Kyle Kim (kimkyle@)

Extending on @madolson‘s comment, we (AWS) would like to gain open source community's feedback on our proposed design for tracking per-slot memory usage. This problem can be broken down into...

@PingXie It's a valid point you raised. We have previously reviewed this approach during our internal design review process, with details attached below. TLDR; Our proposed design candidate still stands...

> Can we also consider some indication in the output that the slot memory usage could be off due to the presence of such a module? A different stat name,...

It's been a while since the last update was posted! Here's our update thus far - Based on open-source community's interests, we have been investing time & efforts into performance...

@madolson Understood. If this is the case - isn't the "zmalloc state" approach identical to @PingXie 's proposed solution earlier, which has been pruned? His comment is attached below; >...

@zuiderkwast Thanks your feedback on the design candidate. Regarding CPU impact, please refer to the performance benchmarking results below. #### Notes - Read commands (ex: `get`, `hget`) are not impacted....

Aside from CPU / TPS impact, memory overhead has been noted down within my previous comment [here](https://github.com/redis/redis/issues/10472#issuecomment-1138070198). > - Introduction of `size_t size` parameter for memory sparse internal data-structures increases...

@madolson That's correct. In fact, the relative performance hit would be slightly better than CMD case. Reason being, CME has a few additional computation related to clusters, which dilutes the...

Here's the summary of the worst-case memory impact, along with its reasonable counter-parts. Please note, while the % may suggest worst-case memory impact being detrimental, the absolute memory difference will...

The following is the breakdown of each data-structure's extra memory consumption. #### Hash & Set (using dict), 16 bytes From `dict`, total of 16 bytes. - 8 bytes from dict->`size_t...