CODEWITHSUNDEEP
couchbasehigh-availability
nntp
nntp

Reputation: 43

Couchbase 3.1: Rack Awareness , Replica and autoFailover behaviors

I have a question on Couchbase(3.1)'s RAZ +Replica+ autoFailover behaviors.

I want to simulate the below configuration and disaster scenario :

2 nodes are on the vm platform that provided by VMWare VCenter(simulate the 1st part members of the couchbase cluster are from a primary production zone). And 3 nodes are on the vm platform that provided by RHEV (KVM based)(simulate the 2nd part members of the couchbase cluster are from a secondary product zone at the same data center)

All these 5 nodes have same OS and same version of couchbase(3.1) bits installed. And configured to one cluster with one bucket created and replica=2 at first round of testing and later changed to 3 for this bucket. "Auto Failover" setting are enabled.

Suppose the three nodes that in the secondary production zone(provided by RHEV platform) are crashed at the same time(simulating secondary zone complete failed, eg. the whole RHEV platform crashed by power outage).

What is the best way to minimize the hurt to the cluster availability and reduce the impaction to the application that connecting/operating the couchbase cluster.(r/w operation).

What is the best combination that fully utilize Rack Awareness, Fine tune Replica numbers and right operation steps for this 3 nodes crashed scenario?

btw: suppose there is no capacity/load problem when doing the failover(or autoFailover) , aka the cascading failure.

Thanks

Upvotes: 2

Views: 103

Answers (1)

NoSQLKnowHow
NoSQLKnowHow

Reputation: 4865

If you are going to use RZA with the two Hypervisors you talk about, I would do this with an even amount. 3 nodes on each HV, all nodes on HV #1 are in Server Group #1 and all nodes on HV #2 are in Server Group #2. You would only need 1 replica though. For the failure scenario you mention when an entire HV goes down, having more than one replica would not help you at all. When running with RZA like I mentioned all of the replicas for the nodes in SG#1 are over on the nodes in SG#2 and vice versa. Hopefully this makes sense.

Also, RZA in the configuration we are talking about is an Enterprise only feature in production so this means you have paid for it (anything over 2 nodes in production has to be licensed). I only mention that because these kinds of things can be brought to your Couchbase Account Manager or the Technical Support team for this and other more detailed help.

Upvotes: 1

Related Questions