Configuring Upgrade Domains

Upgrade Domains allow to group cluster hosts for optimal performance during restarts and upgrades.

Upgrade Domains enable faster cluster restarts, faster Cloudera Runtime upgrades, and seamless OS patching & hardware upgrades across large clusters. Upgrade Domains provide an alternative to the default HDFS block placement policy, distributing data across a set of hosts (potentially larger than a single rack) that Cloudera Manager can upgrade/restart at once without compromising service and data availability. When you select Upgrade Domains as the block placement policy, you also assign an Upgrade Domain group to each DataNode host. The NameNode uses these groups to distribute blocks when writing data, and to orchestrate rolling restarts and upgrades. This feature is useful for very large clusters, or for clusters where rolling restarts happen frequently.

For example, if HDFS is configured with the default replication factor of 3, the NameNode places the replica blocks on DataNode hosts in 3 different Upgrade Domains and on at least two different racks.

note

Cloudera recommends that you assign an approximately equal number of DataNode hosts to each Upgrade Domain.
The number of Upgrade Domains in a cluster should be greater than or equal to the HDFS Replication Factor. When you perform a rolling restart on a cluster, all hosts in an Upgrade Domain group will be restarted simultaneously, followed by the hosts in each remaining Upgrade Domain group.
You should create a sufficient number of Upgrade Domains so that the cluster can still function adequately when all the hosts in a single Upgrade Domain are taken offline. The appropriate number of Upgrade Domains depends on the workloads and capacity of the cluster and may require tuning for optimal performance.
To take advantage of the improved rolling restart performance, Upgrade Domain groups should not duplicate rack assignments. The number of hosts in an Upgrade Domain group should be larger than the number of hosts in a rack.