Is anyone already working on how to intelligently re-partition slots across the cluster? If not, I think providing this kind of data would enable 3rd party developers to start coming up with algorithms for re-partitioning based on slot data size and time spent executing against a slot.
I imagine this working by allowing the user to enable a CLUSTER SLOT SAMPLE mode. After which, each node, would track the time spent for each slot using before/after timestamps for each command. To preserve performance the user could specify a sample percentage (1-100%). The user would also supply a result key to store this data structure in order to retrieve it later... The reason for running this on each node is to account for read-only workloads as well, maybe this could be an option in case the user has no read only requests.
Sorry if this has already been discussed, only did a small amount of digging.