Hi Gaurav, Its more about being able to reduce the required number of replicas when you have a cluster distributed across multiple racks/DCs.
As a basic example if you have a 4 broker cluster across two Racks/DCs/Regions any topic you create you will need to have to have a RF >= 3 and min.insync.replicas of at least 3 to guarantee there was at least one replica on each Racks/DCs/Regions for site failure resilience. That's 3 sets of acks and (associated latency) required for the message to be committed. This gets even trickier the more Racks/DCs/Regions you are spread across. By setting min.insync.racks=2 and min.insync.replicas=2 you only need 2 acks. As a secondary benefit, if you have a topic with a replication factor of 2 then it's possible for both replicas to end up in the one Rack/DC/Region if the original partition placement logic doesn't spread it across multiple racks or if someone has manually moved partitions around for whatever reason. Karl ________________________________ From: Gaurav Narula <[email protected]> Sent: 05 March 2026 14:50 To: [email protected] <[email protected]> Subject: Re: [DISCUSS] KIP-1290: Rack-Aware Minimum In-Sync Replicas Hi Karl, Thanks for the KIP. I'm intrigued behind the motivation. Usually we see the number of replicas to be <= number of fault domains. I'm curious to know what's the motivation behind running with a higher replica count - as in what's the extra number of copies getting you? Regards, Gaurav > On 5 Mar 2026, at 13:39, Karl Sorensen <[email protected]> wrote: > > Hi, > > I haven't had any replies on this so just bumping. > > I've got a patch that I've tested. > > Thanks > Karl > > > ________________________________ > From: Karl Sorensen <[email protected]> > Sent: 25 February 2026 11:27 > To: [email protected] <[email protected]> > Subject: [DISCUSS] KIP-1290: Rack-Aware Minimum In-Sync Replicas > > Hi, > > I'd like to start a discussion about KIP-1290 > https://cwiki.apache.org/confluence/x/ao48G > > This proposes adding a new config min.insync.racks to ensure producers write > to replicas spread across multiple racks. > > Let me know if you have any questions. > > Thanks > Karl
