Node Management APIs

(Rajesh Nataraja) #1

Are there APIs to add and remove Nodes from Elastic Search cluster? Or modifying and restarting elasticsearch is the only way?

(Mark Walkom) #2

There is no API, it needs to be done in the config.

(Rajesh Nataraja) #3

What is the best way to reload elasticsearch? Is there an API that causes ElasticSearch to reprocess the yml file?

(Mark Walkom) #4

You need to restart the service.

(Rajesh Nataraja) #5

Interesting, What are the ways to add a node into cluster without downtime? Doesnt restarting elasticsearch cause downtime? In some dynamic environments it may be difficult to implement a coordination of restarting one after the other.

(Mark Walkom) #6

There is no downtime required to add more nodes to a cluster. What makes you think that?

(Rajesh Nataraja) #7

So if the only way is to modify elasticsearch.yml then dont I have to restart es everywhere where that node needs to be reflected. That is after modifying es.yml everywhere.

If not what is the other way? Also in our scenario VIP points to all the elastic nodes and restarting elasticsearch at one place can cause failure of requests to that node which may have been restarting at that time. How is this avoided?

(Ivan Brusic) #8

I am assuming Rajesh might be worried about updating the unicast node list
on the existing nodes.

If the new node contains a list of valid nodes in the cluster, it will join
the cluster and the cluster state will be shared with all the nodes in the
cluster to each other. The new node just needs to know one single current
host. Of course, the maintainability might suffer in the long run if
cluster continues to expand and each older nodes does not have the newer
nodes in its unicast host list, but the cluster state should work smoothly.

(Rajesh Nataraja) #9

When a new node is getting added, We can populate entire node list into this new node and start it fresh. This will let the new node discover entire cluster.

For the existing nodes in the cluster, from Ivan's reply I understood, they can be untouched and they will essentially get their cluster figured out due to the new node having the information.

This is good.

When a node is decommissioned, how is this handled?


(Mark Walkom) #10

This is a good example of why dedicated masters makes sense.
That or abstract the list to a DNS alias.

(system) #11

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.