Data Repository cluster configuration in different network sites

Document ID : KB000096530
Last Modified Date : 08/06/2018
Show Technical Document Details
Introduction:
For continuity and recovery purposes can the nodes in a Vertica Data Repository cluster reside in different network sites?

For example with a 5 node cluster sample 3 would be in one site, 2 in another site.
Question:
Is this a supported configuration?
Environment:
All supported CA Performance Management releases
Answer:
This is not unsupported. It is also highly recommended that a configuration such as this is NOT used.

A primary reason this is not recommended how to do with how Vertica K-Safety configurations work. Out implementation uses a K-Safety value of 1. K-Safety of 1 in a Vertica Data Repository means if more than 1 node is down the cluster shuts down. If a site becomes inaccessible for any reason 1 or more of 5 nodes will be down and the other 2 will shut down, bringing the entire database down.

Vertica work across networks, but if this is put into place even with recommended resources in place don't expect high performance. Vertica highly recommends them being located together and in the same network. 

It is also recommended not to locate the DA and DR systems remote from each other.

From operational experience we found that at one site they move off one of the nodes. It was ALWAYS down and unavailable. It was usually up for an hour at most before some network issue was observed causing it to once more go down. This was for a bank where they really looked after their network. As such the database was always trying to restore one of the nodes, never really succeeding. 

Vertica is very sensitive to having a node disappear off of the system during even minor network hiccups that other systems might be less sensitive to the interruption.

The recommended solution is to utilize the Fail-Over and DR setup available under CA Performance Management r3.5 if multiple data center configurations are required. This is documented here:

https://docops.ca.com/ca-performance-management/3-5/en/administrating/fault-tolerance