Running 4.6.4, I have an AWS deployment that is over 3 AZ’s, meaning 3 groups and I have 2 replicas. Now this appears to mean I have 1024 vbuckets for active and 2048 vbuckets for replica. What this also means is I only have 48% of my active documents in resident memory and 33% in replica…
Now 
My primary cluster is across 2 AZ’s and has a single replica (2 groups), this cluster has 75% active docs resident in memory and 53.5% replica docs in resident memory
Same cluster configs, same memory footprint, same counts of data. Yet the 2 replica, 3 group cluster is performing pretty poorly and the resident memory shows me, that is a big point of contention.
I’d rather have 75% of my active data in memory and 25-30% of the replica (we don’t miss a lot when we ask the system for the active data!)
So at this time I think the 2 replicas are kicking me in the backside and I need to figure it out and increasing the cluster size is not really an option (and note that it’s working well with 1 replica/2 groups, just seems to be weighted more for replica than active in a 2 replica/3 group config.
Any assistance in tuning this? Get more active docs in resident memory?
Couchbase suggested the 2 replica, 3 groups in a 3 AZ environment (makes sense), but it needs to be cleaner and again lean more towards giving active data RAM vs replica data.
Guess adding a visual might help.
vbuckets
