Spark on Yarn开发运维过程中遇到的问题汇总

Spark Yarn

1. 启动nodemanager报错 No space left on device

   使用df -h命令判断nodemanager运行日志和启动日志磁盘空间是否足够。

2. 使用pyspark读取kafka对应topic数据报错java.lang.NoClassDefFoundError: org/apache/kafka/common/message/KafkaLZ4BlockOutputStream

   • 更改之前

     ./bin/spark-submit --jars lib/spark-streaming-kafka_2.10-1.6.1.jar,lib/kafka_2.10-0.8.2.1.jar,lib/metrics-core-2.2.0.jar --deploy-mode client ./project/stream.py

   • 更改之后

     ./bin/spark-submit --jars lib/spark-streaming-kafka_2.10-1.6.1.jar,lib/kafka_2.10-0.8.2.1.jar,lib/metrics-core-2.2.0.jar,lib/kafka-clients-0.8.2.1.jar --deploy-mode client ./project/stream.py

3. 使用hdfs balancer显示Another Balancer is running.. Exiting ...

   HDFS HA模式与reblancer不兼容 参考资料

   bin/hdfs balancer -Dfs.defaultFS=hdfs://namenode:8020 -Ddfs.nameservices="" -threshold 10

4. 使用MaxMindGeoIP解析报错

   java.lang.NoSuchMethodError: com.fasterxml.jackson.databind.node.ArrayNode.(Lcom/fasterxml/jackson/databind/node/JsonNodeFactory;Ljava/util/List;)

   com.maxmind.geoip2 版本需要是 2.5.0，以便和spark本身兼容

   <dependency>
       <groupId>com.maxmind.geoip2</groupId>
       <artifactId>geoip2</artifactId>
       <version>2.5.0</version>
   </dependency>