Everyone knows that a big cluster state might be a problem, bug how big it can safely go? How much is "too big" and how much is "ok"?
I guess this depends a lot in the network bandwidth between the nodes, but I would like to find some sort of formula that can answer the first question here.
So, I would like contributions to collect some performance numbers using the following command-line. Maybe we can find some sort of general threshold:
My reference cluster has 10 nodes with 7GB heap, and SSDs, all nodes are connected with gigabit interface. The current performance is good.
ES=es-node:9200 time curl -s http://$ES/_cluster/state?pretty | wc -lc 62955 2540673 real 0m0.087s user 0m0.000s sys 0m0.008s ES=es-node:9200 curl -s http://$ES/_cat/shards | wc -l 364