Graceful Cluster Shutdown


#1

What is the preferred method now(1.7.x) for a graceful shutdown of an entire cluster?

Looks like the _shutdown option is going away soon...
https://www.elastic.co/guide/en/elasticsearch/reference/current/cluster-nodes-shutdown.html


(Magnus B├Ąck) #2

As the ES 2.0 documentation page for the _shutdown API indicates, install ES as a service and shut it down by shutting down that service. In the Unix case this will in the end translate to a SIGTERM signal to the ES process.


#3

install ES as a service and shut it down by shutting down that service
Understood and am already doing this.

I was more concerned in the order of shutting things down in a larger sized cluster where master/client/data nodes are on separate machines to prevent a master election and/or shard re-allocation.


(Mark Walkom) #4

Disable allocation and then shut things down, it doesn't really matter in what process but it'd make sense to do clients then data then masters.


(system) #5