0

We have a uniform high availability configuration [somewhat like 1] with the difference we have two different subnets, vlan and vsphere for two different sites, X and Y. The host in site X has ip 10.192.168.5 and similar host in site Y has ip 10.150.6.16). The hosts are connected to an active-active storage where the luns are seen in the two different sites. A client will hit a url which will have a public IP then this will be mapped to the ip on site X 10.192.168.5. However, when Site X fails it does not redirect traffic to Site Y since we cannot get the same ip in Site Y which is in a different subnet. We trying to minimize routing i.e. when site X goes down then go to firewall and add rules to redirect traffic to Site Y.

  1. What is the best way to have HA on two different sites with different subnets where the client does not experience any downtime i.e. seamless from the client point of view?
  2. Whats the minimum changes that we can make to accommodate the above design?

[1] https://www.yellow-bricks.com/2012/11/13/vsphere-metro-storage-cluster-uniform-vs-non-uniform/

2 Answers 2

1

Very often, something like that is done with a virtual IP. The virtual IP gets mapped to one of the hosts (using NAT with routing, or ARP with a shared L2 segment) and remapped on failover.

If you're using public IP addresses from your ISP, any rerouting needs to happen at that ISP. Different ISPs require you to have your own public IP addressing (at least /24) and your ISPs to advertise those addresses.

A different approach is to use a DNS host name and remap the A record to different IP addresses. That requires a short TTL for the record (to avoid lengthy caching and a lenghty propagation delay). Still, DNS-based failover is a matter of minutes however, not seconds.

Of course, seamless failover requires continuous synchronization of the server states - host configurations including virtualization are explicitly off-topic here however.

2
  • To remap to the failover node, do we need to do that manually? Can it be automated? According to cisco.com/c/en/us/support/docs/availability/high-availability/… - "Both primary and secondary must be on the same subnet."
    – schien
    Apr 18 at 19:01
  • Yes, that can be automated - there's a health monitor that triggers the failover. That Cisco link describes a scheme with a virtual IP address failing over locally, so both primary and secondary need to be on the same subnet. There are many other schemes, including route-based ones where both nodes can be quite distant. Note that we cannot recommend any product or even resources as that's explicitly off-topic here (for attracting opinions, advocacy and spam).
    – Zac67
    Apr 18 at 19:49
3

There are a few options— which one you choose depends on many details. This is the kind of problem that people hire consultants for. You shouldn’t expect to get a solution for free from strangers on the Internet.

That said, one possible solution is to use load balancers that forward traffic to the active servers. The load balancers can perform NAT so the client always sees the same address.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.