join操作时指定驱动表:

   SELECT /*+ STREAMTABLE(a) */ a.val, b.val, c.valFROM a

   JOIN b ON (a.key = b.key1)JOIN c ON (c.key = b.key1)

   指定此HINT后,原先默认的右表驱动会失效。

9. left Semi-Join HIVE-870

   Left Semi-Join是可以高效实现IN/EXISTS子查询的语义。以下SQL语义:

(1)SELECT a.key, a.value FROM a WHERE a.key in (SELECT b.key FROM b);

   未实现Left Semi-Join之前,Hive实现上述语义的语句是:  

   SELECT t1.key, t1.value FROM a t1

   left outer join (SELECT distinct key from b) t2

   on t1.id = t2.id where t2.id is not null;

(2)可被替换为Left Semi-Join如下:

   SELECT a.key, a.valFROM a LEFT SEMI JOIN b on (a.key = b.key)

   这一实现减少至少1次MapReduce过程,注意Left Semi-Join的Join条件必须是等值。

10.Skew Join优化 HIVE-964 ,数据倾斜

   优化skewed join key为map join。开启hive.optimize.skewjoin=true可优化倾斜的数据。Skew Join优化需要额外的mapjoin操作,且不能节省shuffle的代价。

11.Sorted merge (map) join HIVE-1194

   (对关键表key排序)

   如果MapJoin中的表都是有序的,这一特性使得Join操作无需扫描整个表,这将大大加速Join操作。可通过hive.optimize.bucketmapjoin.sortedmerge=true开启这个功能,获得高的性能提升。

hive优化方式和使用技巧_第1张图片

12.支持ALTER TABLE修改分区的InputFormat/OutputFormat定义

   这一特性使得我们可以用压缩方式(SequenceFileInputFormat)存储后续表分区的数据,同时又不需要对以前的表分区做修改,即透明切换到压缩格式。

13.支持并发提交没有依赖关系的MR过程HIVE-549

   此前的Hive仅仅顺序提交MR任务。这一增强使得没有依赖关系的多次MR过程(例如Union all语义中的多个子查询)可以并发提交。某些情况下可以提高单条HQL命令的响应速度。以下参数对并发提交功能启作用:

   hive.exec.parallel[=false]

   hive.exec.parallel.thread.number[=8]

14.Sorted Group byHIVE-931

  (中间表的预处理)

   对已排序的字段做Group by可以不再额外提交一次MR过程。这种情况下可以提高执行效率。

15.UDTF支持

   UDTF即User defined table function,是一种UDF,区别是这种UDF可以返回多条记录。这一修改使得当前很多Transform脚本可以被替换为更通用、更高效、更用户友好的UDTF实现。UDTF是一种1:n输出,可用于行转列等。

   UDTF不支持UDTF/列混合的select、不支持嵌套、不支持相同子查询中的GROUP BY / CLUSTER BY /DISTRIBUTE BY / SORT BY。

UDTF可与Lateral View相结合。

16.支持动态分区HIVE-1002HIVE-1238              

   动态分区可通过设定hive.exec.dynamic.partition=true打开DP特性。使用方法:

   INSERT OVERWRITETABLE tbl partition (col1[=value][, col2[=value] …])

   使用hive.exec.dynamic.partition.mode = nonstrict动态分区有一定风险,包括小文件、覆盖数据等。默认分区开关:

   hive.exec.default.dynamic.partition.name

17.插入强制排序HIVE-1193

   只需要打开hive.enforce.sorting选项即可。这一特性对Sorted merge bucket (map) join非常有用

18.支持视图功能

   可用于字段级别的权限控制

19.支持持笛卡尔积join(1.0特性       SELECT a.*, b.*FROM aCROSS JOIN b

CREATE VIEW [IF NOT EXISTS] view_name
[ (column_name [COMMENT column_comment], … ) ]
[COMMENT ‘view_comment’]
AS SELECT …
[ ORDER BY …  LIMIT … ]

 

三. hive优化方式总结

1. 多表join优化代码结构:

   select .. from JOINTABLES (A,B,C) WITH KEYS (A.key, B.key, C.key) where ....

关联条件相同多表join会优化成一个job

2. LeftSemi-Join是可以高效实现IN/EXISTS子查询的语义

   SELECT a.key,a.value FROM a WHERE a.key in (SELECT b.key FROM b);

(1)未实现Left Semi-Join之前,Hive实现上述语义的语句是:

   SELECT t1.key, t1.valueFROM a  t1

   left outer join (SELECT distinctkey from b) t2 on t1.id = t2.id

   where t2.id is not null;

(2)可被替换为Left Semi-Join如下:

   SELECT a.key, a.valFROM a LEFT SEMI JOIN b on (a.key = b.key)

   这一实现减少至少1次MR过程,注意Left Semi-Join的Join条件必须是等值。

3. 预排序减少map  join和group by扫描数据HIVE-1194

(1)重要报表预排序,打开hive.enforce.sorting选项即可

(2)如果MapJoin中的表都是有序的,这一特性使得Join操作无需扫描整个表,这将大大加速Join操作。可通过

     hive.optimize.bucketmapjoin.sortedmerge=true开启这个功能,获得高的性能提升。

set hive.mapjoin.cache.numrows=10000000;
set hive.mapjoin.size.key=100000;
Insert overwrite table pv_users
Select /*+MAPJOIN(pv)*/ pv.pageid,u.age 
from page_view pv
join user u on (pv.userid=u.userid;

(3)Sorted Group byHIVE-931    

    对已排序的字段做Group by可以不再额外提交一次MR过程。这种情况下可以提高执行效率。

4. 次性pv uv计算框架

(1)多个mr任务批量提交

     hive.exec.parallel[=false]

     hive.exec.parallel.thread.number[=8]

(2) 一次性计算框架,结合multi group by

     如果少量数据多个union会优化成一个job;

     反之计算量过大可以开启批量mr任务提交减少计算压力;

     利用两次group by 解决count distinct 数据倾斜问题 

Set hive.exec.parallel=true;
Set hive.exec.parallel.thread.number=2;
From(
	Select
		Yw_type,
		Sum(case when type=’pv’ then ct end) as pv,
		Sum(case when type=’pv’ then 1 end) as uv,
		Sum(case when type=’click’ then ct end) as ipv,
		Sum(case when type=’click’ then 1 end) as ipv_uv
	from (
		select 
			yw_type,log_type,uid,count(1) as ct
		from (
			select ‘total’ yw_type,‘pv’ log_type,uid from pv_log 
			union all
			select ‘cat’ yw_type,‘click’ log_type,uid from click_log
		) t group by yw_type,log_type
	) t group by yw_type
) t			
Insert overwrite table tmp_1 
Select pv,uv,ipv,ipv_uv 
Where yw_type=’total’

Insert overwrite table tmp_2
Select pv,uv,ipv,ipv_uv
Where yw_type=’cat’;  


5. 控制hive中的map和reduce数

(1)合并小文件

set mapred.max.split.size=100000000;
set mapred.min.split.size.per.node=100000000;
set mapred.min.split.size.per.rack=100000000;
set hive.input.format=
org.apache.hadoop.hive.ql.io.CombineHiveInputFormat;

    hive.input.format=……表示合并小文件。大于文件块大小128m的,按照128m来分隔,小于128m,大于100m的,按照100m来分隔,把那些小于100m的(包括小文件和分隔大文件剩下的),进行合并,最终生成了74个块

(2)耗时任务增大map数

    setmapred.reduce.tasks=10;

6. 利用随机数减少数据倾斜

   大表之间join容易因为空值产生数据倾斜 

select 
	a.uid
from big_table_a a
left outer join big_table_b b
on b.uid = case when a.uid is null or length(a.uid)=0
		then concat('rd_sid',rand()) else a.uid end;


四. 小技巧

1.空值处理, 结果表\N用空字符串代替

   ALTER TABLE a SETSERDEPROPERTIES('serialization.null.format' = '');

2. 避免暴力扫描分区

   今日全量=昨日全量+今日增量

   30数据=前一个30日数据-31日数据+今日数据

   适用场景:需求稳定,需要访问30天或1年数据

3. 利用动态分区减少任务执行时间

 

五. 通过JobTracker 源数据找出低效代码

1. On条件没写或者扫描过多分区情况

   Uv计算参考一次性pv uv计算框架解决方案,on或者分区条件没写去掉即可

select
	id as 天网id,prgname as 任务路径,viewname as 显示名称,job_id ,job_name,job_value,
	length(trim(inputdir))-length(replace(trim(inputdir),',',''))+1 as pathcnt
from (
	select
		t1.id,t1.prgname,t1.viewname,
		t3.job_id,t3.job_name ,
		t3.job_value,
		DBMS_LOB.SUBSTR(t3.job_value,4000) as inputdir
	from(
		 select
			id,prgname,paravalue,viewname from dwa.etl_task_program t
		 where priority in('xx','xxx') --##统计的时候输入自己的业务基线id
			and appflag=0
	) t1,
	dwa.hdp_job_map t2,
	dwa.hdp_job_conf t3
	where t1.id = t2.id
		and t2.job_id = t3.job_id
		and t2.gmtdate = trunc(sysdate-1)
		and t3.gmtdate = trunc(sysdate-1)
		and t3.job_name = 'mapred.input.dir'
)
where length(trim(inputdir))-length(replace(trim(inputdir),',','')) > 10;

2. 同一个脚本相同单表被扫描多次

   尽量把所需要的数据一次性读出来

select sky_id as 天网id,viewname as 天网显示名称,
	tab_name as 被扫描表,on_duty as 负责人,count(1) as  扫描次数 
from(
    select distinct a.tab_name,c.sql_id,a.sub_sql_id,c.sky_id,e.viewname,e.on_duty
    from dwa.meta_tab a,
    dwa.meta_sqlsub b,
    (select * from 
        (select sky_id,sql_id,sql_src,
	    row_number() over(partition by sky_id,length(sql_src) order by sql_id) rn 
	 from dwa.meta_sqlfull
    )where rn=1) c,
    dwa.meta_col d,dwa.etl_task_program e
    where e.priority  in('xx','xxx') --##统计的时候输入自己的业务基线id
        and e.appflag=0 and e.id=c.sky_id
        and a.sub_sql_id=b.sub_sql_id and a.tab_id=d.tab_id and a.sub_sql_id=d.sub_sql_id and b.sqlfull_id=c.sql_id 
        and a.tab_name not like '%-%'  and b.sql_type='select'
    order by c.sky_id,c.sql_id,a.sub_sql_id
)group by sky_id,viewname,tab_name,on_duty
having count(1) >1
order by cnt desc;

3. Job数过多

   尽量一次性读取所需数据

   才有union方式合并任务

   Left  outer join on条件相同会合并成一个job

 SELECT /*+ parallel(t,32) */ 
	groupname,
	id,
	BIZ_SORTID,
	ON_DUTY,
	PRGNAME,
	job_cnt,
	JOB_TOTAL_MAPS,
	JOB_TOTAL_REDUCES,
	TOTAL_TIME,
	HDFS_BYTES_READ,
	HDFS_BYTES_WRITTEN,
	TOTAL_MAP_TIME,
	TOTAL_REDUCE_TIME,
	MAP_INPUT_RECORDS,
	MAP_OUTPUT_RECORDS,
	REDUCE_INPUT_RECORDS,
	REDUCE_OUTPUT_RECORDS,
	time,
	row_number() over(partition by groupname order by TIME desc) rn_time,
	row_number() over(partition by groupname order by TOTAL_MAP_TIME+TOTAL_REDUCE_TIME desc) rn_slots
from(
	select 
	  DWA.ETL_TASK_BASELINE.name as  groupname,
	  DWA.HDP_JOB_MAP.ID,
	  DWA.ETL_TASK_PROGRAM.BIZ_SORTID,
	  DWA.ETL_TASK_PROGRAM.ON_DUTY,
	  DWA.ETL_TASK_LOG.PRGNAME,
	  count(DWA.HDP_JOB_MAP.job_id) job_cnt,  --天网任务的job数
	  sum(DWA.HDP_JOB_STAT.JOB_TOTAL_MAPS) JOB_TOTAL_MAPS,
	  sum(DWA.HDP_JOB_STAT.JOB_TOTAL_REDUCES) JOB_TOTAL_REDUCES,
	  sum(DWA.HDP_JOB_STAT.TOTAL_TIME) TOTAL_TIME,
	  sum(DWA.HDP_JOB_STAT.HDFS_BYTES_READ) HDFS_BYTES_READ,
	  sum(DWA.HDP_JOB_STAT.HDFS_BYTES_WRITTEN) HDFS_BYTES_WRITTEN,
	  sum(DWA.HDP_JOB_STAT.TOTAL_MAP_TIME) TOTAL_MAP_TIME,
	  sum(DWA.HDP_JOB_STAT.TOTAL_REDUCE_TIME) TOTAL_REDUCE_TIME,
	  sum(DWA.HDP_JOB_STAT.MAP_INPUT_RECORDS) MAP_INPUT_RECORDS,
	  sum(DWA.HDP_JOB_STAT.MAP_OUTPUT_RECORDS) MAP_OUTPUT_RECORDS, --new
	  sum(DWA.HDP_JOB_STAT.REDUCE_INPUT_RECORDS) REDUCE_INPUT_RECORDS,
	  sum(DWA.HDP_JOB_STAT.REDUCE_OUTPUT_RECORDS) REDUCE_OUTPUT_RECORDS, --new
	  trunc((DWA.ETL_TASK_LOG.edate-DWA.ETL_TASK_LOG.sdate)*24*60) time
	FROM
	  DWA.HDP_JOB_MAP,
	  DWA.ETL_TASK_PROGRAM,
	  DWA.ETL_TASK_LOG,
	  DWA.HDP_JOB_STAT,
	  DWA.ETL_TASK_BASELINE
	WHERE
	  ( DWA.HDP_JOB_STAT.JOB_ID=DWA.HDP_JOB_MAP.JOB_ID  )
	  AND  ( DWA.HDP_JOB_MAP.ID=DWA.ETL_TASK_LOG.ID  )
	  AND  ( DWA.ETL_TASK_LOG.ID=DWA.ETL_TASK_PROGRAM.ID  )
	  AND  ( DWA.ETL_TASK_PROGRAM.BASELINE_ID=DWA.ETL_TASK_BASELINE.ID  )
	  AND  
	  (
	   ( ( DWA.HDP_JOB_STAT.GMTDATE ) = trunc(sysdate)  )
	   AND
	   ( ( DWA.HDP_JOB_MAP.GMTDATE ) = trunc(sysdate)  )
	   AND
	   ( ( DWA.ETL_TASK_LOG.GMTDATE ) = trunc(sysdate)  )
	   AND DWA.ETL_TASK_PROGRAM.priority  in('xx','xxx') --##统计的时候输入自己的业务基线id  
	  )
	GROUP BY
	  DWA.ETL_TASK_BASELINE.name,
	  DWA.HDP_JOB_MAP.ID, 
	  DWA.ETL_TASK_PROGRAM.BIZ_SORTID, 
	  DWA.ETL_TASK_PROGRAM.ON_DUTY, 
	  DWA.ETL_TASK_LOG.PRGNAME,
	  (DWA.ETL_TASK_LOG.edate-DWA.ETL_TASK_LOG.sdate)*24*60
 ) t
 where time is not null  and job_cnt>10 --job数量,可以自己定义;

4. From表个数过多(节点入度过高)

select sky_id as 天网id,viewname as 显示名称,
	sum(cnt) as 来源表使用次数,count(cnt) as 来源表个数 
from(
	select sky_id,viewname,tab_name,on_duty,count(1) cnt 
	from(
		select distinct a.tab_name,c.sql_id,a.sub_sql_id,c.sky_id,e.viewname,e.on_duty
		from dwa.meta_tab a,dwa.meta_sqlsub b,
		(
			select * 
			from(
				select sky_id,sql_id,sql_src,
				row_number() over(partition by sky_id,length(sql_src) order by sql_id) rn 
				from dwa.meta_sqlfull)
			where rn=1
		) c,
		dwa.meta_col d,dwa.etl_task_program e
		where e.priority  in('xx','xxx') --##统计的时候输入自己的业务基线id  
			and e.appflag=0 and e.id=c.sky_id
			and a.sub_sql_id=b.sub_sql_id 
			and a.tab_id=d.tab_id and a.sub_sql_id=d.sub_sql_id and b.sqlfull_id=c.sql_id 
			and a.tab_name not like '%-%'  and b.sql_type='select'
		order by c.sky_id,c.sql_id,a.sub_sql_id
	)
	group by sky_id,viewname,tab_name,on_duty
	order by cnt desc
) 
group by sky_id,viewname
order by sum(cnt) desc;

5. Job倾斜情况

空值处理方法:

(1)直接过滤掉

(2)空值加上随机数分散到不同的reduce

解决方法一job2,方法二job1

select   a11.GMTDATE as  任务执行日期,
   a11.GROUP_NAME  as 业务线名称,
   a11.ID as 天网id,
   a11.SORT_ID as 云梯优先级,
   a11.NAME as 天网显示名称,
   a11.JOB_ID as job_id,
   a11.KEY_FLAG  是否关键节点任务,
   a11.USER_NAME  用户名,
   sum(a11.JOB_AVG_TIME)  WJXBFS1,
   sum(a11.JOB_MAX_TIME)  WJXBFS2,
   sum(a11.JOB_AVG_RECORDS)  WJXBFS3,
   sum(a11.JOB_MAX_RECORDS)  WJXBFS4
from   DWA.VIEW_HDP_JOB_STAT   a11
where gmtdate=date'2012-09-27'
	and group_name in ('xxxxx')
--业务线名称即天网任务配置里的“项目”
group by   a11.GMTDATE,
   a11.GROUP_NAME,
   a11.ID,
   a11.SORT_ID,
   a11.NAME,
   a11.JOB_ID,
   a11.KEY_FLAG,
   a11.USER_NAME ;

6. 相同输入字节数的任务抽取与合并

   数据源相同的任务,抽取相同的job进行合并

drop table gv_job_mapinput;
create table gv_job_mapinput as
select 
	id,prgname,job_id,MAP_INPUT_BYTES
from 
(
select 
 DWA.ETL_TASK_BASELINE.name groupname,
  DWA.HDP_JOB_MAP.ID,
  DWA.ETL_TASK_PROGRAM.BIZ_SORTID,
  DWA.ETL_TASK_PROGRAM.ON_DUTY,
  DWA.ETL_TASK_LOG.PRGNAME,
  DWA.HDP_JOB_MAP.job_id,  --天网任务的job数
  sum(DWA.HDP_JOB_STAT.JOB_TOTAL_MAPS) JOB_TOTAL_MAPS,
  sum(DWA.HDP_JOB_STAT.JOB_TOTAL_REDUCES) JOB_TOTAL_REDUCES,
  sum(DWA.HDP_JOB_STAT.TOTAL_TIME) TOTAL_TIME,
  sum(DWA.HDP_JOB_STAT.HDFS_BYTES_READ) HDFS_BYTES_READ,
  sum(DWA.HDP_JOB_STAT.HDFS_BYTES_WRITTEN) HDFS_BYTES_WRITTEN,
  sum(DWA.HDP_JOB_STAT.TOTAL_MAP_TIME) TOTAL_MAP_TIME,
  sum(DWA.HDP_JOB_STAT.TOTAL_REDUCE_TIME) TOTAL_REDUCE_TIME,
  sum(DWA.HDP_JOB_STAT.MAP_INPUT_RECORDS) MAP_INPUT_RECORDS,
  sum(DWA.HDP_JOB_STAT.MAP_INPUT_BYTES) MAP_INPUT_BYTES,
  sum(DWA.HDP_JOB_STAT.MAP_OUTPUT_RECORDS) MAP_OUTPUT_RECORDS, --new
  sum(DWA.HDP_JOB_STAT.REDUCE_INPUT_RECORDS) REDUCE_INPUT_RECORDS,
  sum(DWA.HDP_JOB_STAT.REDUCE_OUTPUT_RECORDS) REDUCE_OUTPUT_RECORDS, --new
  trunc((DWA.ETL_TASK_LOG.edate-DWA.ETL_TASK_LOG.sdate)*24*60) time
FROM
  DWA.HDP_JOB_MAP,
  DWA.ETL_TASK_PROGRAM,
  DWA.ETL_TASK_LOG,
  DWA.HDP_JOB_STAT,
  DWA.ETL_TASK_BASELINE
WHERE
  ( DWA.HDP_JOB_STAT.JOB_ID=DWA.HDP_JOB_MAP.JOB_ID  )
  AND  ( DWA.HDP_JOB_MAP.ID=DWA.ETL_TASK_LOG.ID  )
  AND  ( DWA.ETL_TASK_LOG.ID=DWA.ETL_TASK_PROGRAM.ID  )
  AND  ( DWA.ETL_TASK_PROGRAM.BASELINE_ID=DWA.ETL_TASK_BASELINE.ID  )
  AND  
  (
   ( ( DWA.HDP_JOB_STAT.GMTDATE ) = trunc(sysdate)  )
   AND
   ( ( DWA.HDP_JOB_MAP.GMTDATE ) = trunc(sysdate)  )
   AND
   ( ( DWA.ETL_TASK_LOG.GMTDATE ) = trunc(sysdate)  )
   AND
   DWA.ETL_TASK_PROGRAM.priority  in('xx','xxx')
  --##统计的时候输入自己的业务基线id
  )
GROUP BY
  DWA.ETL_TASK_BASELINE.name,
  DWA.HDP_JOB_MAP.ID, 
  DWA.ETL_TASK_PROGRAM.BIZ_SORTID, 
  DWA.ETL_TASK_PROGRAM.ON_DUTY, 
  DWA.ETL_TASK_LOG.PRGNAME,
  DWA.HDP_JOB_MAP.job_id,
  (DWA.ETL_TASK_LOG.edate-DWA.ETL_TASK_LOG.sdate)*24*60
  )
order by MAP_INPUT_RECORDS desc ,job_id;
 
select * from gv_job_mapinput 
where id exists (
	select id from 
	(select id,prgname,count(job_id) cnt from gv_job_mapinput group by id,prgname)
	where cnt =1 
)
order by MAP_INPUT_BYTES desc;

7. 多个任务只有一个共同的父任务

drop table gvora_view_relation;
create table gvora_view_relation as 
select a.id,a.viewname,a.on_duty,a.sourceid,a.priority,a.parentid,
	b.viewname parentviewname,b.on_duty pon_duty,b.sourceid psourceid,b.priority p_priority 
from(
	select a.id,b.viewname,b.on_duty,b.sourceid,b.priority,a.parentid from 
	dwa.etl_task_relation a,
	dwa.etl_task_program b
	where a.id=b.id
) a,
dwa.etl_task_program b
where a.parentid=b.id;

select a.id as 天网id,a.viewname as 显示名称,rudu,chudu 
from(
	select id,viewname,count(1) rudu from gvora_view_relation
	where priority  in('xx','xxx')
	--##统计的时候输入自己的业务基线id
	group by id,viewname
) a,
(
	select parentid,parentviewname,count(1) chudu from gvora_view_relation
	where priority  in('xx','xxx')
	 --##统计的时候输入自己的业务基线id
	group by parentid,parentviewname
) b
where a.id=b.parentid
order by rudu +chudu desc;


 

 

 

你可能感兴趣的:(优化,hive)