Antony Stubbs

FYI see `PollContext` object passed in to lambda to get failure counts of records.

DLQ tracked here: - #310

Had a quick look at it - [something like this with Reactor and BucketJ](https://github.com/confluentinc/parallel-consumer/compare/master...astubbs:features/rate-limiting#diff-9e6185103da281ca730d012e8c67ca198d099f103a60c62ce81dc7ad5a751275R81) should work.

This is a very large change to the way the system works, and affects a lot of tests. POC is done, but will take some time to polish it.

Hi! Yes, I've thought about this many times, and brought it up myself. I'm not aware of anyone trying it yet - but it seems like a very obvious marriage...

Consider the inverse - teach PC to run connectors...

I have started experimenting on this now. Sources or sinks in particular? I have working hacks with both connect inside PC, and PC inside connect. But I think connect sinks...

Oh my! Yes! This is a breaking issue for the poll and produce use case. I will look into it with priority. Too much testing with just String types :)

> i see that you remove the blocker label. Whoops - mistake :) Still plan on changing this. Will be resuming work on PC this week :)

Btw, workaround is to use your own instance of a producer to send a result back to Kafka, of any type.