Prevention and brain-split of cerebral fissure in Elasticsearch cluster __elasticsearch

What is the brain crack.

In the ES cluster internal network communication problem, the node in the cluster and master communication is disconnected, the node will think that the current cluster does not master, the node from the original cluster will again elect master, so that the original cluster will have more than master, Different nodes are subordinate to different master, which is called Brains (split)

The brain crack can cause the cluster to be in the risk of losing data, because Master is considered to be the highest ruler in the cluster, it is responsible for the generation of new fragments, fragmentation, and so on, if there are two master in the cluster, the data integrity and consistency in the cluster will be broken.

On the problem of brain crack, the best way to deal with it is prevention, by setting configuration parameter minimum_master_nodes to prevent the cluster from appearing multiple master

The Minimum_master_nodes setting is important to the stability of the cluster, which tells the cluster to choose not to select the node as master unless the node is adequately qualified: This is the number of nodes that the node is supported to obtain

(number of master-eligible NODES/2) + 1 (value of all alternate master nodes/2) +1

For example: There are 10 regular nodes (both as data nodes and as master nodes), then Minimum_master_nodes is set to 6

If there are 3 master alternate nodes and 100 data nodes, then the Minimum_master_nodes is set to 2

If there are 2 regular nodes, it is awkward, minimum_master_nodes need to be set to 2, if one of the nodes hung, the cluster will not be available, if set to 1, can not prevent the brain crack

How to configure it.

The first method, which you can add in the configuration file elasticsearch.yml

Discovery.zen.minimum_master_nodes:2

But ES clusters are dynamic and often have operations to add nodes and delete nodes, which causes each node to modify the configuration file and restart the cluster is very bad. Using the second method

The second method, the dynamic API invocation configuration,

Put/_cluster/settings
{
    "persistent": {
        "discovery.zen.minimum_master_nodes": 2
    }
}}

You can use the Curl command

Curl-xput http://localhost:9201/_cluster/settings-d '
{
 "persistent": {
        "Discovery.zen.minimum_master _nodes ": 2
    }
}"

This is a permanent configuration, priority is higher than the configuration in the configuration file, you should modify the configuration, whether you are adding a node or deleting a node.

To view the current minimum_master_nodes value of a configuration cluster

Curl-xget localhost:9200/_cluster/settings

Attach Official document address and picture

Https://www.elastic.co/guide/en/elasticsearch/guide/1.x/_important_configuration_changes.html