How to submit a spark job on a remote master node in yarn client mode?
First of all, if you are setting conf.setMaster(...) from your application code, it takes highest precedence (over the --master argument). If you want to run in yarn client mode, do not use MASTER_IP:7077 in application code. You should supply hadoop client config files to your driver in the following way.
You should set environment variable HADOOP_CONF_DIR or YARN_CONF_DIR to point to the directory which contains the client configurations.
http://spark.apache.org/docs/latest/running-on-yarn.html
Depending upon which hadoop features you are using in your spark application, some of the config files will be used to lookup configuration. If you are using hive (through HiveContext in spark-sql), it will look for hive-site.xml. hdfs-site.xml will be used to lookup coordinates for NameNode reading/writing to HDFS from your job.