如何在运行时配置flink作业？

Question

是否可以在运行时配置flink应用程序？例如，我有一个流应用程序，该应用程序读取输入，进行一些转换，然后过滤掉低于某个阈值的所有元素。但是，我希望此阈值在运行时可配置，这意味着我可以更改此阈值而不必重新启动flink作业。示例代码：

DataStream<MyModel> myModelDataStream = // get input ...
                // do some stuff ...
                .filter(new RichFilterFunction<MyModel>() {
                    @Override
                    public boolean filter(MyModel value) throws Exception {
                        return value.someValue() > someGlobalState.getThreshold();
                    }
                })
                // write to some sink ...

DataStream<MyConfig> myConfigDataStream = // get input ...
                // ...
                .process(new RichProcessFunction<MyConfig>() {
                      someGlobalState.setThreshold(MyConfig.getThreshold());
                })
                // ...

是否有可能实现这一目标？例如，可以通过配置流更改的全局状态。

Answer 1

是的，您可以使用BroadcastProcessFunction。大概是这样的：

    MapStateDescriptor<Void, Threshold> bcStateDescriptor = new MapStateDescriptor<>(
    "thresholds", Types.VOID, Threshold.class);

    DataStream<MyModel> myModelDataStream = // get input ...
    DataStream<Threshold> thresholds = // get input...
    BroadcastStream<Threshold> controlStream = thresholds.broadcast(bcStateDescriptor);

    DataStream<MyModel> result = myModelDataStream
      .connect(controlStream)
      .process(new MyFunction());

    public class MyFunction extends BroadcastProcessFunction<MyModel, Long, MyModel> {    
        @Override
        public void processBroadcastElement(Threshold newthreshold, Context ctx, Collector<MyModel> out) {
            BroadcastState<Void, Threshold> bcState = ctx.getBroadcastState(new MapStateDescriptor<>("thresholds", Types.VOID, Threshold.class));  
            bcState.put(null, newthreshold);
        }

        @Override
        public void processElement(MyModel model, Collector<MyModel> out) {
            Threshold threshold = ctx.getBroadcastState(new MapStateDescriptor<>("threshold", Types.VOID, Threshold.class)).get(null);
            if (threshold.value() == null || model.getData() > threshold.value()) {
                out.collect(model);
            }
        }
    }