In a Hadoop cluster, high availability provides failover support for one or more configured nodes. This section describes how to enable the Designer Cloud Powered by Trifacta® platform to utilize the highly available set of nodes within the Hadoop cluster. High availability enables access to each node of the cluster configured for it, in the event of machine crash or software installation or upgrade.
If high availability is enabled on the Hadoop cluster, you must enable it on the Designer Cloud Powered by Trifacta platform, which prevents integration conflicts between Hadoop-specific components in the platform and their cluster equivalents.
The WebHDFS service does not directly support high availability. You must enable the related HttpFS service and specify a WebHDFS namenode to point to the server hosting the HttpFS service.
NOTE: Avoid enabling the HttpFS service on the primary namenode of the cluster. In the event that the node hosting the namenode fails over, the HttpFS service is no longer available. You may be required to manually set the active namenode and restart the Designer Cloud Powered by Trifacta platform.
For more information, see Enable HttpFS.
Enable HA Service
To begin, you must enable the High Availability service in the Designer Cloud Powered by Trifacta platform for the supported components. In platform configuration, each component has its own feature flag under
In the following example configuration, high availability has been disabled for resourcemanagers and enabled for namenodes:
NOTE: In almost all cases,
feature.highAvailability.resourceManager should be set to
false. For more information, see Example - Configure resource manager below.
Configure HA for Individual Components
High availability in Hadoop works by specifying a nameservice for a highly available component and then enumerating the hosts and ports as children of that nameservice node. These values must be explicitly specified in the platform configuration.
Service names and child names should be specified in the file as they appear in the cluster's configuration files.
Example - Configure namenode
In the following example, the nameservice
namenodeha provides high availability through two namenodes:
nn2. In a high availability environment, these hosts are used for submitting jobs and writing data to HDFS.
Example - Configure resource manager
feature.highAvailability.resourcemanager=true only if the cluster file
yarn.resourcemanager.hostname.highlyavailableyarn. This setting enables the cluster high availability for resourcemanager.
feature.highAvailability.resourcemanager=false for all environments. For HA environments, the resourcemanager hosts specified in the configuration below set the HA servers that are used by the Designer Cloud Powered by Trifacta platform.
The following example specifies two failover nodes for the resource manager:
Configure HA in a Kerberized Environment
If you are enabling high availability in a Kerberized environment, additional configuration is required.
NOTE: WebHDFS does not support high availability/failover. You must enable HttpFS instead. For more information, see Enable HttpFS.
If you have not done so already, acquire
httpfs-site.xmlfrom your Hadoop cluster.
Add the following settings to the file, replacing
]with the value appropriate for your environment:
The above change must also be applied to the
httpfs-site.xmlconfiguration file for the cluster.
Save your changes and restart the platform.
Change Active Namenode
If you must change the active namenode manually, the new namenode must be configured explicitly.
NOTE: This step is not required for a failover event.
NOTE: If the HttpFS service has been tied to the primary namenode of the cluster and that node fails, this setting must be manually configured to the new node and the platform must be restarted. Avoid tying HttpFS to the primary namenode.
In this example, the active namenode has been set to the
nn1 value in the previous configuration:
When high availability has been enabled, you must restart the platform from the command line. For more information, see Start and Stop the Platform.
This page has no comments.