Yuanbo Liu Thanks for your patience. I am not fully convinced that we should do this. Let us look at the downside of not having this patch. As Arpit Agarwal mentioned it can lock up a requests when we are shutting down datanodes. However if datanode is shutting down, queuing the commands is not really useful right ?. Just to make sure that I am not missing something, what do you think is the real benefit of fixing this issue ?
Arpit Agarwal I would love to get your perspective on this, the code change is minor, but I am not sure if we are achieving anything significant. An alternative approach would be to do actually not shutdown the thread whenever we are done balancing or in the cancel path – Arpit Agarwal mentioned the same in an earlier code review.