Single Server Single Node vs Single Server Multiple Node

[sramana235]

As the title suggests, I wonder if it is better to configure a single node or multiple nodes when configuring a node on a single server.

I understood that the reason for configuring multiple nodes can have the advantages of performance and backup by distributing data.

However, I understood that this case is a case of multiple servers, but if it is a single server, can it have the following advantages if it is configured as multiple nodes?

If that happens, I wonder if I can use the replica shard in the single node.

[Rios]

My opinion:

Single node adv:

• simple for configuration and maintenance
• easy to monitor
• less storage, no replica
• cheap
• internal authentication
• excellent for indices 1-50 GB, or a storage in total to 1 TB, good for test environments, big small data

Disadvantages:

• no replica
• no fail tolerance
• no ILM
• no data roles
• ingestion node shouldn't be avoided, it's better to use Logstash on a separated host
• avoid complex queries/dashboards

[DavidTurner]

@Rios

Disadvantages:

• no replica
• no fail tolerance

If you only have one physical machine, a replica is kind of a waste of resources so "no replica" is an advantage IMO, and "no fail tolerance" applies to both options.

[Christian_Dahlqvist]

@sramana235

However, I understood that this case is a case of multiple servers, but if it is a single server, can it have the following advantages if it is configured as multiple nodes?

If your node has multiple disks you can configure multiple Elasticsearch nodes to each use a different disk, which can protect you from disk failures if you have a replica shard configured. The drawback here is that you will use double the storage space to store each shard twice.

If you have a really large node with multiple disks it may therefore make sense to run multiple nodes on it, but I would otherwise generally recommend against it due to the overhead it adds. It is better to deploy multiple nodes across multiple servers as that would allow the cluster to continue operating if a node goes down.

[Rios]

@DavidTurner

If you only have one physical machine, a replica is kind of a waste of resources so "no replica" is an advantage IMO, and "no fail tolerance" applies to both options.

Yes, the single-node without replica is just a feature. From point of CIA, 3nodes are better solution.