I have a 2 nodes cluster and doing stress testing now. There are some appearances I can not explain and ask for help here.
When doing search testing. The search thread pool's queue length perform different between two nodes.
When the node1's queue length up, node 2's down. And then reverse. They are go through the rule in these days.
The search requests to each node is similar.
I feel confuse to the queue length's performance.Is there some explanation about that?
thanks in advance,
I guess there is some kinds of load balance mechanism. Is there some doc about it?
What are you using for load testing?
If you are using something that leverages the transport or node client, or any of our officially provided clients then it will automatically round robin, producing this behaviour.
Thanks for your answer~
We use Jmeter for load testing. The test case open 200 threads to send HTTP requests to our cluster.
We also try another test case: open 200 threads to keep sending HTTP requests to only node 1. In this case, two nodes in the cluster got busy and the queue length still acted in that behavior. It seen like search request can be redirect between two nodes. Is that what you said "transport"?