During the Liberty development cycle sahara implemented a tool that recommends and applies configuration values for cluster templates and node group templates. These recommendations are based on the number of specific instances and on flavors of the cluster node groups. Currently the following plugins support this feature:
- CDH;
- Ambari;
- Spark;
- the Vanilla Apache Hadoop plugin.
By default this feature is enabled for all cluster templates and node group templates. If you want to disable this feature for a particular cluster or node group template you should set the use_autoconfig field to false.
The following describes the settings for which sahara can recommend autoconfiguration:
The Cloudera, Spark and Vanilla Apache Hadoop plugin support configuring dfs.replication (dfs_replication for Cloudera plugin) which is calculated as a minimum from the amount of datanode (HDFS_DATANODE for Cloudera plugin) instances in the cluster and the default value for dfs.replication.
The Vanilla Apache Hadoop plugin and Cloudera plugin support autoconfiguration of basic YARN and MapReduce configs. These autoconfigurations are based on the following documentation: http://docs.hortonworks.com/HDPDocuments/HDP2/HDP-2.0.9.1/bk_installing_manually_book/content/rpm-chap1-11.html
The Ambari plugin has its own strategies on configuration recommendations. You can choose one of ALWAYS_APPLY, NEVER_APPLY, and ONLY_STACK_DEFAULTS_APPLY. By default the Ambari plugin follows the NEVER_APPLY strategy. You can get more information about strategies in Ambari’s official documentation: https://cwiki.apache.org/confluence/display/AMBARI/Blueprints#Blueprints-ClusterCreationTemplateStructure