We deploy Couchbase 7.0.0 Beta nodes on top of several Kubernetes clusters, where the K8S version, OS version, and the underlying hardware is slightly different.
We use the latest CoreOS, Kubernetes version from 1.17 to 1.18 on Intel HNS2600 nodes in general.
Fresh deployments of 1 to 4 nodes clusters usually startup with 1 bucket and then tests may make Couchbase ingest a few 10s or 100s of documents. The number of buckets/scopes/collections or data volume seems to be irrelevant, in - about a handful of days of running (20-30 days) - the CPU load goes up to 3.0-4.0 CPU. This means, for all of our CB .7.0.0 Beta clusters running, irrespective of data size and server load, we pay +25% upkeep cost in terms of CPU resources, even if there are 0 operations on all of them.
We suspect that there is a resource leakage in the long run.