Using Digital Ocean Droplets as Client Nodes

Would a small digital ocean droplet generally work well for a dedicated Kibana/Elasticsearch client node? Would a $20/month (2 GB RAM/2 CPU) work well in most circumstances? Would a $5/month (512 MB RAM/1 CPU) droplet even be viable? I’m unsure of the actual load that splitting search requests & combining/reducing results puts on a client node.

My planned setup is 4-6 ES data nodes (8 CPUs each, ~100 GB RAM total), indexes up to 1 TB each, each index with 1 shard per data node, and serving up dashboards potentially for up to a couple dozen people at a time.


Depends how big your queries are.

So, RAM & CPU needs of a client node will mainly be determined by the amount of data the client node is handling in the search reduce/"gather phase" of a query, and the number of queries being performed simultaneously.

Is the work being performed during the gather phase on a client node comparable to or much less than the work performed while executing the query on one of the data nodes?