Just to clarify, in 0.11.0.1 we have optimized the restoration process so it is actually expected to be faster than 0.10.2.1, so what you observed is indeed very weird. We'd need some more information like log entries to see what's happening there.
Guozhang On Tue, Oct 17, 2017 at 2:50 AM, Damian Guy <damian....@gmail.com> wrote: > Hi Johan, > > Do you have any logs? The state store restoration changed significantly in > 0.11.0.1. If you could get some logs at trace level, that would be useful. > Also if you could provide your topology (removing anything > proprietary/sensitive). > > Thanks, > Damian > > On Tue, 17 Oct 2017 at 05:55 Johan Genberg <johan.genb...@gmail.com> > wrote: > > > Hi, > > > > I'm upgrading a kafka streams application from 0.10.2.1 to 0.11.0.1, > > running against a kafka cluster with version 0.10.2.1. The application > uses > > a couple of state stores. > > > > When stopping/starting the application prior to the upgrade (with > 0.10.2.1 > > client) on 2 instances, it was up and running in less than 30s to a > minute > > on all nodes. However, after client was upgraded to 0.11.0.1, when the > > instances started (during some load), it took about 6 minutes for one of > > the nodes to reach "RUNNING" state, and the second one didn't get there. > > After 10 minutes I had to roll back. > > > > Is it expected that this initial rebalancing takes a little longer with > > 0.11.0.1, and is there a way to configure or tweak the client, or > otherwise > > optimize this to make this go faster? > > > > Thanks, > > Johan > > > -- -- Guozhang
