Container [pid=6263,containerID=container_1494900155967_0001_02_000001] is running beyond virtual memory limits

以Spark-Client模式運行，Spark-Submit時出現了下麵的錯誤：

User:  hadoop  
Name:  Spark Pi  
Application Type:  SPARK  
Application Tags:   
YarnApplicationState:  FAILED  
FinalStatus Reported by AM:  FAILED  
Started:  16-五月-2017 10:03:02  
Elapsed:  14sec  
Tracking URL:  History  
Diagnostics:  Application application_1494900155967_0001 failed 2 times due to AM Container for appattempt_1494900155967_0001_000002 exited with exitCode: -103 
For more detailed output, check application tracking page:http://master:8088/proxy/application_1494900155967_0001/Then, click on links to logs of each attempt. 
Diagnostics: Container [pid=6263,containerID=container_1494900155967_0001_02_000001] is running beyond virtual memory limits. Current usage: 107.3 MB of 1 GB physical memory used; 2.2 GB of 2.1 GB virtual memory used. Killing container.  

意思是說Container要用2.2GB的記憶體，而虛擬記憶體只有2.1GB，不夠用了，所以Kill了Container。

我的SPARK-EXECUTOR-MEMORY設置的是1G，即物理記憶體是1G，Yarn預設的虛擬記憶體和物理記憶體比例是2.1，也就是說虛擬記憶體是2.1G，小於了需要的記憶體2.2G。解決的辦法是把擬記憶體和物理記憶體比例增大，在yarn-site.xml中增加一個設置：

    <property>
        <name>yarn.nodemanager.vmem-pmem-ratio</name>
        <value>2.5</value>
    </property>

再重啟Yarn，這樣一來就能有2.5G的虛擬記憶體，運行時就不會出錯了。