Stretched cluster and Arbiter scenario

[J-Philippe]

Hi, regarding that (old) KB from vmware : https://kb.vmware.com/s/article/51462 (availability zone failure)

What will happen if site 1 and 2 are isolated each other but not from site 3 that host the arbiter.

HA will fail and VM remains power on on each side untill the communication get back between site 1 and 2 ?

Regards,

Jean-Philippe

[AnkiN]

Hi Jean,

Thanks for using HPE SimpliVity Forum.
In HPE SimpliVity stretched cluster environment The arbiter should reside outside of the hosts which make up your stretched cluster....i.e: do not run your arbiter VM on any host which is part of the stretch.

Hope this link would help you in gathering other information.

Regards,
Ankit

[J-Philippe]

Hello yes of course, my question is about have site 1 isolated from site 2 but neither from site 3 where the arbiter reside.

So site 1 can reach site 3 with arbiter

site 2 can reach site 3 with arbiter

site 1 cannot reach site 2

Kind of host isolation, but I suppose that if host isolation response is set to disable in the HA config the VM remains power on.

[AnkiN]

Is your Site 3 part of the SimpliVity federation?

[J-Philippe]

No, site 3 is just hosting the arbiter on a stand alone machine, just taking the same sample than in the vmware KB but want to know what happen in the event when site 1 and 2 cannot reach other while they are able to reach the arbiter.

[AnkiN]

I hope the below scenario would help you to understand.

In the stretched cluster environment Any VM's that have their primary copy on Zone A(Site 1), will have their secondary data copy located in zone B, and any primary copies in Zone B (Site 2) will have their secondary data copies in zone A.

In the event of a catastrophic event that brings the entire Zone A down, then your Zone B will keep ownership of its original primary data copies, but also now take ownership of the secondary data copies of any VM's which were running on Zone A, ensuring business continuity. This all will happen if your site 3 having arbiter is accessible from Zone B(Site 2) during that event.

[J-Philippe]

Sorry I think I'm not clear because I fully understand this scenario, no pb with that topic when a site is down.

My concern is about network isolation.  

Host A in site 1 is up but cannot reach Host B in site 2 because of network failure (it can happen in real world)

Site 3 is hosting the arbiter on Stand Alone machine.

I guess but I would like to have confirmation that VM on Host A in site 1 will remain "Power On" and VM on Host B in site 2 will also remain "Power On". But there will be no split brain because the arbiter will be aware that Host A cannot reach Host B and when the link will be up between site 1 and site 2 the host A and B will be able to synchronize Primary and secondary data of the VM.

[Shivam1]

Hi Jean,

Host A Site 1 and Host B site 2 will be the part of same cluster but they will be part of two zones say zone1 and Zone2 respectively ,which makes them a part of stretch cluster configuration.Both the zones will have same arbiter.

So ,if Host A Site 1 (Zone 1) goes down ,the VM will power ON in (Zone2) i.e HostB Site2 ,arbiter will not cause split brain here ,since the "p" and "s" copy of VM(hives) will be distributed across zones and not within zones.

Let me know if it clarifies !!

Regards

Shivam

[ICPMuenchen · Accepted solution]

---

@J-Philippe wrote:

Hello yes of course, my question is about have site 1 isolated from site 2 but neither from site 3 where the arbiter reside.

So site 1 can reach site 3 with arbiter

site 2 can reach site 3 with arbiter

site 1 cannot reach site 2

Kind of host isolation, but I suppose that if host isolation response is set to disable in the HA config the VM remains power on.

---

In my opinion nothing will/should happen, if you havnt configured anything regarding host isolation etc. in vcenter HA directly.

The arbiter reacts as witness, and because he still can see/reach both hosts in site 1 and 2, everything is fine for the arbiter. The VMs should reside on its place and he shouldnt initiate a failover or anything like that.

Thats my understanding about the arbiter service.

[J-Philippe]

That's also what I believe but I would like to have feedback from HPE. I will test it. Thank You