Setup: Couchbase Server 6.5 Beta EE on two nodes. Nodes connected via switch with 1 Gbps.
This post is about my misconception. It isn’t about “this query is slow” or “how can I create an index to speed things up”.
I intentionally excecuted a slow long-running query. Setup is two nodes, both have the query service. Below are screenshots of how many gets per second are executed. As I understand gets per second is a good measure on how fast this particular query runs. The query is the last one run in the screenshot which took several minutes. Both nodes average around 12.5k gets per second. Together around 25k gets per second.
Setup is a single node. Average gets per second is ~ 30k gets per second.
Setup is two nodes. Only one node has query service. Average gets per seconds and graph spikes are very similar to the first setup.
Conclusion is that a single node is faster than 2 nodes. My conception was that several nodes share the query execution. I also hoped that it scales nearly linearly. Clearly I’m wrong and I’m looking for some insight.