1、HQL子查询别名问题
报错:hive query failed cannot recognize input near '(' 'select' '*' in joinSource
备注:子查询需要加上别名 否则报错。
HQL的书写,select * from (select * from table) ;
应改为:select * from (select * from table) a,执行成功。
2、left join 若后表为前表的基础上进一步查询的结果,不能直接用前表的临时命名,可用with table1 as (select xxxx) 来实现同一查询结果的复用;否则sql语句是执行到哪里,才有一遍结果,故后表查询时,需要把前表查询语句再执行一遍。
报错:table name "a" must be "db.table" format
本质:不能写 (select xx from xx where xx) a left join (select yy from a where yy) b
执行到第二个a时,a表并不实际存在!!
解决:
【法一】再写一遍子查询
(select xx from xx where xx) a left join (select yy from (select xx from xx where xx) a where yy) b
【法二】with a as _ 实现同一查询结果多次复用
with a as (select xx from xx where xx)
a left join (select yy from a where yy) b
3、hive使用正则报错
报错:Hive query failed, Error while compiling statement: FAILED: SemanticException [Error 10002]: Line 28:18 Invalid column reference 'median.*'
语句:select a.*, b.`median.*` from a
left join
(
select brand, region
, percentile(rscore, 0.5) as median_r
,percentile(fscore, 0.5) as median_f
,percentile(mscore, 0.5) as median_m
from a
group by brand, region
) b
on a.brand=b.brand and a.region=b.region
b.`median.*` 代表b表中,以“median”开头的字段
解决:需要在代码段前加上:
set hive.support.quoted.identifiers=none;
4、insert overwrite 报错
报错:Caused by: MetaException(message:Exception thrown when executing query : SELECT DISTINCT 'org.apache.hadoop.hive.metastore.model.MPartition' AS NUCLEUS_TYPE
,A0
.CREATE_TIME
,A0
.LAST_ACCESS_TIME
,A0
.PART_NAME
,A0
.PART_ID
FROM PARTITIONS
A0
LEFT OUTER JOIN TBLS
B0
ON A0
.TBL_ID
= B0
.TBL_ID
LEFT OUTER JOIN DBS
C0
ON B0
.DB_ID
= C0
.DB_ID
WHERE B0
.TBL_NAME
= ? AND C0
.NAME
= ? AND A0
.PART_NAME
= ?)
原因:由于partition字段异常字符导致,分区字段不支持中文
根本原因:源表字段值和输出分区值之间的关系是根据位置,而不是命名来匹配的!!!
例子:
insert overwrite table employees
partition (country, state)
select ..., se.cnty, se.st
from staged_employees se;
5、cannot recognize input near 报错
报错:Error: Error while compiling statement: FAILED: ParseException line 10:8 cannot recognize input near '(' 'select' 'statis_id' in joinSource (state=42000,code=40000)
原因:大多数情况是没有对子查询/join的子表 命别名
实际原因:所以我一直纠结在别名的检查上,谁能想到是 变量名写错 了呢……T^T
6、Error: org.apache.hive.service.cli.HiveSQLException: Error while processing statement: FAILED: Execution Error, return code 1 from org.apache.hadoop.hive.ql.exec.mr.MapRedTask
解决1:set mapred.reduce.tasks=10; -> set mapreduce.job.reduces=10; (hive2)
7、Query failed (#20200429_072245_00729_nh2wp): Error opening Hive split s3://bi-oor-com/warehouse/analysis/inter/union_inter_dw.db/dim_mkt_app_info/dayno=29991231/000003_0 (offset=0, length=25082323): Malformed ORC file. Can not read SQL type integer from ORC stream .system_id of type STRING [s3://bi-oor-com/warehouse/analysis/inter/union_inter_dw.db/dim_mkt_app_info/dayno=29991231/000003_0]
解决:select字段过多,不能 select * ,改成指定某几个字段
8、SemanticException Column app_id Found in more than One Tables/Subqueries (state=42000,code=40000)
一般是因为join的时候,含同字段两张表,没有命别名
但我遇到一个诡异的情况:
A union all B
B=aa left join bb left join cc (均已指定别名)
问题出在,aa中,有 not in (select xx_id from A) 的情况,把该条件拿掉后,不再报上述错误
解决:搞不清为啥,直接把B表逻辑with tmp as 了,不再报此错误。
Hive对in的支持有点问题,写出来的sql一直报错,分开跑也报错,一会儿 column more than one tables,一会儿又说语法错误,一会儿又能跑。。。
所以,以后遇到 in / not in 的查询需求,统一用 join 替换,如下:
-
in:
select A.x_id, A.xxx,……, from (table) A left join (table) B on A.x_id=B.x_id where B.x_id is not null; -
not in:
select A.x_id, A.xxx,……, from (table) A left join (table) B on A.x_id=B.x_id where B.x_id is null;
9、hive insert overwrite / into 都没报错,但是数据没写进去
不知道原因>.<
有可能是分区字段不能含NULL,我把分区字段筛选了不为NULL,就好像写进去了?——<验证不是这个问题>
现在怀疑是set hive 某些参数,导致本应有数据,但select出来就没有,所以也就没有insert进去了。——对!!!去掉参数设置就有数据了!!!是参数出了问题!!!有可能是以下几个参数:
- set hive.strict.checks.cartesian.product = false;
- set hive.mapred.mode = nonstrict;
- set hive.support.quoted.identifiers = none;
明天来排查具体是哪个参数的问题。真的太坑了,巨坑,一切的坑都源于省事儿用 in, 用 in 就要求开启这几个配置,开了就忘了关了,就一直写不进去数据(实际上是就没查出来数据)!!!巨坑!!!巨TM坑!!!明天一定要来搞搞清楚!!!