如何为火花流检查点设置检查点间隔？

Question

我想根据官方文档为我的python spark流脚本设置checkpoint Interval：

对于需要RDD检查点的有状态转换，默认间隔是批处理间隔的倍数，至少为10秒。 可以使用dstream.checkpoint（checkpointInterval）进行设置。 通常，DStream的滑动间隔的5-10倍的检查点间隔是尝试的良好设置。

我的脚本：

import sys

from pyspark import SparkContext
from pyspark.streaming import StreamingContext
from pyspark.streaming.kafka import KafkaUtils

def functionToCreateContext():
    sc = SparkContext(appName="PythonStreamingDirectKafkaWordCount")
    ssc = StreamingContext(sc, 6)
    ssc.checkpoint("./checkpoint")
    kvs = KafkaUtils.createDirectStream(ssc, ['test123'], {"metadata.broker.list": "localhost:9092"})

    kvs = kvs.checkpoint(60) #set the checkpoint interval

    lines = kvs.map(lambda x: x[1])
    counts = lines.flatMap(lambda line: line.split(" ")) \
        .map(lambda word: (word, 1)) \
        .reduceByKey(lambda a, b: a+b)
    counts.pprint()
    return ssc

if __name__ == "__main__":
    ssc = StreamingContext.getOrCreate("./checkpoint", functionToCreateContext)

    ssc.start()
    ssc.awaitTermination()

运行脚本后的输出：

16/05/25 17:49:03 INFO DirectKafkaInputDStream: Slide time = 6000 ms
16/05/25 17:49:03 INFO DirectKafkaInputDStream: Storage level = StorageLevel(false, false, false, false, 1)
16/05/25 17:49:03 INFO DirectKafkaInputDStream: Checkpoint interval = null
16/05/25 17:49:03 INFO DirectKafkaInputDStream: Remember duration = 120000 ms
16/05/25 17:49:03 INFO DirectKafkaInputDStream: Initialized and validated org.apache.spark.streaming.kafka.DirectKafkaInputDStream@1be80174
16/05/25 17:49:03 INFO PythonTransformedDStream: Slide time = 6000 ms
16/05/25 17:49:03 INFO PythonTransformedDStream: Storage level = StorageLevel(false, true, false, false, 1)
16/05/25 17:49:03 INFO PythonTransformedDStream: Checkpoint interval = 60000 ms
16/05/25 17:49:03 INFO PythonTransformedDStream: Remember duration = 120000 ms
16/05/25 17:49:03 INFO PythonTransformedDStream: Initialized and validated org.apache.spark.streaming.api.python.PythonTransformedDStream@69f9a089
16/05/25 17:49:03 INFO PythonTransformedDStream: Slide time = 6000 ms
16/05/25 17:49:03 INFO PythonTransformedDStream: Storage level = StorageLevel(false, false, false, false, 1)
16/05/25 17:49:03 INFO PythonTransformedDStream: Checkpoint interval = null
16/05/25 17:49:03 INFO PythonTransformedDStream: Remember duration = 6000 ms
16/05/25 17:49:03 INFO PythonTransformedDStream: Initialized and validated org.apache.spark.streaming.api.python.PythonTransformedDStream@d97386a
16/05/25 17:49:03 INFO PythonTransformedDStream: Slide time = 6000 ms
16/05/25 17:49:03 INFO PythonTransformedDStream: Storage level = StorageLevel(false, false, false, false, 1)
16/05/25 17:49:03 INFO PythonTransformedDStream: Checkpoint interval = null
16/05/25 17:49:03 INFO PythonTransformedDStream: Remember duration = 6000 ms
16/05/25 17:49:03 INFO PythonTransformedDStream: Initialized and validated org.apache.spark.streaming.api.python.PythonTransformedDStream@16c474ad
16/05/25 17:49:03 INFO ForEachDStream: Slide time = 6000 ms
16/05/25 17:49:03 INFO ForEachDStream: Storage level = StorageLevel(false, false, false, false, 1)
16/05/25 17:49:03 INFO ForEachDStream: Checkpoint interval = null
16/05/25 17:49:03 INFO ForEachDStream: Remember duration = 6000 ms
..........

DStream检查点间隔仍为空。 对此有何看法？

Answer 1

在创建流后，尝试将此行向下移动几行： ssc.checkpoint("./checkpoint")

基本上，在完全准备好流之后执行检查点。