Why Do We Need Dedicated Master Nodes?

In Elasticsearch, the master node is responsible for cluster management, not for serving search results.

The master node handles things like:

• Creating and deleting indices
• Assigning shards to nodes
• Keeping track of which node has which shard
• Managing cluster state and metadata
• Detecting node failures and reacting to them

What the master node does not do:

• It does not execute searches.
• It does not store index data.
• It does not serve documents.

Because of this, in real systems:

We want the master node to be as stable, light, and undisturbed as possible.

That is why we often configure:

• Some nodes as dedicated data nodes
• Some nodes as dedicated master nodes

Step 1: Understanding the Node Role Configuration

In Elasticsearch, node roles are controlled using:

node.roles

By default, if you do not set anything, a node can do everything.

Dedicated Master Node

node.roles: [ master ]

This means:

• This node is only a master
• It is:
  • Not a data node
  • Not an ingest node
  • Not a coordinating data holder
• It will:
  • Never store shards
  • Never hold index data

Dedicated Data Nodes with Voting Only

node.roles: [ data, master ]
node.voting_only: true

This sounds confusing at first, so let us explain it properly.

• In order to vote in a master election, a node must be master-eligible.
• That is why master is still present in the role list.
• But:
  • node.voting_only: true means: “This node can vote, but it will never be elected as master.”

What Is the Coordinating Role?

Whenever you send a request to Elasticsearch, for example:

• A search request
• A bulk indexing request
• A multi-get request
• A query that touches multiple shards

That request always goes to one node first. That node becomes the coordinator for that request.

The coordinating node is responsible for:

• Accepting the client request.
• Figuring out which shards and which nodes need to be involved.
• Sending the request to all relevant shards (scatter phase).
• Collecting partial results from all shards.
• Merging, sorting, and combining the results (gather phase).
• Sending the final combined response back to the client.

In simple words:

The coordinating node is the “request manager” or “traffic controller” of Elasticsearch queries.

Important points

• Without a master, the cluster cannot manage its state.
• Without a master, it cannot answer cluster-level questions.
• Search and read operations can continue even when the master is down.
• The master node is only for cluster management, not for serving data.
• Without a master:
  • You cannot:
    • Create indices
    • Delete indices
    • Change mappings
    • Change shard allocation