Hi Neha and Guozhang, As long as stickiness is maintain consistently to a particular partition in target DC that is great so we can do per DC and across DC aggregation.
How about non hash based instead of range based partitioning ? eg Key start with "a" then send message to partition 1 to 10, if key starts with b then partition 11 to 20 and so on & so forth... Is this case how does MM handle copying data ? This is just for FYI for now we are in process of upgrading to new producer then how will MM distribute data to target DC if partition number are different etc ? Basically, how can I inject MM with my custom partitioning logic ? Thanks for your help !! Thanks, Bhavesh On Mon, Aug 11, 2014 at 10:20 PM, Guozhang Wang <wangg...@gmail.com> wrote: > Bhavesh, > > As Neha said, with more partitions on the destination brokers, events that > are belong to the same partition in the source cluster may be distributed > to different partitions in the destination cluster. > > Guozhang > > > On Mon, Aug 11, 2014 at 9:35 PM, Neha Narkhede <neha.narkh...@gmail.com> > wrote: > > > Bhavesh, > > > > For keyed data, the mirror maker will just distribute data based on > > hash(key)%num_partitions. If num_partitions is different in the target DC > > (which it is), a message that lived in partition 0 in the source cluster > > might end up in partition 10 in the target cluster. > > > > Thanks, > > Neha > > > > > > On Mon, Aug 11, 2014 at 7:23 PM, Bhavesh Mistry < > > mistry.p.bhav...@gmail.com> > > wrote: > > > > > Hi Guozhang, > > > > > > We are using Kafka 0.8.1 for all producer consumer and MM. > > > > > > We have 32 partition in source (local) per DC and we have 100 in target > > > (Central) DC. > > > > > > Is there any configuration on MM for this etc ? > > > > > > Thanks, > > > > > > Bhavesh > > > > > > > > > On Mon, Aug 11, 2014 at 4:33 PM, Guozhang Wang <wangg...@gmail.com> > > wrote: > > > > > > > Hi Bhavesh, > > > > > > > > What is the number of partitions on the source and target clusters, > and > > > > what version of Kafka MM are you using? > > > > > > > > Guozhang > > > > > > > > > > > > On Mon, Aug 11, 2014 at 1:21 PM, Bhavesh Mistry < > > > > mistry.p.bhav...@gmail.com> > > > > wrote: > > > > > > > > > HI Kafka Dev Team, > > > > > > > > > > > > > > > > > > > > We have to aggregate events (count) per DC and across DCs for one > of > > > > topic. > > > > > We have standard Linked-in data pipe line producers --> Local > Brokers > > > --> > > > > > MM --> Center Brokers. > > > > > > > > > > > > > > > > > > > > So I would like to know How MM handles messages when custom > > > partitioning > > > > > logic is used as below and number of partition in target DC is SAME > > vs > > > > > different > > > > > than the source DC ? > > > > > > > > > > > > > > > > > > > > If we have key based messages and custom partitioning logic ( > > hash(key) > > > > % > > > > > number of partition per topic source topic) we want to count event > > > > > similar > > > > > event by hashing to same partition and count events, and but when > > same > > > > > event is MM to target DC will it go to same partition even though > > > number > > > > of > > > > > partition is different in target DC (meaning does MM will use > > hash(key > > > > > message) % number of partition) ? > > > > > > > > > > > > > > > > > > > > According to this reference, I do not have way to configure this or > > to > > > > > control which partitioning logic to use when MM data ? > > > > > > > > > > > > > > > > > > > > https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=27846330 > > > > > > > > > > > > > > > Thanks, > > > > > > > > > > > > > > > > > > > > Bhavesh > > > > > > > > > > > > > > > > > > > > > -- > > > > -- Guozhang > > > > > > > > > > > > > -- > -- Guozhang >
