How it works…

The coordinator node is a special node that works as a proxy/pass thought for the cluster. Its main advantages are as follows:

• It can easily be killed or removed from the cluster without causing any problems. It's not a master, so it doesn't participate in cluster functionalities and it doesn't contain data, so there are no data relocations/replications due to its failure.
• It prevents the instability of the cluster due to a developers' /users bad queries. Sometimes, a user executes aggregations that are too large (that is, date histograms with a range of some years and intervals of 10 seconds). Here, the Elasticsearch node could crash. (In its newest version, Elasticsearch has a structure called circuit breaker to prevent similar issues, but there are always borderline cases that can bring instability using scripting, for example. The coordinator node is not a master and its overload doesn't cause any problems for cluster stability.
• If the coordinator or client node is embedded in the application, there are less round trips for the data, speeding up the application.
• You can add them to balance the search and aggregation throughput without generating changes and data relocation in the cluster.