Consumers can fetch messages up to the high watermark, which is dependent on the in sync replicas, but not directly dependent on `min.insync.replicas` (e.g. if there are 3 in sync replicas, the high watermark is the min of the log end offset of the 3 replicas, even if min in sync replicas is 2).
Ismael On Tue, Sep 26, 2017 at 1:34 PM, Denis Bolshakov <bolshakov.de...@gmail.com> wrote: > By default kafkf does not allow dirty reads for clients, so while > `min.insync.replicas` > is not achieved consumers don't see new messages. > > On 26 September 2017 at 11:09, Sameer Kumar <sam.kum.w...@gmail.com> > wrote: > > > Thanks Stevo for pointing me out to correct link. > > In this case, how would exactly once feature of streams would behave > since > > they configure producers with acks=all. I think they would fail and would > > need to be resumed once the broker comes back. > > > > -Sameer. > > > > On Tue, Sep 26, 2017 at 1:09 PM, Stevo Slavić <ssla...@gmail.com> wrote: > > > > > Hello Sameer, > > > > > > Behavior depends on min.insync.replicas configured for the topic. > > > Find more info in the documentation > > > https://kafka.apache.org/documentation/#topicconfigs > > > > > > Kind regards, > > > Stevo Slavic. > > > > > > On Tue, Sep 26, 2017 at 9:01 AM, Sameer Kumar <sam.kum.w...@gmail.com> > > > wrote: > > > > > > > In case one of the brokers fail, the broker would get removed from > the > > > > respective ISR list of those partitions. > > > > In case producer has acks=all, how would it behave? would the > producers > > > be > > > > throttled and wait till the broker get backed up. > > > > > > > > -Sameer. > > > > > > > > > > > > > -- > //with Best Regards > --Denis Bolshakov > e-mail: bolshakov.de...@gmail.com >
