Simple producer benchmark

[svroonland]

It's not much, but it gives us something to compare against.

Relates to #1304

[erikvanoosten]

Great start! I propose we:

• move construction of ProducerRecord to outside the benchmark
• in the benchmark we publish 100 chunks of 5000 small messages

WDYT?

[erikvanoosten]

I just realized that there is one super-tiny thing we can improve: use a binary key/value. Because we now using string, there is the tiny overhead of converting string to byte array. I am not sure if you want to put in the effort, and this is most probably already good.

[erikvanoosten]

FYI, I have removed the first run of this new bench from the charts at https://zio.github.io/zio-kafka/dev/bench/ because it was producing a lot less data and is therefore not comparable with later runs.

[svroonland]

I just realized that there is one super-tiny thing we can improve: use a binary key/value. Because we now using string, there is the tiny overhead of converting string to byte array. I am not sure if you want to put in the effort, and this is most probably already good.

I think it doesn't really matter since this isn't currently a benchmark about how much performance we can squeeze out of the Producer. It's just to have a reference for future changes.

[erikvanoosten]

You're right, good point 👍

[svroonland]

Would be nice to have a "max performance" benchmark though, using the *async methods, batching and clever pipelining.