Flink DataStream 如何将一个自定义的 POJO 组合成另一个 DataStream

Question

I want convert a DataStream to DataStream with schema info

input

args[0] DataStream

{"fields":["China","Beijing"]}

args[1] schema

message spark_schema {
  optional binary country (UTF8);
  optional binary city (UTF8);
}

expect output

{"country":"china", "city":"beijing"}

my code like this

public DataStream<String> convert(DataStream source, MessageType messageType) {

        SingleOutputStreamOperator<String> dataWithSchema = source.map((MapFunction<Row, String>) row -> {
            JSONObject data = new JSONObject();
            this.fields = messageType.getFields().stream().map(Type::getName).collect(Collectors.toList());
            for (int i = 0; i < fields.size(); i++) {
                data.put(fields.get(i), row.getField(i));
            }
            return data.toJSONString();
        });
        return dataWithSchema;
    }

Exception Errors

Exception in thread "main" org.apache.flink.api.common.InvalidProgramException: Object com.xxxx.ParquetDataSourceReader$$Lambda$64/1174881426@d78795 is not serializable
    at org.apache.flink.api.java.ClosureCleaner.ensureSerializable(ClosureCleaner.java:180)
    at org.apache.flink.streaming.api.environment.StreamExecutionEnvironment.clean(StreamExecutionEnvironment.java:1823)
    at org.apache.flink.streaming.api.datastream.DataStream.clean(DataStream.java:188)
    at org.apache.flink.streaming.api.datastream.DataStream.map(DataStream.java:590)

But the code below works fine

public DataStream<String> convert(DataStream source, MessageType messageType) {
        if (this.fields == null) {
            throw new RuntimeException("The schema of AbstractRowStreamReader is null");
        }

        List<String> field = messageType.getFields().stream().map(Type::getName).collect(Collectors.toList());
        SingleOutputStreamOperator<String> dataWithSchema = source.map((MapFunction<Row, String>) row -> {
            JSONObject data = new JSONObject();
            for (int i = 0; i < field.size(); i++) {
                data.put(field.get(i), row.getField(i));
            }
            return data.toJSONString();
        });
        return dataWithSchema;
    }

The Flink map operator how to combine a external complex POJO?

Answer 1

For Flink to distribute the code across tasks, the code needs to be completely Serializable . In your first example, it isn't; in the second it is. In particular, Type::getName will generate a lambda that is not Serializable .

To get a lambda that is Serializable , you need to explicitly cast it to a serializable interface (eg Flink MapFunction ) or use cast it with (Serializable & Function)

Since the second one is also saving computations, it would be better in any case. Convert will be executed only once during job compilation, while DataStream#map is called for each record. If that is not clear, I recommend to execute it in an IDE and play around with breakpoints.