背景
- spark streaming + kafka 有两种方案接收kafka数据-基于receiver的方案和direct方案(no receiver方案)。
-
基于receiver的方案,属于比较老的方案,其采用Kafka’s high-level API通过专门的Rceiver去接收kafka数据。
采用 KafkaUtils.createStream -
direct方案,是当前的主流用法,其采用Kafka’s simple consumer API,创建的RDD partitions数与kafka partitions数一致。性能比前者好。
采用 KafkaUtils.createDirectStream具体的介绍可以查看官方介绍:Spark Streaming + Kafka Integration Guide
- 对于第二种方案,spark streaming启动后当kafka的topic partition数从A升到B时,新增的分区上的数据会丢失。
spark streaming只从topic中读取原来的A个分区数据,新增的分区并不能被spark streaming感知到。
更具体的原因是
DirectKafkaInputDStream#compute生成的KafkaRDD, 其partitions数与spark streaming启动时topic的partitions数一致,topic的partitions和offset保存在currentOffsets map变量中,
这个变量在启动时初始化,后续不会根据topic的partition变化进行更新。所以导致kafka新增的partitions数据
会丢失。
更详细的分析,可以参考下面的文章:
Spark Streaming 自适应上游 kafka topic partition 数目变化
因为spark streaming是24x7运行的,如何让spark streaming不重启的情况下,自适应topic partitions的变化?
且能够符合如下要求:
- 不需要重新编译spark源码, 因为源代码不是我们维护的。
- 能够灵活的部署
解决方案
思路可以参考
Spark Streaming 自适应上游 kafka topic partition 数目变化
总的解决方案如下:
- MTDirectKafkaInputDStream继承DirectKafkaInputDStream,override compute方法,在每次生成KafkaRDD时,更新currentOffsets中的分区信息。
- 在org.apache.spark.streaming.kafka路径下,新建一个KafkaUtils.scala文件,里面的代码直接将spark源码中的KafkaUtils源码复制过来。 修改新建的KafkaUtils.scala,将createDirectStream中new DirectKafkaInputDStream,替换为 new MTDirectKafkaInputDStream.
具体的实现步骤如下
-
新建maven工程 kafkastreamingadpter, 配置好scala相关的pom配置。
-
新建package org.apache.spark.streaming.kafka
备注:这里需要说明,因为下面创建的MTDirectKafkaInputDStream需要继承DirectKafkaInputDStream,而DirectKafkaInputDStream是private[streaming]修饰的。这里的规避技巧就是,将需要继承DirectKafkaInputDStream的子类对应的package设置为和DirectKafkaInputDStream一致,即可规避private无法继承的问题。
-
新建MTDirectKafkaInputDStream继承DirectKafkaInputDStream
override compute方法,在每次生成KafkaRDD时,更新currentOffsets中的分区信息。
-
在org.apache.spark.streaming.kafka路径下,新建一个KafkaUtils.scala文件,里面的代码直接复制spark源码中的KafkaUtils源码。修改新建的KafkaUtils.scala,将createDirectStream中new DirectKafkaInputDStream,替换为 new MTDirectKafkaInputDStream
备注:streaming代码中是通过 KafkaUtils.createDirectStream来创建stream的。因此希望KafkaUtils.createDirectStream返回的是MTDirectKafkaInputDStream。所以要修改KafkaUtils代码,将createDirectStream中new DirectKafkaInputDStream,修改为 new MTDirectKafkaInputDStream。
我们不希望修改spark源码,所以这里用来个小技巧,直接将KafkaUtils代码copy一份出来进行修改,需要保持KafkaUtils的package路径一致2,4中用到的小技巧参考了
[2016中国云计算技术大会-腾讯林立伟-Spark-Streaming在腾讯广点通的应用.pdf]
-
新建spark streaming + kafka应用实例DirectKafkaWordCount1
实例需要配置 spark、kafka streaming相关的依赖。
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-streaming-kafka-0-8_2.11</artifactId>
<version>2.2.0</version>
</dependency>
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-core_2.11</artifactId>
<version>2.2.0</version>
</dependency>
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-streaming_2.11</artifactId>
<version>2.2.0</version>
</dependency>
- mvn clean package 打包 生成kafka-streaming-adpter-1.0.jar
对应的配置文件如下
<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
<modelVersion>4.0.0</modelVersion>
<groupId>com.sensetime.iva</groupId>
<artifactId>kafka-streaming-adpter</artifactId>
<version>1.0</version>
<packaging>jar</packaging>
<properties>
<scala.version>2.11.8</scala.version>
<commons.codec.version>1.8</commons.codec.version>
<grizzled.version>1.0.1</grizzled.version>
<slf4j-log4j12.version>1.7.5</slf4j-log4j12.version>
</properties>
<dependencies>
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-streaming-kafka-0-8_2.11</artifactId>
<version>2.2.0</version>
</dependency>
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-core_2.11</artifactId>
<version>2.2.0</version>
</dependency>
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-streaming_2.11</artifactId>
<version>2.2.0</version>
</dependency>
<dependency>
<groupId>org.scala-lang</groupId>
<artifactId>scala-library</artifactId>
<version>${scala.version}</version>
</dependency>
<dependency>
<groupId>junit</groupId>
<artifactId>junit</artifactId>
<version>3.8.1</version>
<scope>test</scope>
</dependency>
<dependency>
<groupId>commons-codec</groupId>
<artifactId>commons-codec</artifactId>
<version>${commons.codec.version}</version>
</dependency>
<dependency>
<groupId>org.clapper</groupId>
<artifactId>grizzled-slf4j_2.10</artifactId>
<version>${grizzled.version}</version>
</dependency>
<dependency>
<groupId>org.slf4j</groupId>
<artifactId>slf4j-log4j12</artifactId>
<version>${slf4j-log4j12.version}</version>
</dependency>
</dependencies>
<build>
<sourceDirectory>src/main/scala</sourceDirectory>
<testSourceDirectory>src/test/scala</testSourceDirectory>
<plugins>
<plugin>
<groupId>org.scala-tools</groupId>
<artifactId>maven-scala-plugin</artifactId>
<executions>
<execution>
<goals>
<goal>compile</goal>
<goal>testCompile</goal>
</goals>
</execution>
</executions>
<configuration>
<scalaVersion>${scala.version}</scalaVersion>
</configuration>
</plugin>
</plugins>
</build>
</project>
- 上传jar包到spark on k8s的根路径下的jars,进行镜像打包
node 73节点
cd /home/huangchuibi/k8s/spark-2.2.0-k8s-0.5.0-bin-2.7.3
./sbin/build-push-docker-images.sh -r register.sensetime.sz.test-portus.com/spark-on-k8s -t 0.1 build
./sbin/build-push-docker-images.sh -r register.sensetime.sz.test-portus.com/spark-on-k8s -t 0.1 push
- 在k8s各个节点上pull 镜像,因为spark on k8s,配置镜像拉取策略为本地有就不再拉取。所以需要手动进行拉取。
docker pull register.sensetime.sz.test-portus.com/spark-on-k8s/spark-driver:0.1
docker pull register.sensetime.sz.test-portus.com/spark-on-k8s/spark-executor:0.1
- 创建topic
ssh node91
kubectl -n snappydata exec -it kafka-0 /bin/bash
cd /opt/kafka_2.11-0.10.2.0
bin/kafka-topics.sh --create --zookeeper zookeeper-0.zookeeper.snappydata.svc.cluster.local:2181,zookeeper-1.zookeeper.snappydata.svc.cluster.local:2181,zookeeper-2.zookeeper.snappydata.svc.cluster.local:2181/kafka --replication-factor 1 --partitions 1 --topic test8
- 运行streaming 应用
bin/spark-submit \
--deploy-mode cluster \
--class org.apache.spark.examples.streaming.DirectKafkaWordCount1 \
--master k8s://http://172.20.2.91:8001 \
--kubernetes-namespace snappydata \
--conf spark.executor.instances=3 \
--conf spark.app.name=spark-streaming \
--conf spark.kubernetes.driver.docker.image=register.sensetime.sz.test-portus.com/spark-on-k8s/spark-driver:0.1 \
--conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \
--conf spark.kubernetes.executor.docker.image=register.sensetime.sz.test-portus.com/spark-on-k8s/spark-executor:0.1 \
--conf spark.kubernetes.initcontainer.docker.image=register.sensetime.sz.test-portus.com/spark-on-k8s/spark-init:0.1 \
--conf "spark.driver.extraJavaOptions=-Xss10m" \
--conf "spark.executor.extraJavaOptions=-Xss10m" \
--conf "spark.driver.extraClassPath=/opt/spark/examples/jars/kafka-streaming-adpter-1.0.jar" \
--conf "spark.executor.extraClassPath=/opt/spark/examples/jars/kafka-streaming-adpter-1.0.jar" \
--conf "spark.executor.userClassPathFirst=true" \
--conf "spark.driver.userClassPathFirst=true" \
local:///opt/spark/examples/jars/kafka-streaming-adpter-1.0.jar kafka-0.kafka.snappydata.svc.cluster.local:9092,kafka-1.kafka.snappydata.svc.cluster.local:9092,kafka-2.kafka.snappydata.svc.cluster.local:9092 test8
备注 这里增加的四个conf配置,是为了让kafka-streaming-adpter-1.0.jar里的KafkaUtils优先级高于spark源码中的KafkaUtils。但是实际测试发现,应用还是以同一个jar包中的KafkaUtils为优先级最高。所以下面的四个配置也可以去掉。
--conf "spark.driver.extraClassPath=/opt/spark/examples/jars/kafka-streaming-adpter-1.0.jar" \
--conf "spark.executor.extraClassPath=/opt/spark/examples/jars/kafka-streaming-adpter-1.0.jar" \
--conf "spark.executor.userClassPathFirst=true" \
--conf "spark.driver.userClassPathFirst=true" \
- 动态增加分区
kafka-topics.sh --zookeeper zookeeper-0.zookeeper.snappydata.svc.cluster.local:2181,zookeeper-1.zookeeper.snappydata.svc.cluster.local:2181,zookeeper-2.zookeeper.snappydata.svc.cluster.local:2181/kafka -alter -partitions 2 --topic test8
网友评论