Hyperflex stretched Cluster

[mumbai.support]

Hello,
We are implementing the hyperflex stretched Cluster of HX240c M6 All Flash Node,5 hx-servers at DC , 5 hx-servers at DR & 2 edge server for witness VM.
now our DC site & witness site is ready to configure HX server, But DR site is not ready.
so can we configure stretched Cluster on DC site without DR so we can add the DR site later in stretched cluster after site is ready?

[RedNectar]

Hi @mumbai.support ,

There are four main stages of a stretched Cluster Install, with distinct breaks between them. You can stop and restart after step 1 or 2 or 3.

1. The UCS Manager competent of Site A
2. The UCS Manager competent of Site B
3. The ESXi and HyperFlex Cluster install and config
4. "Post-install" steps

The "post-install" steps include setting up the witness, configuring vMotion, and configuring site affinity and a few other bits.

But you can certainly start the configuration for your DC - and I suggest this is a GOOD IDEA - any UCS problems you might see will probably get sorted then.

BUT YOU WILL NOT BE ABLE TO USE the DC site until the DR site is built.

---

One other thing about your query that bothers me.

"2 edge server for witness VM."

The witness VM is a VM - you can certainly run it on a HyperFlex Edge cluster at a remote site, but it does NOT need a whole cluster to itself.  That would be like buying two luxury apartments to house one cat.  Now if you fill the apartments with paying guests, (or fill the Edge cluster with more VMs) then that's OK.

The Witness VM is only ever needed if the connectivity between the two clusters fails, so it IS important, but any reliable site, such as your favourite cloud provider, would probably be more economical.

[mumbai.support]

Hello,

Thanks for your replay.

Now we are planning , we will keep all the 10 hx servers at DC site temporary including DR servers . then we will configure the stretched Cluster at DC. if site DR is ready then we will move the 5 Hx server to DR site.

is it possible by moving the server to DR in stretched Cluster?

if yes, then how many hx servers we can move at same time to DR?

can we do power off all 5 hx servers to move DR site or we have to move one by one?

what should be replication factor for 10 node cluster in stretched Cluster?

[Kirk J]

Greetings.

You do not want to move the witness into the prod or DR cluster.  This will result in the whole cluster going down if just a single site (housing the witness) goes down.

Please keep in mind the witness IP address cannot be changed after deployment, so make sure your ip assigned to the witness, is supported at the location you intend to permanently house the witness VM.

I suspect in future version of HX we will see support for ability to change witness IP.

Kirk...

[Steven Tardy]

You are asking a lot of questions and trying to do things in an order which are untested so this is your fair warning to (borrowing a quote from CBS Big Brother) "expect the unexpected".

HX stretch cluster is meant to be DC<>DC, not really DC<>DR. IMO two clusters with HX replication would make more sense for DC<>DR. I'm sure Sales and Engineering will say "Yes, DC<>DR is okay", but leaves me (who has to fix any goof ups when they goof) many questions like: is your DR site as hardened/maintained as your DC site?

Moving the servers of an entire "site" for a stretch cluster is a thing and there is a documented process for "shutting down a side of a stretch cluster". It would make most sense to move all the DR servers from their temporary DC home to the permanent DR home. But IMO may make more sense to wait for the DR site to be fully functional.

Stretch cluster will always be replication factor (RF) 4 (really RF2 + RF2 if you want to get technical).

[RedNectar]

Hi @mumbai.support ,

With such expert advice from guys that know way better than me, (@Kirk J  and @Steven Tardy) I probably can't add anything of value. But here goes:

---

Now we are planning , we will keep all the 10 hx servers at DC site temporary including DR servers . then we will configure the stretched Cluster at DC. if site DR is ready then we will move the 5 Hx server to DR site.

Potentially could work, See @Steven Tardy 's answer.

is it possible by moving the server to DR in stretched Cluster?

 https://www.youtube.com/watch?v=nQq0gsW-vXw

Technically it probably COULD be done. But when it breaks Cisco don't want to have to help you put it back together.

if yes, then how many hx servers we can move at same time to DR?

I've started 3 attempts to answer this. Each time I keep thinking "that is such a bad idea" I didn't complete it. If this is a test environment and you don't mind loosing all your data, I could give you some tips, but they all involve completely wiping your system at some stage, although there is apparently a documented process.

can we do power off all 5 hx servers to move DR site or we have to move one by one?

Just don't do it!

what should be replication factor for 10 node cluster in stretched Cluster?

---

That's easy to answer, there are always 4 copies of your data kept, two at each site.

A better way

Have you considered using Replication instead? You'd have to do some talking to Cisco to get the licences changed over, but if you took this approach you could build your DC using 5 HX nodes, and when the DR site is ready, build another cluster using the other 5 nodes.  The advantage of doing it this way is that don't have to worry about latency lag between sites, and each site is independent.

You possibly could even build your DR site at the DC, configure and test the Replication, and when the DR site is ready, carefully shut down the DR Cluster (with assistance from TAC) and move it.  @Steven Tardy might have a better idea if this is possible.