We generally recommend the same value for request and limit of memory.

Customers may configure different requests and limits for good reasons, so our recommendations should follow that.

If current memory request = X_1 and current memory limit = Y_1, and the recommended request is X_2, then the recommended limit should NOT be Y_2 = X_2 but Y_2 = X_2*Y_1/X_1 (i. e.  keep the same request/limit ratio as in the current value).

This should be optional and configurable via settings, something like "adjust memory recommendations to minimum usage = true/false". We should do this per cluster (ie part of the cost model), not globally.