Hi Dong,

Thanks for your reply.

Yeah I agree with you that the total disk capacity can be useful
> particularly if it is different across brokers but it is probably of
> limited use in most cases. I also expect that most users would have their
> own customized tool across to determine the new partition reassignment
> after retrieving the partition distribution using DescribeDirsRequest.

By not providing a tool, you're just forcing people to write their own. So
your expectation will be self-fulfilling. Surely it would be better if the
project provided a tool (perhaps one which did the boring bits and gave
people the option to provide their own customized algorithm).

> And
> that customized tool can probably be easily provided with the configuration
> (e.g. disk capacity, IO parameters) of the disks in the cluster when user
> runs it.

Sure, but it would be better if a tool could discover this for itself. At
best you're forcing people into getting the information out-of-band (e.g.
via JMX), but worse would be if they end up using static data that doesn't
change as their cluster evolves over time.

> I am relatively neural on whether or not we should add this field. If there
> is no strong reason to add this field, I will add it if one or more
> committer recommends to do this.

I don't think we should add it to KIP-113: It could be added at a later
date easily enough.



Reply via email to