Kylo Spark Properties¶
Overview¶
The kylo-spark-shell process compiles and executes Scala code for schema detection and data transformations.
Configuration¶
The default location of the configuration file is at /opt/kylo/kylo-services/conf/spark.properties.
The process will run in one of three modes depending on which properties are defined. The default mode is Server which requires the process to be started and managed separately, typically using the included init script. In the other two modes, Managed and Multi-User, the kylo-services process will start and manage the kylo-spark-shell processes. Typically Server mode will only be used in development environments and Managed or Multi-User will be used in production. The following sections further describe these modes and their configuration options.
Server Mode¶
The kylo-spark-shell process will run in Server mode when the below properties are defined in spark.properties. In this mode the process should be started by using the included init script.
When using the Kylo sandbox it is sufficient to run service kylo-spark-shell start.
The properties from the other sections in this document are ignored when Server mode is enabled. To modify the Spark options, edit the run-kylo-spark-shell.sh file and add them to the
spark-submit call on the last line. Note that only the local Spark master is supported in this configuration.
| Property | Type | Default | Description |
|---|---|---|---|
| server.port | Number | 8450 | Port for kylo-spark-shell to listen on. |
| spark.shell.server.host | String | Host name or address where the kylo-spark-shell process is running as a server. |
|
| spark.shell.server.port | Number | 8450 | Port where the kylo-spark-shell process is listening. |
| spark.ui.port | Number | 8451 | Port for the Spark UI to listen on. |
Advanced options are available by using Spring Boot properties.
Example spark.properties configuration:
spark.shell.server.host = localhost
spark.shell.server.port = 8450
Managed Mode¶
In Managed mode, Kylo will start one kylo-spark-shell process for schema detection and another for executing data transformations.
Once the process has started it will call back to kylo-services and register itself. This allows Spark to run in yarn-cluster mode as the driver can run on any node in the cluster.
The auth-spark Spring profile must be enabled for the Spark client to start.
| Property | Type | Default | Description |
|---|---|---|---|
| spark.shell.appResource | String | Path to the kylo-spark-shell-client jar file. This is only needed if Kylo is unable to determine the location automatically. The default location for Spark 1.x is /opt/kylo/kylo-services/lib/ app/kylo-spark-shell-client-v1-*.jar. There is a v2 jar for Spark 2.x. |
|
| spark.shell.deployMode | String | Whether to launch a kylo-spark-shell process locally ( client) or on one of the worker machines inside the cluster ( cluster). Set to cluster when enabling user impersonation. |
|
| spark.shell.files | String | Additional files to be submitted with the Spark application. Multiple files should be separated with a comma. |
|
| spark.shell.javaHome | String | The JAVA_HOME for launching the Spark application. |
|
| spark.shell.idleTimeout | Number | 900 | Indicates the amount of time in seconds to wait for a user request before terminating a kylo-spark-shell process. Any user request sent to the process will reset this timeout. This is only used in yarn-cluster mode. |
| spark.shell.jars | String | Additional jars to be submitted with the Spark application. Multiple jars should be separated with a comma. |
|
| spark.shell.master | String | Where to run Spark executors locally (local) or inside a YARN cluster ( yarn). Set to yarn when enabling user impersonation. |
|
| spark.shell.portMin | Number | 45000 | Minimum port number that a kylo-spark-shell process may listen on. |
| spark.shell.portMax | Number | 45999 | Maximum port number that a kylo-spark-shell process may listen on. |
| spark.shell.propertiesFile | String | A custom properties file with Spark configuration for the application. |
|
| spark.shell.registrationUrl | String | Kylo Services URL for registering the Spark application once it has started. This defaults to http://<server-address>:8400/proxy/v1/spark/ shell/register |
|
| spark.shell.sparkArgs | String | Additional arguments to include in the Spark invocation. |
|
| spark.shell.sparkHome | String | A custom Spark installation location for the application. |
|
| spark.shell.verbose | Boolean | false | Enables verbose reporting for Spark Submit. |
The default property values should work on most systems. An error will be logged if Kylo is unable to determine the correct value from the environment. Example spark.properties configuration:
#spark.shell.server.host = localhost
#spark.shell.server.port = 8450
spark.shell.deployMode = cluster
spark.shell.master = yarn
Multi-User Mode¶
Kylo will start a separate process for each Kylo user in Multi-User mode. This ensures that users only have access to their own tables and cannot interfere with each other.
The auth-spark Spring profile must be enabled for the Spark client to start.
In a Kerberized environment Kylo will need to periodically execute kinit to ensure there is an active Kerberos ticket. Spark does not support supplying both a keytab and a proxy user on the command-line. See Spark User Impersonation Configuration for more information on configuring user impersonation in a Kerberized environment.
The options from Managed Mode are also supported.
| Property | Type | Default | Description |
|---|---|---|---|
| spark.shell.proxyUser | Boolean | false | Set to true to enable Multi-User mode. |
Example spark.properties configuration:
#spark.shell.server.host = localhost
#spark.shell.server.port = 8450
spark.shell.deployMode = cluster
spark.shell.master = yarn
spark.shell.proxyUser = true
spark.shell.sparkArgs = --driver-java-options -Djavax.security.auth.useSubjectCredsOnly=false
Hadoop must be configured to allow the kylo user to proxy users:
$ vim /etc/hadoop/conf/core-site.xml
<property>
<name>hadoop.proxyuser.kylo.groups</name>
<value>*</value>
</property>
<property>
<name>hadoop.proxyuser.kylo.hosts</name>
<value>*</value>
</property>
Kerberos¶
Kerberos is supported in both Managed and Multi-User modes.
| Property | Type | Default | Description |
|---|---|---|---|
| kerberos.spark.kerberosEnabled | Boolean | false | Indicates that an active Kerberos ticket is needed to start a kylo-spark-shell process. |
| kerberos.spark.kerberosPrincipal | String | Name of the principal for acquiring a Kerberos ticket. |
|
| kerberos.spark.keytabLocation | String | Local path to the keytab for acquiring a Kerberos ticket. |
|
| kerberos.spark.initInterval | Number | 43200 | Indicates the amount of time in seconds to cache a Kerberos ticket before acquiring a new one. Only used in Multi-User mode. A value of 0 disables calling kinit. |
| kerberos.spark.initTimeout | Number | 10 | Indicates the amount of time in seconds to wait for kinit to acquire a ticket before killing the process. Only used in Multi-User mode. |
| kerberos.spark.retryInterval | Number | 120 | Indicates the amount of time in seconds to wait before retrying to acquire a Kerberos ticket if the last try failed. Only used in Multi-User mode. |
| kerberos.spark.realm | String | Name of the Kerberos realm to append to usernames. |
Example spark.properties configuration:
kerberos.spark.kerberosEnabled = true
kerberos.spark.kerberosPrincipal = kylo
kerberos.spark.keytabLocation = /etc/security/keytabs/kylo.headless.keytab