The way I do it now is:
For each combination of topic name and partition number, I get
LogEndOffset and LogStartOffset from the JMX metrics, and do the
difference between them. I believe this will reflect the queue lengths.
Does that sound reasonable?
Second issue - I've found something called Offset Explorer, which also
seems to provide the queue length for each topic / partition. The
numbers it provides are a perfect match for my results, at least so far.
I am only using it to check the results of my automation.
https://www.kafkatool.com/
Is that tool considered reliable?
--
Florin Andrei
https://florin.myip.org/
