Hi, Jason! > On 17 Aug 2016, at 21:53, Jason Gustafson <ja...@confluent.io> wrote: > > Hi Andrey, > > Thanks for picking this up and apologies for the late comment. > > One thing worth mentioning is that the consumer actually sends multiple > parallel fetch requests, one for each broker that is hosting some of the > assigned partitions. Unless you were planning to modify this behavior, this > KIP actually changes the maximum memory used by the consumer from > > max.partition.fetch.bytes * num_partitions > > to > > fetch.response.max.bytes * num_brokers > > I guess it's really the minimum of the two values since > max.partition.fetch.bytes is still supported. I think this is still a very > helpful feature, but it's probably worth calling this out in the KIP.
Good point. I’ll add comment about it. > > Also, one question on naming: would it make sense to change > "fetch.response.max.bytes" to "max.fetch.bytes"? Seems to fit nicer with > "max.partition.fetch.bytes”. > I have no objections. However, we already initiated voting procedure on this KIP, so I am a bit unsure wether can I change KIP now. Jun, what do you think? > > Thanks, > Jason > > Thanks, Andrey.
