On this page you will find cluster configurations that provide ways you can help setup your quorum/witness alongside existing nodes within an AWS cloud environment.
Key for diagrams below:
2 Node, Single Region Deployment
If both nodes all reside in the same region, then the witness node should reside in same region. If any availability zone in the region fails, you still have one node and a witness node. If the entire region fails you have no nodes anyway, so failover becomes irrelevant.
3 Node, 2 Region Deployment
In a 3-node cluster where two nodes live in one region and the 3rd lives in a different region then a witness node does not need to be added to the cluster. This is because we already have an odd number of votes. Failover to the DR region will always be a manual process in the event of the failover of the entirety of region 1.
2 Node w/Witness, Multi Region Deployment
In a 2-node cluster where one node lives in region 1 and node 2 lives in region 2, the witness should reside in region 3 if you have it, but region 1 if you don’t have it. The reason to not put it in region 2 is that a failure of the network between region 1 and region 2 will cause a failover to region 2 unnecessarily. Having a 3rd region eliminates that possibility. If the 3rd region in not practical, then putting it in the primary region is preferable to eliminate false failover. However, in the event of a complete failure of region 1, you will have to bring the secondary server online manually. When this event occurs, please be sure to log the time/date of the failure and specifically what was done to bring the server online.
Alternative Method
2 nodes and storage disk
Amazon S3 has the concept of globals, S3 buckets are globally accessible. When using Amazon S3, it is required to use S3 in a different region than the instance where LifeKeeper is running. However, in environments without internet connectivity, accessing S3 in another region requires specifying an endpoint. NFS or EFS can be set up in the same region, but placing it in a different AZ (Availability Zone) from the instance (one zone storage class for EFS) will have an ability to handle AZ failures.
Post your comment on this topic.