Sparkling Water Configuration Properties¶
The following configuration properties can be passed to Spark to configure Sparking Water.
Configuration properties independent of selected backend¶
Property name |
Default value |
Description |
|---|---|---|
Generic parameters |
||
|
|
This option can be set either to
|
|
Generated unique name |
Name of H2O cluster. |
|
|
Limit for number of threads used by
H2O, default |
|
|
Decides whether H2O REPL is initiated or not. |
|
|
Number of parallel REPL sessions started at the start of Sparkling Water |
|
|
Decides whether listener which kills H2O cluster on the change of the underlying cluster’s topology is enabled or not. This configuration has effect only in non-local mode. |
|
|
Enables check if run-time Spark version matches build time Spark version. |
|
|
If unsupported Spark parameter is detected, then application is forced to shutdown. |
|
|
Path to Java KeyStore file. |
|
|
Password for Java KeyStore file. |
|
|
Alias to certificate in keystore to secure H2O Flow. |
|
|
Enable hash login. |
|
|
Enable LDAP login. |
|
|
Enable Kerberos login. |
|
|
Login configuration file. |
|
|
Username used for the backend H2O cluster and to authenticate the client against the backend. |
|
|
Password used to authenticate the client against the backend. |
|
|
Path to a file containing H2O or Sparkling Water internal security configuration. |
|
|
Automatically generate the required key store and password to secure H2O flow by SSL. |
|
|
H2O internal log level used for H2O nodes except the client. |
|
|
Location of H2O logs on H2O nodes except on the client. |
|
|
Interval for getting heartbeat from the H2O backend. |
|
|
Timeout (in msec) for cluster formation. |
|
|
Subnet selector for H2O running inside Spark executors. This disables using IP reported by Spark but tries to find IP based on the specified mask. |
|
|
Interval specifying how often stack traces are taken on each H2O node. -1 means that no stack traces will be taken. |
|
|
Context path to expose H2O web server. |
|
|
Decide whether the Scala cells in H2O Flow will run synchronously or Asynchronously. Default is synchronously. |
|
|
Number of max parallel Scala cell jobs The value -1 means not limited. |
|
|
Offset between the API(=web) port and
the internal communication port on the
client node;
|
|
|
Base port used for individual H2O nodes. |
|
|
If a scoring MOJO instance is not used within a Spark executor JVM for a given timeout in milliseconds, it’s evicted from executor’s cache. Default timeout value is 10 minutes. |
|
|
Timeout in milliseconds specifying how often we check whether the the client is still connected. |
|
|
A string containing extra parameters passed to H2O nodes during startup. This parameter should be configured only if H2O parameters do not have any corresponding parameters in Sparkling Water. |
|
|
Extra HTTP headers that will be used in communication between the front-end and back-end part of Flow UI. The headers should be delimited by a new line. Don’t forget to escape special characters when passing the parameter from a command line. |
|
|
Enables secure communications among
H2O nodes. The security is based on
automatically generated keystore
and truststore. This is equivalent for
|
H2O client parameters |
||
|
|
Directory where flows from H2O Flow are saved. |
|
|
IP of H2O client node. |
|
|
Location of iced directory for the driver instance. |
|
|
H2O internal log level used for H2O client running inside Spark driver. |
|
|
Location of H2O logs on the driver machine. |
|
|
Port on which H2O client publishes its API. If already occupied, the next odd port is tried on so on. |
|
|
Exact client port to access web UI.
The value |
|
|
The client outputs verbose log output
directly into console. Enabling the
flag increases the client log level to
|
|
|
Subnet selector for H2O client, this disables using IP reported by Spark but tries to find IP based on the specified mask. |
|
|
Ignore SPARK_PUBLIC_DNS setting on the H2O client. The option still applies to the Spark application. |
|
|
Enable or disable web on h2o client node. It is enabled by default. Disabling the web just on the client node just restricts everybody from accessing flow, the internal ports between client and rest of the cluster remain open. |
|
|
Allows to override the base URL address of Flow UI, including the scheme, which is showed to the user. |
|
|
A string containing extra parameters passed to H2O client during startup. This parameter should be configured only if H2O parameters do not have any corresponding parameters in Sparkling Water. |
|
|
Whether certificates should be verified before using in H2O or not. |
Internal backend configuration properties¶
Property name |
Default value |
Description |
|---|---|---|
Generic parameters |
||
|
|
Expected number of workers of H2O cluster. Value None means automatic detection of cluster size. This number must be equal to number of Spark executors. |
|
|
Multiplication factor for dummy RDD
generation. Size of dummy RDD is
|
|
|
Number of retries for creation of an RDD spread across all existing Spark executors. |
|
|
Starting size of cluster in case that size is not explicitly configured. |
|
|
Subsequent successful tries to figure out size of Spark cluster, which are producing the same number of nodes. |
|
|
Enable or disable web on H2O worker nodes. It is disabled by default for security reasons. |
|
Either a string with the Path to a file with Hadoop HDFS configuration or the org.apache.hadoop.conf.Configuration object. Useful for HDFS credentials settings and other HDFS-related configurations. |
|
H2O nodes parameters |
||
|
|
Location of iced directory for H2O nodes on the Spark executors. |
External backend configuration properties¶
Property name |
Default value |
Description |
|
|
ip:port of arbitrary H2O node to identify external H2O cluster. |
|
|
Number of H2O nodes to start in
|
|
|
Timeout for confirmation of write operation (Spark frame => H2O frame) on external cluster. |
|
|
Timeout in seconds for starting H2O external cluster. |
|
|
Full path to a file which is used sd the notification file for the startup of external H2O cluster. |
|
|
Amount of memory assigned to each H2O node on YARN/Hadoop. |
|
|
Path to the directory on HDFS used for storing temporary files. |
|
|
If this option is set to |
|
|
Path to H2O driver used during
|
|
|
Yarn queue on which external H2O cluster is started. |
|
|
IP address of H2O driver in case of external cluster in automatic mode. |
|
|
If true, the client will try to kill the cluster and then itself in case some nodes in the cluster report unhealthy status. |
|
|
Kerberos Principal. |
|
|
Kerberos Keytab. |
|
|
Impersonated Hadoop user. |
|
|
Ip address or network of mapper->driver callback interface. Default value means automatic detection. |
|
|
Port of mapper->driver callback interface. Default value means automatic detection. |
|
|
Range portX-portY of mapper->driver callback interface; eg: 50000-55000. |
|
|
This option is a percentage of
|
|
|
The size of blocks representing
data traffic from Spark nodes to
H2O-3 nodes. The value must be
represented in the same format as
JVM memory strings with a size unit
suffix “k”, “m” or “g” (e.g.
|
|
|
Timeout for confirmation from
worker nodes when stopping the
external backend. It is also
possible to pass |
|
|
Name or path to path to a hadoop executable binary which is used to start external H2O backend on YARN. |