Role of ZooKeeper

The ZooKeeper utility provides configuration and state management and distributed coordination services to Dgraph nodes of the Big Data Discovery cluster. It ensures high availability of the query processing by the Dgraph nodes in the cluster.

ZooKeeper is part of the Hadoop package. The Hadoop package is assumed to be installed on all Hadoop nodes in the BDD cluster deployment. Even though ZooKeeper is installed on all Hadoop nodes in the BDD cluster, it may not be running on all of these nodes. To ensure high availability of a clustered Dgraph deployment, configure an odd number (at least three) of Hadoop nodes to run ZooKeeper instances. This will prevent ZooKeeper from being a single point of failure.

ZooKeeper has the following characteristics:

It is a shared information repository that provides a set of distributed coordination services. It ensures synchronization, event notification, and coordination between the nodes. The communication and coordination mechanisms continue to work in the case when connections or Dgraph-hosting nodes fail.
In the case a leader Dgraph instance fails, Zookeeper informs the Dgraph Gateway of leader Dgraph failure, and the Dgraph Gateway starts the process for automatic Dgraph leader re-election, until a new Dgraph leader node is elected (if the BDD cluster has sufficient number of Dgraph nodes. At least three Dgraph nodes are recommended for high availability).

To summarize, in order to run, ZooKeeper requires a majority of its hosting nodes to be active. The optimal number of Hadoop nodes hosting ZooKeeper instances is an odd number that is at least 3.