Question

我正在Spark Streaming应用程序中读取来自Kafka的消息。

SparkBatch持续时间：15秒。 SparkWindow：60秒。

var dstream = KafkaUtils.createDirectStream() // ignore the argumtns
var windowedStream = dstream.window(SparkWindow)
// delete data from REDIS
windowedStream.foreachRDD(rdd -> {
        if(!rdd.isEmpty()) {
                JavaFutureAction<Void> v = rdd.foreachPartitionAsync(t -> {
                // collect error data across partitions and write those to REDIS
        })// foreachPartitionAsync ends
    }
})
// fetchFromREDISAndProcess() --Once foreachRDD ends. fetch error data from REDIS and process them

有一个矛盾之处，我必须首先在火花窗口中从每个分区和RDD收集错误记录，然后在驱动程序上对其进行处理。

我将在每个Spark Window中获得4个RDD。

问题：我想在每个窗口之后从REDIS中读取数据，并在继续下一个窗口之前对其进行处理。有没有办法确保每次火花窗口结束时都执行我的代码？

Answer 1

您可以使用以下逻辑：::

var dstream = KafkaUtils.createDirectStream() // ignore the argumtns
var windowedStream = dstream.window(SparkWindow)
// delete data from REDIS
var partitions = 4;
var currentPart = 0;

windowedStream.foreachRDD(rdd -> {
        if(!rdd.isEmpty()) {
                JavaFutureAction<Void> v = rdd.foreachPartitionAsync(t -> {
                // collect error data across partitions and write those to REDIS
        })// foreachPartitionAsync ends

        if(++currentPart % partitions == 0)    //It will be true at every 4th RDD where window will end.
            //Read data from REDIS and process here as after this new window will start.
    }
})
// fetchFromREDISAndProcess() --Once foreachRDD ends. fetch error data from REDIS and process them

如何知道火花流窗口中的RDD是否已完成火花工作？

1 个答案: