Frequently Asked Questions¶

Where do I find the Spark logs?

Standalone mode: Spark executor logs are located in the $SPARK_HOME/work/app-<AppName> directory (where <AppName> is the name of your application). The location also contains stdout/stderr from H2O.

YARN mode: The executors logs are available via the yarn logs -applicationId <appId> command. Driver logs are by default printed to console; however, H2O also writes logs into current_dir/h2ologs.

The location of H2O driver logs can be controlled via the Spark property spark.ext.h2o.client.log.dir (passed via --conf).
How can I display Sparkling Water information in the Spark History Server?

Sparkling Water reports the information already. You just need to add the sparkling-water classes on the classpath of the Spark history server. To see how to configure the Spark application for logging into the History Server, please see the Spark documentation for Monitoring and Instrumentation
Spark is very slow during initialization or H2O does not form a cluster. What should I do?
Configure the Spark variable SPARK_LOCAL_IP. For example:
export SPARK_LOCAL_IP='127.0.0.1'
How do I increase the amount of memory assigned to the Spark executors in Sparkling Shell?
Sparkling Shell accepts common Spark Shell arguments. For example, to increase the amount of memory allocated by each executor, use the spark.executor.memory parameter:
bin/sparkling-shell --conf "spark.executor.memory=4g"
How do I change the base port that H2O uses to find available ports?

H2O accepts the spark.ext.h2o.port.base parameter via Spark configuration properties: bin/sparkling-shell --conf "spark.ext.h2o.port.base=13431". For a complete list of configuration options, refer to Sparkling Water Configuration Properties.
How do I use Sparkling Shell to launch a Scala test.script that I created?
Sparkling Shell accepts common Spark Shell arguments. To pass your script, use -i option of the Spark Shell:
bin/sparkling-shell -i test.script
How do I increase PermGen size for Spark driver?

Specify --conf spark.driver.extraJavaOptions="-XX:MaxPermSize=384m"

How do I add Apache Spark classes to the Python path?

Configure the Python path variable PYTHONPATH:

export PYTHONPATH=$SPARK_HOME/python:$SPARK_HOME/python/build:$PYTHONPATH
export PYTHONPATH=$SPARK_HOME/python/lib/py4j-0.9-src.zip:$PYTHONPATH

While importing a class from the hex package in Sparkling Shell, I’m getting the following message: missing arguments for method hex in object functions; follow this method with '_' if you want to treat it as a partially applied
In this case you are probably using Spark 1.5+, which is importing SQL functions into the Spark Shell environment. Please use the following syntax to import a class from the hex package:
import _root_.hex.tree.gbm.GBM
While attempting to run Sparkling Water on HDP YARN cluster, I’m getting an error: java.lang.NoClassDefFoundError: com/sun/jersey/api/client/config/ClientConfig

The YARN time service is not compatible with libraries provided by Spark. Please disable time service by setting spark.hadoop.yarn.timeline-service.enabled=false. For more details, please visit https://issues.apache.org/jira/browse/SPARK-15343.
Getting non-deterministic H2O Frames after the Spark Data Frame to H2O Frame conversion.
This is caused by what we think is a bug in Apache Spark. On specific kinds of data combined with a higher number of partitions, we can see non-determinism in BroadCastHashJoins. This leads to jumbled rows and columns in the output H2O frame. We recommend disabling broadcast based joins, which seem to be non-deterministic as:
sqlContext.sql("SET spark.sql.autoBroadcastJoinThreshold=-1")
The issue can be tracked as PUBDEV-3808. On the Spark side, the following issue is related to the problem: Spark-17806.

How can I configure the Hive metastore location?

Spark SQL context (in fact Hive) requires the use of metastore, which stores metadata about Hive tables. In order to ensure this works correctly, the ${SPARK_HOME}/conf/hive-site.xml needs to contain the following configuration. We provide two examples, how to use MySQL and Derby as the metastore.

For MySQL, the following configuration needs to be located in the ${SPARK_HOME}/conf/hive-site.xml configuration file:
<property>
  <name>javax.jdo.option.ConnectionURL</name>
  <value>jdbc:mysql://{mysql_host}:${mysql_port}/{metastore_db}?createDatabaseIfNotExist=true</value>
  <description>JDBC connect string for a JDBC metastore</description>
</property>

<property>
  <name>javax.jdo.option.ConnectionDriverName</name>
  <value>com.mysql.jdbc.Driver</value>
  <description>Driver class name for a JDBC metastore</description>
</property>

<property>
  <name>javax.jdo.option.ConnectionUserName</name>
  <value>{username}</value>
  <description>username to use against metastore database</description>
</property>

<property>
  <name>javax.jdo.option.ConnectionPassword</name>
  <value>{password}</value>
  <description>password to use against metastore database</description>
</property>
where:

{mysql_host} and {mysql_port} are the host and port of the MySQL database.

{metastore_db} is the name of the MySQL database holding all the metastore tables.

{username} and {password} are the username and password to MySQL database with read and write access to the {metastore_db} database.

For Derby, the following configuration needs to be in the the ${SPARK_HOME}/conf/hive-site.xml configuration file:
<property>
  <name>javax.jdo.option.ConnectionURL</name>
  <value>jdbc:derby://{file_location}/metastore_db;create=true</value>
  <description>JDBC connect string for a JDBC metastore</description>
</property>

<property>
  <name>javax.jdo.option.ConnectionDriverName</name>
  <value>org.apache.derby.jdbc.ClientDriver</value>
  <description>Driver class name for a JDBC metastore</description>
</property>
where:

{file_location} is the location to the metastore_db database file.

After conversion of Spark Data Frame to H2O Frame, I see only 100 columns on the console.
If your Spark Data Frame has more than 100 columns, we don’t treat it any different. We always fully convert the Spark Data Frame to H2O Frame. We just limit the number of columns we send to the client as it’s hard to read that many columns in the console plus it optimizes the amount of data we transfer betweeen the client and backend. If you wish to configure how many columns are sent to the client, you can specify it as part of the conversion method as:
h2o_context.as_h2o_frame(dataframe, "Frame_Name", 200):
The last parameter specifies the number of columns to sent for the preview.
I’m getting the following exception when trying to start PySparkling installed via pip: No matching distribution found for pyspark<=2.4.0,>=UBST_SPARK_MAJOR_VERSION.0 (from h2o-pysparkling-UBST_SPARK_MAJOR_VERSION==2.4.12-SNAPSHOT-98)

h2o_pysparkling_2.4 package now correctly requires PySpark package. If you want to use PySparkling from installation from pip, please also make sure to install PySpark package.
I’m getting java.lang.reflect.InvocationTargetException via java.lang.IllegalArgumentException saying that “YOUR_SPARK_ML_STAGE parameter locale given invalid value YOUR_LOCALE.” when using a Spark stage in my ML pipeline.
Set the default locale for JVM of Spark driver to a valid combination of a language and country:
--conf spark.driver.extraJavaOptions="-Duser.language=en -Duser.country=US"