How to enable spark-history server for standalone cluster non hdfs mode

问题:

I have setup Spark2.1.1 cluster (1 master 2 slaves) following http://paxcel.net/blog/how-to-setup-apache-spark-standalone-cluster-on-multiple-machine/ in standalone mode. I do not have a pre-Hadoop setup on of the machine. I wanted to start spark-history server. I run it as follows:

roshan@bolt:~/spark/spark_home/sbin$ ./start-history-server.sh 

and in the spark-defaults.conf I set this:

spark.eventLog.enabled           true 

But it fails with the error:

7/06/29 22:59:03 INFO SecurityManager: SecurityManager: authentication disabled; ui acls disabled; users  with view permissions: Set(roshan); groups with view permissions: Set(); users  with modify permissions: Set(roshan); groups with modify permissions: Set() 17/06/29 22:59:03 INFO FsHistoryProvider: History server ui acls disabled; users with admin permissions: ; groups with admin permissions Exception in thread "main" java.lang.reflect.InvocationTargetException     at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)     at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62)     at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)     at java.lang.reflect.Constructor.newInstance(Constructor.java:423)     at org.apache.spark.deploy.history.HistoryServer$.main(HistoryServer.scala:278)     at org.apache.spark.deploy.history.HistoryServer.main(HistoryServer.scala) Caused by: java.io.FileNotFoundException: Log directory specified does not exist: file:/tmp/spark-events Did you configure the correct one through spark.history.fs.logDirectory?     at org.apache.spark.deploy.history.FsHistoryProvider.org$apache$spark$deploy$history$FsHistoryProvider$$startPolling(FsHistoryProvider.scala:214) 

What I should set to spark.history.fs.logDirectory and spark.eventLog.dir

Update 1:

spark.eventLog.enabled           true spark.history.fs.logDirectory   file:////home/roshan/spark/spark_home/logs spark.eventLog.dir               file:////home/roshan/spark/spark_home/logs 

but I am always getting this error:

java.lang.IllegalArgumentException: Codec [1] is not available. Consider setting spark.io.compression.codec=snappy at org.apache.spark.io.Co 

回答1:

By default spark defines file:/tmp/spark-events as the log directory for history server and your log clearly says spark.history.fs.logDirectory is not configured

first of all you need to create spark-events folder in /tmp (which is not a good idea as /tmp is refreshed everytime a machine is rebooted) and then add spark.history.fs.logDirectory in spark-defaults.conf to point to that directory. But I suggest you create another folder which spark user has access to and update spark-defaults.conf file.

You need to define two more variables in spark-defaults.conf file

spark.eventLog.dir              file:path to where you want to store your logs spark.history.fs.logDirectory   file:same path as above 

Suppose you want to store in /opt/spark-events where spark user has access to then above parameters in spark-defaults.conf would be

spark.eventLog.enabled          true spark.eventLog.dir              file:/opt/spark-events spark.history.fs.logDirectory   file:/opt/spark-events 

You can find more information in Monitoring and Instrumentation

回答2:

Try setting

spark.io.compression.codec=org.apache.spark.io.SnappyCompressionCodec  

in spark-defaults.conf

文章来源: How to enable spark-history server for standalone cluster non hdfs mode