Spark Shell - __spark_libs__.zip does not exist

Question

I\'m new to Spark and I\'m busy setting up a Spark Cluster with HA enabled.

When starting a spark shell for testing via: bash spark-shell --master yarn --deploy

Answer 1

This error was due to the config in the core-site.xml file.

Please note that to find this file your HADOOP_CONF_DIR env variable must be set.

In my case I added HADOOP_CONF_DIR=/opt/hadoop-2.7.3/etc/hadoop/ to ./conf/spark-env.sh

See: Spark Job running on Yarn Cluster java.io.FileNotFoundException: File does not exits , eventhough the file exits on the master node

core-site.xml

    
        fs.default.name
        hdfs://master:9000
     

If this endpoint is unreachable, or if Spark detects that the file system is the same as the current system, the lib files will not be distributed to the other nodes in your cluster causing the errors above.

In my situation the node I was on couldn't reach port 9000 on the specified host.

Debugging

Turn the log level up to info. You can do this by:

1. Copy ./conf/log4j.properties.template to ./conf/log4j.properties

2. In the file set log4j.logger.org.apache.spark.repl.Main = INFO

Start your Spark Shell as normal. If your issue is the same as mine, you should see an info message such as: INFO Client: Source and destination file systems are the same. Not copying file:/tmp/spark-c1a6cdcd-d348-4253-8755-5086a8931e75/__spark_libs__1391186608525933727.zip

This should lead you to the problem as it starts the train reaction that results from the missing files.