Scaling Instances to save $

I’m inheriting our MemSQL clusters from a team of developers, and their use is not well established yet in the org. Trying to get them managed/rightsized.

What are practices people are using to save money on MemSQL cluster infra costs, both prod and non-prod? Shutting down at night? Scaling down some nodes?

Is there a way to dynamically scale CPU allocation? Or a good/safe way to power down unused clusters on nights/weekends (eg, dev instances)?

We have a 10x8vCPU Prod cluster and 2x 6x8vCPU DEV and UAT cluster. I think they all have more vCPU than is necessary at our current state, but I have no insights into how to safely size/scale the infrastructure as needed.

How do you size your master/child nodes vs. your leaf nodes, from a CPU/Memory/Disk standpoint? Right now we have identical nodes in terms of CPU/Memory/SSD, but that doesn’t seem necessary.

Any advice on these terms would be appreciated.

For longer-term adjustments to cluster capacity, adding or removing leaf nodes and then doing a REBALANCE typically works well.