Row is very special data type, because Flink cannot extract the field types automatically based on Java generics. By default it is serialized by Kryo, you need to specify the field types using Typles.ROW(Types.STRING, ...) and pass this information in your `.returns()` methods instead of `Row.class`.

Lambdas are sometimes a problem for Flink, see https://ci.apache.org/projects/flink/flink-docs-release-1.4/dev/java8.html.

In your example it might makes sense to use the Table API for pre-processing. The Table API has a built-in CSV input format and deals with all the type handling for you. You can convert a table back to DataStream.


Am 05.08.17 um 13:49 schrieb Егор Литвиненко:
> No errors, and no
data in table.

2017-08-05 14:49 GMT+03:00 Егор Литвиненко <e.v.litvinenk...@gmail.com <mailto:e.v.litvinenk...@gmail.com>>:


    I try to test Flink and have a problems.
    Why this example doesn't work -
    Logs - https://pastebin.com/iuBZhfeG <https://pastebin.com/iuBZhfeG>
    No errors, and no

    One more question. Examples intentionally use anonymous classes,
    instead of lambdas. Because with lambda it also doesn't work. I
    did small investigation and found out that Flink has different
    processing for lambdas. And in one moment when Flink processes
    types, they are empty.  Do you have full lambdas support?

    In best regards, Egor Litvinenko.

Reply via email to