Madelyn Olson

@kyle-yh-kim As we talked about internally, I only think we need to be tracking TPS, which is a proxy for the execution time of individual commands on the server side....

I'm aware of both of those comments, and they aren't addressing that there are relatively easy ways to identify those other types of memory allocations. Data structures are fairly well...

@kyle-yh-kim I also don't think the used_memory_redisdb can realistically work, as there are places where zmalloc is called for both purposes. Like dict, which stores the main dictionary as well...

@oranagra @yossigo @soloestoy @zuiderkwast Pinging some individual people this time. The general conclusion of this thread is that in order to best track the memory used per-slot in CME, the...

@kyle-yh-kim Did you run any tests where CME was enabled but metrics collection was disabled? I presume the performance would then be equivalent to the CMD case. I also want...

Update from an internal core notes: * Performance generally looks ok * Memory usage is hard to follow. We want to review an updated memory impact highlighting the potential worst...

Thanks for this data @kyle-yh-kim. Overall this data seems really good to me. There is the separate project for improving main memory efficiency of the dictionary, so if these two...

> btw, lists do have dense encoding (quicklist is an evolution of both linked-list and ziplist, should have the best of both worlds). By the definition of dense in this...

> why do you sum the extra pointer in hashDictType / setDictType / zsetDictType? it's one instance per server (not per key) Agree with Oran about this, last time we...

> i think we generally want to avoid adding more MULTI_ARG_CONFIGs and we have a general understanding that the interface of client-output-buffer-limit is horrible. If you have any type of...