Bartlomiej Plotka

Yea, there is also [csproto](https://github.com/CrowdStrike/csproto) on top of [vtprotobuf](https://github.com/planetscale/vtprotobuf), plus changing out of gogo can change things a lot. I plan to setup Prometheus + receiver code indeed and benchmark...

Yes, there will be always option to negotiate/configure what protocol to use. However for your microcontroller case, new protocol will be only better. You can't partially decode proto anyway, and...

Did some early benchmark without any extra optimizations, just all plugins in fastest possible mode. Using `unmarshal_unsafe` on vtproto which is only on main (not released yet) yields the best...

However, what's the final proto form with that []string idea? I would propose something like https://github.com/bwplotka/go-proto-bench/blob/main/proto/prw/v2testvtproto/write.proto, so PRW 2.0, with another package e.g. `remotewrite.v2` (!), separate to remote read even....

Ayay, sorry for delay. Will try to look this/next week. Thanks!

> Is label ordering a 'hard requirement', i.e. should such samples be considered invalid and we should refuse to append them and treat them similarly to empty / duplicate labels?...

Amazing, thanks, seems this issue can be closed then!

Just TL:DR: I would be really careful with replacing APIs. Perhaps there is a way to built on top of what we have right now AND define stable identifier and...

Side note: If we are considering changing that number - what about making this (e.g. hidden for now) flag?

I would say we need a flag.