schreter

BTW, in our project, we also noticed performance degradation because of this regression (small 16B `Result`s returned via the stack instead of registers), so I vote for a quick fix....

> I would sincerely appreciate it if this was quantified in some manner or if you could give a more complex example. @workingjubilee Yes, we need to quantify this, but...

@workingjubilee I've tried to implement something with `black_box`, unfortunately, Rust happily optimizes it out, so I have to find a different approach (didn't have much time yet, sorry). What we...

> This requires more thought on how to do it right, I'm worried about cases of multiple writers to a single commit-id, where there could be cases that one person...

Just a note: please no `Vec`. We need to get rid of them in all places in favor of something zero-cost, like `impl IntoIterator` or similar, since a `Vec` is...

> Does It have to have another generic type argument if we use the trait IntoIterator to describe the type of rpc? You are right, it is not optimal to...

Well, I was planning to suggest implementing lease read myself :-). This indeed can speed up the reads considerably. OTOH if I understand it correctly, there is the additional cost...

> Raft defines its own time with term and log index. Hehe, I didn't intend to make a pun on "time" :-). The "transaction time" we use is a kind...

> Maybe the election timeout can be replaced with some external event source that triggers re-election. That's indeed a good idea and it would help in our project too (we...

> In short, replace tokio runtime with a customized one, right? Yes. There are just a couple primitives that are needed in the implementation. Implementing them trivially for test purposes...