Cluster Elasticsearch


(Samir) #1

Hello EveryOne,

I have a question about elasticsearch clusters which are based on different geographical sites.
What are the real consequences ? Only on the performance of ElasticSearch or also on the data?

And if the connection between 2 sites is very reliable and almost identical than on a single site, is it possible to base the cluster on these 2 sites ?

Thank you very much.


(Luca Wintergerst) #2

Yes, things will be slower, but also more likely to fail.

Let me say first that we do not recommend having clusters span multiple datacenters.
Everyone tells us that "MY connection is reliable and fast!". And that might be right. Until it's not.

The network between datacenters is usually much more susceptible to failure or interruptions and that will cause problems in the long run.

That said, we see a lot of people doing it regardless. From my experience anything below 10ms latency works well, provided that the network is stable.

AWS has a concept called availability zones, where they have separate datacenters right next to each other. Each of these has their own power supply and uplink. Since they are so close to each other we DO recommend to span a cluster across them in this case.
If you DCs are in San Francisco and New York, then I would advise against doing that.

We are also working on making cross datacenter replication a thing, so you can have two separate cluster which contain the same data.

Let me know if that answer helps you,

Luca


(system) #3

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