Hive解析多重嵌套JSON数组

最近在工作中遇到一个很难解析的JSON，他是一个嵌套的JSON数组的JSON，要使用Hive来进行解析，用Presto写了一次，逻辑就很清晰，因为Presto自带了JSON数据类型，转换数组就很方便，而Hive解析完JSON数组后是一个字符串，只能使用split方法来对string类型的数据进行切分，所以如果遇到多层嵌套的数组，要注意切分方法，不然就会乱套。

需要解析的JSON

{
	"base": {
		"code": "xm",
		"name": "project"
	},
	"list": [{
		"ACode": "cp1",
		"AName": "Product1",
		"BList": [{
			"BCode": "gn1",
			"BName": "Feature1"
		}, {
			"BCode": "gn2",
			"BName": "Feature2"
		}]
	}, {
		"ACode": "cp2",
		"AName": "Product2",
		"BList": [{
			"BCode": "gn1",
			"BName": "Feature1"
		}]
	}]
}

解析出来的结果应该如下表所示

code	name	ACode	Aname	Bcode	Bname
xm	project	cp1	Product1	gn1	Feature1
xm	project	cp1	Product1	gn2	Feature2
xm	project	cp2	Product2	gn1	Feature1

解决方案

首先使用get_json_object方法，把需要解析的数组解析出来，然后使用regexp_replace将}]},{替换成}]}||{，然后再使用split方法对||进行分割，分割成数组后，使用lateral view explode方法对其进行展开成多列即刻。

SELECT
    code
  , name
  , ai.ACode
  , ai.AName
  , bi.BCode
  , bi.BName
FROM
    (
        SELECT
            get_json_object(t.value, '$.base.code') AS code
          , get_json_object(t.value, '$.base.name') AS name
          , get_json_object(t.value, '$.list')      AS list
        FROM
            (
                SELECT
                    '{"base":{"code":"xm","name":"project"},"list":[{"ACode":"cp1","AName":"Product1","BList":[{"BCode":"gn1","BName":"Feature1"},{"BCode":"gn2","BName":"Feature2"}]},{"ACode":"cp2","AName":"Product2","BList":[{"BCode":"gn1","BName":"Feature1"}]}]}' as value
            )
            t
    )
    t lateral view explode(split(regexp_replace(regexp_extract(list,'^\[(.+)\]$',1),'\}\]\}\,\{', '\}\]\}\|\|\{'),'\|\|')) list as a 
	lateral view json_tuple(a,'ACode','AName','BList') ai as ACode
    , AName
    , BList lateral view explode(split(regexp_replace(regexp_extract(BList,'^\[(.+)\]$',1),'\}\,\{', '\}\|\|\{'),'\|\|')) BList as b 
    lateral view json_tuple(b,'BCode','BName') bi as BCode
    , BName
;

执行完

xm	project	cp1	Product1	gn1	Feature1
xm	project	cp1	Product1	gn2	Feature2
xm	project	cp2	Product2	gn1	Feature1
Time taken: 0.787 seconds, Fetched: 3 row(s)

总结

1. 尝试切分为数组后，使用lateral view posexplode方案，逐层解析，但这样会导致笛卡尔。所以必须一次性全部解析好，而不是套用多个子查询逐层解析；
2. 使用OUTER字段，能使LATERAL VIEW不忽略NULL

include OUTER in the query to get rows with NULL values

something like,

select *  FROM table LATERAL VIEW OUTER explode (  split (  email  ,','  ) ) email AS email_id;