spark 集群安装

下载安装文件
配置Spark-env.sh 和spark-default.properties

sbin/start-master.sh

slaver节点启动 worker

./start-slave.sh spark://cnsz046690:7077

---

scp -r spark-2.1.1-bin-hadoop2.6 cnsz046691:~
scp -r spark-2.1.1-bin-hadoop2.6 cnsz046745:~
scp -r spark-2.1.1-bin-hadoop2.6 cnsz046746:~

---

spark 2.1.1 参数修改

import scala.collection.JavaConversions._

show partitions base.UDS_B_I_TRADE_FUND_MOVT;

show partitions base.UDS_B_I_TRADE_FUND_MOVT;
select count(1) from base.UDS_B_I_TRADE_FUND_MOVT;

spark-sql –files /etc/spark/log4j.properties

spark-sql -Dlog4j.configuration=/etc/spark/log4j.properties

Spark 环境部署和动态资源分配配置

• spark2.2及以后，Java要求最低要Java8
• spark-sql 不支持custer模式

mv spark-2.2.0-bin-hadoop2.6 /usr/libln -s spark-2.2.0-bin-hadoop2.6 sparkmv /opt/app/spark/conf/* .ln -s /etc/spark/conf confln -s /var/log/spark logs

启动history server

/usr/lib/spark/sbin/start-history-server.sh

如果使用yarn模式，好像不用修改
修改文件log4j.properties，将日志级别调整为WARN
log4j.rootCategory=INFO, console

添加全局路径
export PATH=$PATH:/usr/lib/spark/bin

Spark log配置

log4j.rootCategory=INFO, RFAlog4j.appender.RFA=org.apache.log4j.RollingFileAppenderlog4j.appender.RFA.File=/appcom/log/spark/spark-${user.name}.loglog4j.appender.RFA.MaxFileSize=256MBlog4j.appender.RFA.MaxBackupIndex=20log4j.appender.RFA.layout=org.apache.log4j.PatternLayout# Pattern format: Date LogLevel LoggerName LogMessagelog4j.appender.RFA.layout.ConversionPattern=%d{ISO8601} %p %c: %m%n# Debugging Pattern format#log4j.appender.RFA.layout.ConversionPattern=%d{ISO8601} %-5p %c{2} (%F:%M(%L)) - %m%n

问题排查

Spark2.0 yarn方式启动报错

优雅的解决方法

Jersey problem

If you try to run a spark-submit command on YARN you can expect the following error message:

Exception in thread “main” java.lang.NoClassDefFoundError: com/sun/jersey/api/client/config/ClientConfig
Jar file jersey-bundle-*.jar is not present in the $SPARK_HOME/jars. Adding it fixes this problem:

sudo -u spark wget http://repo1.maven.org/maven2/com/sun/jersey/jersey-bundle/1.19.1/jersey-bundle-1.19.1.jar -P $SPARK_HOME/jars
January 2017 – Update on this issue:
If the following is done, Jersey 1 will be used when starting Spark History Server and the applications in Spark History Server will not be shown. The folowing error message will be generated in the Spark History Server output file:

WARN servlet.ServletHandler: /api/v1/applicationsjava.lang.NullPointerException        at org.glassfish.jersey.servlet.ServletContainer.service(ServletContainer.java:388)

This problem occurs only when one tries to run Spark on YARN, since YARN 2.7.3 uses Jersey 1 and Spark 2.0 uses Jersey 2

One workaround is not to add the Jersey 1 jar described above but disable the YARN Timeline Service in spark-defaults.conf

spark.hadoop.yarn.timeline-service.enabled false

[解决方法二](https://my.oschina.net/xiaozhublog/blog/737902)jar包冲突导致：Exception in thread "main" java.lang.NoClassDefFoundError: com/sun/jersey/api/client/config/ClientConfig解决办法：

cp /usr/lib/hadoop-yarn/lib/jersey-client-1.9.jar /usr/lib/spark/jarscp /usr/lib/hadoop-yarn/lib/jersey-core-1.9.jar /usr/lib/spark/jarsmv /usr/lib/spark/jars/jersey-client-2.22.2.jar /usr/lib/spark/jars/jersey-client-2.22.2.jar.bak

Spark 获取hive元数据失败

Caused by: MetaException(message:Version information not found in metastore. )    at org.apache.hadoop.hive.metastore.ObjectStore.checkSchema(ObjectStore.java:6664)    at org.apache.hadoop.hive.metastore.ObjectStore.verifySchema(ObjectStore.java:6645)    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)    at java.lang.reflect.Method.invoke(Method.java:498)

解决方法：
关闭 hive.metastore.schema.verification 参数即可，这个参数会根据hive的版本去检查元数据。

环境测试：

spark-submit --class org.apache.spark.examples.SparkPi \    --master yarn \    --deploy-mode cluster \    --driver-memory 2g \    --executor-memory 2g \    --executor-cores 1 \    --queue default \    /usr/lib/spark/examples/jars/spark-examples_*.jar \    10spark-submit --class org.apache.spark.examples.SparkPi \    --master yarn \    --deploy-mode client \    --driver-memory 2g \    --executor-memory 2g \    --executor-cores 1 \    --queue default \    /usr/lib/spark/examples/jars/spark-examples_*.jar \    10spark-sql --master yarn --deploy-mode client \  --driver-memory 2g \  --executor-memory 2g \  --num-executors 8 

Spark 动态资源配置

• yarn-site.xml 修改

  <property>    <name>yarn.nodemanager.aux-services</name>    <value>mapreduce_shuffle,spark_shuffle</value>  </property>  <property>    <name>yarn.nodemanager.aux-services.spark_shuffle.class</name>    <value>org.apache.spark.network.yarn.YarnShuffleService</value>  </property>  <property>    <name>spark.shuffle.service.port</name>    <value>7337</value>  </property>

• 发布spark shuffle jar

chmod a+x /usr/lib/spark/lib/*.jarcp /usr/lib/spark/lib/spark-1.6.3-yarn-shuffle.jar /usr/lib/hadoop-yarn/  并同步到所有NM节点

• 配置spark-defaults.conf 开启动态资源分配

spark.shuffle.service.enabled true   spark.shuffle.service.port 7337 spark.dynamicAllocation.enabled true  spark.dynamicAllocation.minExecutors 1  spark.dynamicAllocation.maxExecutors 100  spark.dynamicAllocation.schedulerBacklogTimeout 1sspark.dynamicAllocation.sustainedSchedulerBacklogTimeout 5s