Hi Alexander,

Please try with turning on the union type:

ALTER SESSION SET `exec.enable_union_type` = true;

Kind regards
Vitalii

2016-07-08 10:50 GMT+00:00 Holy Alexander <[email protected]>:

> My JSON data looks - simplified - like this
>
> {"ID":1,"a":"some text"}
> {"ID":2,"a":"some text","b":"some other text"}
> {"ID":3,"a":"some text"}
>
> Column b is only physically serialized when it is not null.
> It is the equivalent of a NULLable VARCHAR() column in SQL.
>
> I run queries like these:
>
> SELECT b
> FROM dfs.`D:\MyData\test.json`
> WHERE b IS NOT NULL
>
> And normally all is fine.
> However, among my thousands of data files, I have two files where the
> first occurrence of b happens a few thousand records down the file.
> These two data files would look like this:
>
> {"ID":1,"a":"some text"}
> {"ID":2,"a":"some text"}
> ... 5000 more records without column b ...
> {"ID":5002,"a":"some text","b":"some other text"}
> {"ID":5003,"a":"some text"}
>
> In this case, my simple SQL query above fails:
>
> [30027]Query execution error. Details:[
> DATA_READ ERROR: Error parsing JSON - You tried to write a VarChar type
> when you are using a ValueWriter of type NullableIntWriterImpl.
> File  /D:/MyData/test.json
> Record 5002 Fragment ...
>
> It seems that the Schema inference mechanism of Drill only samples a
> certain amount of bytes (or records) to determine the schema.
> If the first occurrence of a schema detail happens to far down things go
> boom.
>
> I am now looking for a sane way to work around this.
> Preferred by extending the query and not by altering my massive amounts of
> data.
>
> BTW, I tried altering the data by chaning the first line:
> {"ID":1,"a":"some text","b":null}
> does not help.
>
> Of course, changing the first line to
> {"ID":1,"a":"some text","b":""}
> solves the problem, but this is not a practical solution.
>
> Any help appreciated.
> Alexander
>

Reply via email to