Strong consistency in XDR architecture


#1

I was reading the various XDR topologies and I’ve got a query regarding the consistency guarantees(assuming I’m using the enterprise version). I was interested in the simple Active-Active topology. What happens in case the link goes down between the 2 clusters? Since clients can write on any of the clusters, therefore, how will the consistency be ensured?


#2

XDR does async replication between datacenters and there is no coordination between the XDR writes and the client reads. So, it cannot achieve strong consistency across clusters. If you are configuring strong consistency, it is only applicable to the local cluster. XDR writes will be like one more client writing to that cluster.


#3

Actually in my use case, strong consistency is must. But I want to understand this feature(XDR). What exactly is the benefit of having 2 separate clusters(having n nodes each) in XDR as compared to 1 cluster(having 2n nodes)?


#4

If the nodes are geographically close, say in the same building or across buildings where network latency is <1 or 2 ms, it is definitely better to have them as a single cluster. XDR is mainly meant to maintain data locality and disaster recovery across geographically distributed nodes where forming a single cluster is not possible or optimal. And XDR has its limitations on what it can guarantee and strong consistency is one of it.

Seems you are an enterprise customer as you are talking about SC and XDR. Please reach out via support channel. We can include our solution architect team to go over your usecase and see how we can handle your requirements.


#5

Could you describe your project? What level of consistency guantees does it require within a DC and what level across DCs (linearizabe, sequential, causal, etc)?


#6

This topic was automatically closed 6 days after the last reply. New replies are no longer allowed.