Optimise SVM-DR without stretched layer 2 Network | NetApp | Page 1

tidal stratus Jan 16, 2025, 10:32 AM

#

Hi All,
Just wondering if we could optimise our DR. At the moment we run Site A and Site B but there is no stretch layer 2 network. The problem that this brings is NFS clients needed to unmount before a site/SVM DR failover. We have our SVMS setup with a single IP and a CNAME that we switch the DR. Is there any clever way to avoid this?

noble dock Jan 17, 2025, 5:12 AM

#

Can you provide more details? What are your NFS clients? VMware?

pallid phoenix Jan 17, 2025, 8:57 AM

#

If you don't have layer2 connectivity between sites, using BGP can help, it announces single /32 routes of all VIPs independent of where they are. But it would require you to set up BGP across both sites in your environment if you are not already using it

tidal stratus Jan 17, 2025, 2:29 PM

#

Bog standard unix Rhel clients, some NFSv4 some NFSv3 no VMware

#

Woudl that still work with NFS and give me a single cname? e/g nfs.filer?

noble dock Jan 17, 2025, 3:57 PM

#

The problem is a DNS is a construct used to return an IP and that IP is what is used by the system for its NFS mounts. Once a mount is made to an IP, it doesn't matter what you change the CNAME to, this is why you have to remount after the change to get the new IP. There are two options to this: stretch Layer-2, or reroute Layer-3 traffic using dynamic routing, i.e. BGP.

pallid phoenix Jan 17, 2025, 5:04 PM

#

tidal stratus Woudl that still work with NFS and give me a single cname? e/g nfs.filer?

with BGP the IP would not change, so everything would just continue to work

unkempt jungle Jan 18, 2025, 5:31 PM

#

Do your NFS clients change IP during DR ? We have similar design and our NFS clients failover to DR with new IP. Our problem is the DNS cache on SVMs don't flush quick enough and NFS clients have issue access nfs mounts until caches are removed.

tidal stratus Jan 18, 2025, 8:55 PM

#

Found that because its mounted via a cname and the underlying IP changes they hang and need remounted

pallid phoenix Jan 18, 2025, 9:01 PM

#

yeah this is standard behavior with DNS

#

you didn't specify the type of DR, so I'm assuming something asynchronous like SVM DR or SnapMirror... in that case you want to reboot your clients anyway because otherwise they would continue to work on old data and that is a recipe for data corruption

tidal stratus Jan 18, 2025, 9:21 PM

#

Sorry, We use a CNAME to point to the prod SVM or DR SVM in a failover. We are using SVM-DR underlying

pallid phoenix Jan 18, 2025, 9:22 PM

#

yeah, then your DR volumes are always behind, and you should never switch to them without rebooting the clients (or at least unmounting and remounting but a restart is usually easier)

tidal stratus Jan 18, 2025, 9:23 PM

#

How would the BGP option work? If I was starting again from fresh lets say?

pallid phoenix Jan 18, 2025, 9:25 PM

#

even with BGP you would still have to reboot your clients in a DR scenario.
As for how to setup BGP, you basically configure BGP between your datacenter switches (primary and DR) and peer both NetApps to the corresponding AS. In a DR scenario the other NetApp would take the IP address and announce it in the other AS which the BGP routers pick up and distribute across your network

#

and for BGP to be of any use, you would need to use the same IP address on both the primary and the DR SVMs

tidal stratus Jan 20, 2025, 10:14 AM

#

So we would need the SVMs etc to use the same IPs? Would that be possible without the L2 stretch network?

#Optimise SVM-DR without stretched layer 2 Network