Configuring a multiple data center Elasticsearch cluster

Important

Elasticsearch is only supported up to version 6.8, which is no longer supported by Elastic. We therefore recommend using Storage as a Service instead, or if you do want to continue using self-managed storage even though this is not recommended, using OpenSearch.

When the nodes in the Elasticsearch cluster are hosted in different data centers, racks, or zones, you should use allocation awareness. This will ensure that the data is correctly spread between the different locations and that you will still have all your data in case the connection to a location is lost.

There are two kinds of allocation awareness: shard allocation awareness and forced awareness. The difference between these two is the way they handle the shards when a location is suddenly unreachable. Shard allocation awareness will assign the shards of the missing replicas nodes that can still be reached in the other locations. This can cause a big load on those nodes. If your nodes would not be able to handle this, you can solve this by using forced awareness instead. Forced awareness will never allow copies of the same shard to be allocated to the same locations.

Important

We recommend using shard allocation awareness, as we have not yet tested setups with forced awareness.

Tip

For more information , refer to Allocation Awareness in the Elasticsearch documentation.

Note

A similar configuration exists on OpenSearch. There you can also configure allocation awareness.

Configuring allocation awareness

To configure allocation awareness, you will need to add some parameters in the elasticsearch.yaml file on each node. To do so:

Stop the Elasticsearch service on the Linux server.

For Debian-based systems, you can do this with following command:
```
sudo systemctl stop elasticsearch.service
```
Open the elasticsearch.yaml file. By default this is located in the /etc/elasticsearch/ folder.
Adjust the configuration as necessary.

For example, below you can see what the configuration would look like in a 6-node cluster spread over 2 data centers:
- Node 1, 2, and 3 in data center 1:
```
node.attr.DC: DC1
cluster.routing.allocation.awareness.attributes: DC
```
- Node 4, 5, and 6 in data center 2:
```
node.attr.DC: DC2
cluster.routing.allocation.awareness.attributes: DC
```
With this example configuration, a new attribute called DC is created for the allocation awareness, and it is configured with the values DC1 and DC2.
When each elasticsearch.yaml file has been changed, start the Elasticsearch service again.

For Debian-based systems, you can use following command:
```
sudo systemctl start elasticsearch.service
```

Notify the Elasticsearch cluster that this has been configured. You can do this with a PUT message.

This can be done from Kibana or from a Postman session:

PUT http://[IP of node]:9200/_cluster/settings
{ 
    "persistent": 
        { 
            "cluster.routing.allocation.awareness.attributes" : "DC" 
        } 
}

The Elasticsearch cluster will now start moving data. Depending on how much data is already stored in your database, this can take a while.

Table of Contents

Configuring a multiple data center Elasticsearch cluster

Important

Important

Tip

Note

Configuring allocation awareness