10/19/2022, 12:51 PM
Odd issue - Running rancher 2.6.7 with a very large cluster. I have noticed that over time, in a cluster with 15k workloads deploying and undeploying, that one random etcd node will always consume much more memory than others over time, and this node will continue to use more and more memory until restarted. ETCD has periodic auto-compaction, so the size of the DB is not the issue. Is there some sort of cache that can be cleared? It goes away if you restart it, but that shouldn't be the solution.