python - pySpark Kafka Direct Streaming update Zookeeper / Kafka Offset

Question

currently I'm working with Kafka / Zookeeper and pySpark (1.6.0). I have successfully created a kafka consumer, which is using the KafkaUtils.createDirectStream().

There is no problem with all the streaming, but I recognized, that my Kafka Topics are not updated to the current offset, after I have consumed some messages.

Since we need the topics updated to have a monitoring here in place this is somehow weird.

In the documentation of Spark I found this comment:

   offsetRanges = []

     def storeOffsetRanges(rdd):
         global offsetRanges
         offsetRanges = rdd.offsetRanges()
         return rdd

     def printOffsetRanges(rdd):
         for o in offsetRanges:
             print "%s %s %s %s" % (o.topic, o.partition, o.fromOffset, o.untilOffset)

     directKafkaStream
         .transform(storeOffsetRanges)
         .foreachRDD(printOffsetRanges)

You can use this to update Zookeeper yourself if you want Zookeeper-based Kafka monitoring tools to show progress of the streaming application.

Here is the documentation: http://spark.apache.org/docs/1.6.0/streaming-kafka-integration.html#approach-2-direct-approach-no-receivers

I found a solution in Scala, but I can't find an equivalent for python. Here is the Scala example: http://geeks.aretotally.in/spark-streaming-kafka-direct-api-store-offsets-in-zk/

Question

But the question is, how I'm able to update the zookeeper from that point on?

Answer 1

I write some functions to save and read Kafka offsets with python kazoo library.

First function to get singleton of Kazoo Client:

ZOOKEEPER_SERVERS = "127.0.0.1:2181"

def get_zookeeper_instance():
    from kazoo.client import KazooClient

    if 'KazooSingletonInstance' not in globals():
        globals()['KazooSingletonInstance'] = KazooClient(ZOOKEEPER_SERVERS)
        globals()['KazooSingletonInstance'].start()
    return globals()['KazooSingletonInstance']

Then functions to read and write offsets:

def read_offsets(zk, topics):
    from pyspark.streaming.kafka import TopicAndPartition

    from_offsets = {}
    for topic in topics:
        for partition in zk.get_children(f'/consumers/{topic}'):
            topic_partion = TopicAndPartition(topic, int(partition))
            offset = int(zk.get(f'/consumers/{topic}/{partition}')[0])
            from_offsets[topic_partion] = offset
    return from_offsets

def save_offsets(rdd):
    zk = get_zookeeper_instance()
    for offset in rdd.offsetRanges():
        path = f"/consumers/{offset.topic}/{offset.partition}"
        zk.ensure_path(path)
        zk.set(path, str(offset.untilOffset).encode())

Then before starting streaming you could read offsets from zookeeper and pass them to createDirectStream for fromOffsets argument.:

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


def main(brokers="127.0.0.1:9092", topics=['test1', 'test2']):
    sc = SparkContext(appName="PythonStreamingSaveOffsets")
    ssc = StreamingContext(sc, 2)

    zk = get_zookeeper_instance()
    from_offsets = read_offsets(zk, topics)

    directKafkaStream = KafkaUtils.createDirectStream(
        ssc, topics, {"metadata.broker.list": brokers},
        fromOffsets=from_offsets)

    directKafkaStream.foreachRDD(save_offsets)


if __name__ == "__main__":
    main()