我尝试使用Kerberized Hadoop集群中的Spark Streaming应用程序使用来自Kafka的消息时遇到了一些问题。我尝试了两种方法listed here:
KafkaUtils.createStream
KafkaUtils.createDirectStream
基于接收者的方法(KafkaUtils.createStream
)抛出两种类型的异常(不同的例外,无论我处于本地模式(--master local[*]
)还是YARN模式(--master yarn --deploy-mode client
):
kafka.common.BrokerEndPointNotAvailableException
在两种模式(本地或YARN)中,直接方法(KafkaUtils.createDirectStream
)返回无法解释的EOFException
(详见下文)。
我的最终目标是在YARN 上启动 Spark Streaming作业,因此我将抛开Spark本地工作。
这是我的测试环境:
我正在处理单节点群集(hostname = quickstart.cloudera
)以进行测试。对于有兴趣重现测试的人,我正在使用基于cloudera/quickstart
(Git repo)的自定义Docker容器。
以下是我在spark-shell
中使用的示例代码。当然,当没有启用Kerberos时,此代码可以工作:由kafka-console-producer
生成的消息由Spark应用程序接收。
import org.apache.spark.streaming.kafka.KafkaUtils
import org.apache.spark.streaming.{Seconds, StreamingContext}
import org.apache.spark.storage.StorageLevel
import kafka.serializer.StringDecoder
val ssc = new StreamingContext(sc, Seconds(5))
val topics = Map("test-kafka" -> 1)
def readFromKafkaReceiver(): Unit = {
val kafkaParams = Map(
"zookeeper.connect" -> "quickstart.cloudera:2181",
"group.id" -> "gid1",
"client.id" -> "cid1",
"zookeeper.session.timeout.ms" -> "5000",
"zookeeper.connection.timeout.ms" -> "5000"
)
val stream = KafkaUtils.createStream[String, String, StringDecoder, StringDecoder](ssc, kafkaParams, topics, StorageLevel.MEMORY_ONLY_2)
stream.print
}
def readFromKafkaDirectStream(): Unit = {
val kafkaDirectParams = Map(
"bootstrap.servers" -> "quickstart.cloudera:9092",
"group.id" -> "gid1",
"client.id" -> "cid1"
)
val directStream = KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder](ssc, kafkaDirectParams, topics.map(_._1).toSet)
directStream.print
}
readFromKafkaReceiver() // or readFromKafkaDirectStream()
ssc.start
Thread.sleep(20000)
ssc.stop(stopSparkContext = false, stopGracefully = true)
启用Kerberos后,此代码不起作用。我遵循了本指南:Configuring Kafka Security,并创建了两个配置文件:
jaas.conf
:
KafkaClient {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="/home/simpleuser/simpleuser.keytab"
principal="simpleuser@CLOUDERA";
};
client.properties
:
security.protocol=SASL_PLAINTEXT
sasl.kerberos.service.name=kafka
我可以用以下内容生成消息:
export KAFKA_OPTS="-Djava.security.auth.login.config=/home/simpleuser/jaas.conf"
kafka-console-producer \
--broker-list quickstart.cloudera:9092 \
--topic test-kafka \
--producer.config client.properties
但我不能从Spark Streaming应用程序中使用这些消息。要在spark-shell
模式下启动yarn-client
,我刚创建了一个新的JAAS配置(jaas_with_zk_yarn.conf
),带有Zookeeper部分(Client
),并且对keytab的引用是只有文件的名称(keytab然后通过--keytab
选项传递):
KafkaClient {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="simpleuser.keytab"
principal="simpleuser@CLOUDERA";
};
Client {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="simpleuser.keytab"
principal="simpleuser@CLOUDERA";
};
这个新文件在--files
选项中传递:
spark-shell --master yarn --deploy-mode client \
--num-executors 2 \
--files /home/simpleuser/jaas_with_zk_yarn.conf \
--conf "spark.executor.extraJavaOptions=-Djava.security.auth.login.config=jaas_with_zk_yarn.conf" \
--conf "spark.driver.extraJavaOptions=-Djava.security.auth.login.config=jaas_with_zk_yarn.conf" \
--keytab /home/simpleuser/simpleuser.keytab \
--principal simpleuser
我使用了与之前相同的代码,只是我添加了两个其他Kafka参数,对应于consumer.properties
文件的内容:
"security.protocol" -> "SASL_PLAINTEXT",
"sasl.kerberos.service.name" -> "kafka"
启动Spark Streaming Context后, readFromKafkaReceiver()
会抛出以下错误(无法连接到Zookeeper):
ERROR scheduler.ReceiverTracker: Deregistered receiver for stream 0: Error starting receiver 0 - org.I0Itec.zkclient.exception.ZkTimeoutException: Unable to connect to zookeeper server within timeout: 5000
at org.I0Itec.zkclient.ZkClient.connect(ZkClient.java:1223)
at org.I0Itec.zkclient.ZkClient.<init>(ZkClient.java:155)
at org.I0Itec.zkclient.ZkClient.<init>(ZkClient.java:129)
at kafka.utils.ZkUtils$.createZkClientAndConnection(ZkUtils.scala:89)
at kafka.utils.ZkUtils$.apply(ZkUtils.scala:71)
at kafka.consumer.ZookeeperConsumerConnector.connectZk(ZookeeperConsumerConnector.scala:191)
at kafka.consumer.ZookeeperConsumerConnector.<init>(ZookeeperConsumerConnector.scala:139)
at kafka.consumer.ZookeeperConsumerConnector.<init>(ZookeeperConsumerConnector.scala:156)
at kafka.consumer.Consumer$.create(ConsumerConnector.scala:109)
at org.apache.spark.streaming.kafka.KafkaReceiver.onStart(KafkaInputDStream.scala:100)
at org.apache.spark.streaming.receiver.ReceiverSupervisor.startReceiver(ReceiverSupervisor.scala:148)
at org.apache.spark.streaming.receiver.ReceiverSupervisor.start(ReceiverSupervisor.scala:130)
at org.apache.spark.streaming.scheduler.ReceiverTracker$ReceiverTrackerEndpoint$$anonfun$9.apply(ReceiverTracker.scala:575)
at org.apache.spark.streaming.scheduler.ReceiverTracker$ReceiverTrackerEndpoint$$anonfun$9.apply(ReceiverTracker.scala:565)
at org.apache.spark.SparkContext$$anonfun$38.apply(SparkContext.scala:2003)
at org.apache.spark.SparkContext$$anonfun$38.apply(SparkContext.scala:2003)
at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
at org.apache.spark.scheduler.Task.run(Task.scala:89)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:214)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
at java.lang.Thread.run(Thread.java:745)
有时建立与ZK的连接(没有达到超时),但是没有收到任何消息。
一旦调用此方法, readFromKafkaDirectStream()
就会抛出以下错误::
org.apache.spark.SparkException: java.io.EOFException
at org.apache.spark.streaming.kafka.KafkaCluster$$anonfun$checkErrors$1.apply(KafkaCluster.scala:366)
at org.apache.spark.streaming.kafka.KafkaCluster$$anonfun$checkErrors$1.apply(KafkaCluster.scala:366)
at scala.util.Either.fold(Either.scala:97)
at org.apache.spark.streaming.kafka.KafkaCluster$.checkErrors(KafkaCluster.scala:365)
at org.apache.spark.streaming.kafka.KafkaUtils$.getFromOffsets(KafkaUtils.scala:222)
at org.apache.spark.streaming.kafka.KafkaUtils$.createDirectStream(KafkaUtils.scala:484)
at $iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC.readFromKafkaDirectStream(<console>:47)
没有更多解释,只有EOFException
。我认为Spark和Kafka经纪人之间存在沟通问题,但没有更多解释。我也尝试了metadata.broker.list
而不是bootstrap.servers
,但没有成功。
也许我在JAAS配置文件或Kafka参数中遗漏了什么?也许Spark选项(extraJavaOptions
)无效?我尝试了很多可能性,我有点失落。
如果有人能帮助我解决至少其中一个问题(直接接近或基于接收者),我会很高兴。谢谢:))
答案 0 :(得分:2)
Spark 1.6不支持它,如Cloudera docs中所述:
Spark Streaming无法从安全的Kafka中消耗,直到它开始使用Kafka 0.9 Consumer API
1.6中的Spark-streaming使用旧的消费者API,不支持安全消费。
您可以使用支持安全Kafka的Spark 2.1: https://blog.cloudera.com/blog/2017/05/reading-data-securely-from-apache-kafka-to-apache-spark/