I see, thanks. Yes, re-assignment. Got my terminology off. On Tue, Nov 24, 2015 at 4:45 PM, Gwen Shapira <g...@confluent.io> wrote:
> ah, re-assignment! > > If your re-assignment involves moving leaders to other servers, there will > be a tiny downtime when the new leader election happens. > Otherwise the main risk is copying the partition over to the new node - it > can use lots of network and IO. On a busy system we recommend scripting > this to do very few (maybe one) partition at a time. > > On Tue, Nov 24, 2015 at 4:42 PM, Dillian Murphey <crackshotm...@gmail.com> > wrote: > > > Not adding. Taking some of the partitions from one kafka server and > > spreading them to another. > > > > On Mon, Nov 23, 2015 at 5:40 PM, Gwen Shapira <g...@confluent.io> wrote: > > > > > By re-partition you mean adding partitions to an existing topics? > > > > > > There are two things to note in that case: > > > 1. It is "hitless" because all it does is create new partitions where > > > future records can go, it does not actually move data around. > > > 2. You could be "hit" if your consumer code assumes that all data for a > > > specific key will be in the same partition. Because all future keys may > > get > > > assigned to different partitions than what they had before adding > > > partitions. > > > > > > Hope this helps. > > > > > > Gwen > > > > > > On Mon, Nov 23, 2015 at 9:38 AM, Dillian Murphey < > > crackshotm...@gmail.com> > > > wrote: > > > > > > > Can I do this on a production system and not have downtime? I'm > using > > > > kafkamanager to make this easier, but it's just running the > > re-partition > > > > task. > > > > > > > > > >
