Guozhang, i'm not sure i follow... Global stores aren't per task, they are per application instance and should be fully restored before the stream threads start processing. They don't go through a rebalance as it is manual assignment of all partitions in the topic.
On Thu, 9 Nov 2017 at 17:43 Guozhang Wang <wangg...@gmail.com> wrote: > Instead of restoring the global store during registration, could we also do > this after the rebalance callback as in the main loop? By doing this we can > effectively swallow-and-retry-in-next-loop as we did for non-global stores. > Since global stores are per task not per thread, we would not process the > task after the global store is bootstrapped fully. > > > Guozhang > > On Thu, Nov 9, 2017 at 7:08 AM, Bill Bejeck <bbej...@gmail.com> wrote: > > > Thanks for the KIP Matthias, +1 from me. > > > > > > -Bill > > > > On Thu, Nov 9, 2017 at 8:40 AM, Ted Yu <yuzhih...@gmail.com> wrote: > > > > > lgtm > > > > > > bq. pass both parameter > > > > > > parameter should be in plural. > > > Same with 'two new configuration parameter' > > > > > > Cheers > > > > > > On Thu, Nov 9, 2017 at 4:20 AM, Damian Guy <damian....@gmail.com> > wrote: > > > > > > > Thanks Matthias, LGTM > > > > > > > > On Thu, 9 Nov 2017 at 11:13 Matthias J. Sax <matth...@confluent.io> > > > wrote: > > > > > > > > > Hi, > > > > > > > > > > I want to propose a new KIP to make Streams API more resilient to > > > broker > > > > > disconnections. > > > > > > > > > > > > > > > https://cwiki.apache.org/confluence/display/KAFKA/KIP- > > > > 224%3A+Add+configuration+parameters+%60retries%60+and+% > > > > 60retry.backoff.ms%60+to+Streams+API > > > > > > > > > > > > > > > -Matthias > > > > > > > > > > > > > > > > > > > > > > > -- > -- Guozhang >