闪烁键控流密钥为空

Question

我正在尝试对Flink中的KeyedStream执行映射操作：

stream.map(new JsonToMessageObjectMapper())
                    .keyBy("keyfield")
                    .map(new MessageProcessorStateful())

JsonToObjectMapper运算符的输出是类MessageObject的POJO，它有一个字符串字段'keyfield'。然后在该字段上键入流。

public class MessageAdProcessorStateful extends RichMapFunction<MessageObject, Tuple2<String, String>> {

    private transient MapState<String, Tuple2<Tuple3<String, String, String>, Tuple2<Double, Long>>> state;
    ...
    @Override
    public void open(Configuration config) throws Exception {
        MapStateDescriptor<String, Tuple2<Tuple3<String, String, String>, Tuple2<Double, Long>>> descriptor =
                    new MapStateDescriptor<>(
                        "state",                                                                                     // the state name
                            TypeInformation.of(new TypeHint<String>() {}),
                            TypeInformation.of(new TypeHint<Tuple2<Tuple3<String, String, String>, Tuple2<Double, Long>>>() {}) ); // type information
                    state = getRuntimeContext().getMapState(descriptor);

        state.put(...); // Insert a key, value here. Exception here!

    }
}

代码抛出NullPointer异常：

Caused by: java.lang.NullPointerException: No key set. This method should not be called outside of a keyed context.
    at org.apache.flink.util.Preconditions.checkNotNull(Preconditions.java:75)
    at org.apache.flink.runtime.state.heap.CopyOnWriteStateTable.checkKeyNamespacePreconditions(CopyOnWriteStateTable.java:528)
    at org.apache.flink.runtime.state.heap.CopyOnWriteStateTable.computeHashForOperationAndDoIncrementalRehash(CopyOnWriteStateTable.java:722)
    at org.apache.flink.runtime.state.heap.CopyOnWriteStateTable.get(CopyOnWriteStateTable.java:265)
    at org.apache.flink.runtime.state.heap.CopyOnWriteStateTable.get(CopyOnWriteStateTable.java:306)
    at org.apache.flink.runtime.state.heap.HeapMapState.put(HeapMapState.java:75)
    at org.apache.flink.runtime.state.UserFacingMapState.put(UserFacingMapState.java:52)
    at org.myorg.quickstart.MessageStreamProcessor$MessageAdProcessorStateful.open(MessageStreamProcessor.java:226)
    at org.apache.flink.api.common.functions.util.FunctionUtils.openFunction(FunctionUtils.java:36)
    at org.apache.flink.streaming.api.operators.AbstractUdfStreamOperator.open(AbstractUdfStreamOperator.java:102)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.openAllOperators(StreamTask.java:393)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:254)
    at org.apache.flink.runtime.taskmanager.Task.run(Task.java:718)
    at java.lang.Thread.run(Thread.java:748)

似乎其中一个KeyedStream的keyedState中的键为null，尽管我已经验证了'keyfield'始终是有效字符串。根据Flink文档，Rest似乎是正确的。知道这是怎么回事吗？

Answer 1

问题是您试图访问open()方法中的键控状态。

Keyed state为每个密钥维护一个状态实例。在您的示例中，您使用的是mapstate。因此每个键都有一个mapstate实例。在访问状态时，您将始终获得与当前处理的记录的键相对应的状态实例。在MapFunction（如您的示例）中，这将是传递给Map()方法的记录。

由于未使用记录调用open()，因此open()中的当前键为null并且无法访问键控状态。