0

I'm currently running an Elasticsearch v2.3 cluster (using docker containers) on a single host. (1 gateway, 3 master nodes and 3 data nodes)

In the near future I will be upgrading towards Elasticsearch v5.x and I am questioning whether keeping the cluster-configuration is actually the most optimal way to go in this case.

The cluster-configuration is designed to have a fall-back if a node decides to go down, the only case where I see that happening is if the host might be having issues, which again would result in all the nodes going down, so this advantage becomes invalid.

In my opinion the benefit of running a single Elasticsearch node would be that more RAM becomes available for the node to use? (Since the RAM the other nodes in the cluster setup becomes available...)

Please leave your thoughts and experiences regarding on this...

PS: In the mid-long future I am planning to move towards a PaaS solution (read: Cloud based Elasticsearch service)

Kenny
  • 143
  • 2
  • 14

1 Answers1

0

For Elasticsearch it all depends on your index size. ES will try to fit everything into memory so more memory the better. If you don't have enough to fit your whole dataset your index should at least be in memory. The index is like a table of contents and if ES can quickly look up the index then getting the data will seem fast. Not as fast as if it was all in memory though.

If the index can't fit in memory, ES has to load it from disk each request and that's where you get really slow.

So for memory it all depends on the size of your dataset and index(es) in elasticsearch.

Mike
  • 21,910
  • 7
  • 55
  • 79