Dev Ojha

Yes! Been mildly delayed due to this breaking an api in the sdk, and our tooling messed up on both sdk and comet patches at once. Going to try again...

I just did this on some known hot loop locations I saw due to merge conflicts, and saw a very significant improvement: https://github.com/osmosis-labs/cometbft/commit/30a125b6f4702c417d67422e37643597048b9e45 Savings were over 20GB of RAM. We...

Yeah I think just doing it on the hot spot ones sounds great to me!

Ah sorry missed this ping (please feel free to ping on slack if I miss something in the future) Definitely noticing the cost of logging in: - `peerstate.SetHasVote` (which got...

I've removed several more that are bottlenecks in the p2p stack. Forgot to post here, will reply with more as I remember / re-see them. (Maybe when we do our...

I recommend we close this and do some simpler PR's to reduce complexity. It may be helpful to examine some of the simplification as what can be re-used between the...

I think what we should do here is: - Maintain statistics via Atomic instructions in the mempool for "how many bytes", "how many gas" is rechecked, and "are we done...

Agreed, I also think that in practice this may also need #2955

LazySprintF allocates all the arguments as a slice unfortunately, so it is still doing heap allocations and is expensive in hotroutines or with several args. That should fix the first...

Heres a screenshot for SetHasVote, sorry for not including that in OP:  We see that 5GB are in (what I assume is) allocating args inside LazySprintF. (Otherwise it may...