Hi Ian, That is correct. The consumer fetcher threads get data from multiple partitions in batches, and put batches in to the stream one-by-one, hence when stream iterates, it will likely give a sequence of messages from one partition, then from another one, and so on.
Guozhang On Thu, May 1, 2014 at 4:14 PM, Ian Friedman <i...@flurry.com> wrote: > Hey gang, quick consumer behavior question: > > If I have a Consumer that owns several partitions but only creates a > single stream for them, in what order will that stream get its messages? > From my experiments it seems like the Consumer fetches some amount of > messages for all N partitions it holds, and then processes all the messages > for Stream 1, then Stream 2, etc. until it has run out of messages, then > fetches more. Is that accurate? > > --Ian -- -- Guozhang
