Question

存储在Hive中的文件：

[
  {
    "occupation": "guitarist",
    "fav_game": "football",
    "name": "d1"
  },
  {
    "occupation": "dancer",
    "fav_game": "chess",
    "name": "k1"
  },
  {
    "occupation": "traveller",
    "fav_game": "cricket",
    "name": "p1"
  },
  {
    "occupation": "drummer",
    "fav_game": "archery",
    "name": "d2"
  },
  {
    "occupation": "farmer",
    "fav_game": "cricket",
    "name": "k2"
  },
  {
    "occupation": "singer",
    "fav_game": "football",
    "name": "s1"
  }
]

hadoop中的CSV文件：

name,age,city
d1,23,delhi
k1,23,indore
p1,23,blore
d2,25,delhi
k2,30,delhi
s1,25,delhi

我单独询问他们，工作正常。然后，我尝试了加入查询：

select * from hdfs.`/demo/distribution.csv` d join hive.demo.`user_details` u on d.name = u.name

我遇到了以下问题：

org.apache.drill.common.exceptions.UserRemoteException：SYSTEM ERROR：DrillRuntimeException：Join仅支持1.之间的隐式强制转换。数字数据2. Varchar，Varbinary数据3.日期，时间戳数据左类型：INT，右类型： VARCHAR。添加显式强制转换以避免此错误片段0：0 [错误ID：bITdb9c8-fb35-4ef8-a1c0-31b68ff7ae8d on IMPETUS-DSRV03.IMPETUS.CO.IN:31010]

Answer 1

请参阅此https://drill.apache.org/docs/data-type-conversion/ 我们需要做明确的类型转换来处理这种情况。

考虑我们有一个JSON文件employee.json和一个csv文件sample.csv。为了同时查询两者，在一个查询中我们需要进行类型转换。

0: jdbc:drill:zk=local> select emp.employee_id, dept.department_description, phy.columns[2], phy.columns[3] FROM cp.`employee.json` emp , cp.`department.json` dept, dfs.`/tmp/sample.csv` phy where CAST(emp.employee_id AS INT) =  CAST(phy.columns[0] AS INT) and emp.department_id = dept.department_id;

这里我们进行类型转换 CAST（emp.employee_id AS INT）= CAST（phy.columns [0] AS INT），以便相等不会失败。

有关详细信息，请参阅此处： - http://www.devinline.com/2015/11/apache-drill-setup-and-SQL-query-execution.html#multiple_src

Answer 2

你需要施放，即使默认它已经采取了varchar。试试这个：

select * from hdfs.`/demo/distribution.csv` d join hive.demo.`user_details` u on cast(d.name as VARCHAR) = cast(u.name as VARCHAR)

但是你不能直接从csv引用列名。你需要考虑列[0]的名称。

在apache钻取中的JOIN查询中出现问题

2 个答案: