如何更新/刷新 Flink 应用程序中的参数

Question

I have a Flink application on AWS Kinesis Analytics service. I need to filter some values on a data stream based on a threshold. Also, I'm passing the threshold parameter using AWS Systems Manager Parameter Store service. For now, I got this:

• In my Main class:

val threshold: Int = ssmParameter.getParameterRequest(ssmClient, "/kinesis/threshold").toInt

val kinesis_deserialization_schema = new KinesisDeserialization[ID]
            val KinesisConsumer = new FlinkKinesisConsumer[ID](
                "Data-Stream",
                kinesis_deserialization_schema,
                consumerProps
            )
            val KinesisSource = env.addSource(KinesisConsumer).name(s"Kinesis Data")
val valid_data = KinesisSource
          .filter(new MyFilter[ID](threshold))
          .name("FilterData")
          .uid("FilterData")

• Filter class:

import cl.mydata.InputData
import org.apache.flink.api.common.functions.FilterFunction

class MyFilter[ID <: InputData](
                                  threshold: Int
                                ) extends FilterFunction[ID] {
  override def filter(value: ID): Boolean = {
      value.myvalue > threshold
    }
  }
}

This works fine, the thing is that I need to update the threshold parameter every hour, because that value can be changed by my client.

Answer 1

Perhaps you can implement the ProcessingTimeCallback interface in the MyFilter class, which supports timer operations, and you can update the threshold in the onProcessingTime function

public class MyFilter extends FilterFunction<...> implements ProcessingTimeCallback { 
    int threshold;

    @Override
    public void open(Configuration parameters) throws Exception {
        scheduler.scheduleAtFixedRate(this, 1, 1, TimeUnit.HOURS);

        final long now = getProcessingTimeService().getCurrentProcessingTime();
        getProcessingTimeService().registerTimer(now + 3600000, this);
    }

    @Override
    public boolean filter(IN xxx) throws Exception {
        return xxx > threshold;
    }

    @Override
    public void onProcessingTime(long timestamp) throws Exception {
        threshold = XXXX;

        final long now = getProcessingTimeService().getCurrentProcessingTime();
        getProcessingTimeService().registerTimer(now + 3600000, this);
    }
}

Answer 2

You could turn the FilterFunction into a BroadcastProcessFunction , and broadcast new thresholds as they become available.