Datastore Redundancy Question

[MySongRanHills]

I cant seem to google the correct terms for I would like to accomplish. Can someone point me in the right direction on setting up redundant datastores for a Vmware HA Cluster?

Say I have a NAS that is providing VM storage via iSCSI for 2 hosts in a HA cluster. What would I need to do if I'd like to add another datastore on a different NAS so that if I rebooted the original NAS, VMs would be able to continue to run from the 2nd NAS without interruption?

edit: I think I'm looking to form a datastore cluster, but how does it work - is it like RAID 1, where all data is updated to both datastores in sync?

 

[Child of Wonder]

You're looking for VPLEX or something along those lines that makes two replicated datastores appear as one. vSphere 6 now has FT that would replicate the data, too, on a VM by VM basis.

 

[MySongRanHills]

Vmware doesn't have anything like this natively in 5.5? Seems like an awful lot of trouble to provide redundancy at every level and then have a single point of failure by using 1 datastore.

 

[NetJunkie]

Vmware doesn't have anything like this natively in 5.5? Seems like an awful lot of trouble to provide redundancy at every level and then have a single point of failure by using 1 datastore.

You need to define "without interruption". It's easy to do array replication and then boot a VM off that second copy if the first one dies. It's a lot more work to have the backend storage failover to a second copy on the fly without any interruption. That's where things like EMC VPLEX come in.

Or...you could do something like Neverfail or other application based replication and failover. But honestly, a datastore isn't a SPoF. It's usually on redundant disk behind redundant controllers connected via redundant fabrics. Then you back that data up in case you somehow kill the entire array frame.

 

[Riccochet]

AFAIK replication happens at the datastore level. At least that's how we are doing it with HP and Equalogic units.

 

[peanuthead]

I usually see something like that done at the SAN level.

 

[lopoetve]

AFAIK replication happens at the datastore level. At least that's how we are doing it with HP and Equalogic units.

Or VM level, depending on hardware solution.

Vmware doesn't have anything like this natively in 5.5? Seems like an awful lot of trouble to provide redundancy at every level and then have a single point of failure by using 1 datastore.

It's actually very difficult to do. That being said, in most enterprises, if the array has gone down entirely, many many things have failed (not just one)- multiple controllers, multiple power supplies, raid, etc...

And even if you do mirror writes - how do you catch up if one goes down for a while? What's that do to performance? How do you throttle that? Is guest IO more important than redundancy IO? what if you never catch up? What latency penalties can you stand?

Enterprise solutions for this are very expensive and very complex for a reason

 

[MySongRanHills]

You need to define "without interruption". It's easy to do array replication and then boot a VM off that second copy if the first one dies. It's a lot more work to have the backend storage failover to a second copy on the fly without any interruption. That's where things like EMC VPLEX come in.

Or...you could do something like Neverfail or other application based replication and failover. But honestly, a datastore isn't a SPoF. It's usually on redundant disk behind redundant controllers connected via redundant fabrics. Then you back that data up in case you somehow kill the entire array frame.

It's actually very difficult to do. That being said, in most enterprises, if the array has gone down entirely, many many things have failed (not just one)- multiple controllers, multiple power supplies, raid, etc...

And even if you do mirror writes - how do you catch up if one goes down for a while? What's that do to performance? How do you throttle that? Is guest IO more important than redundancy IO? what if you never catch up? What latency penalties can you stand?

Enterprise solutions for this are very expensive and very complex for a reason

Well fuck. I really didn't think what i wanted to do would be all that hard/rare. You both make good points though.

I was looking for true FT not a HA type solution where a VM could be restarted if original datastore fails.

I was just hoping it would be as easy as copying VM to a 2nd datastore and configuring vSphere to use both , or possibly configure something in SDRS to switch datastores based on availability.

 

[Child of Wonder]

Well fuck. I really didn't think what i wanted to do would be all that hard/rare. You make good points though.

I was looking for true FT not a HA type solution where a VM could be restarted if original datastore fails.

I was just hoping it would be as easy as copying VM to a 2nd datastore and configuring vSphere to use both , or possibly configure something in SDRS to switch datastores based on availability.

Then you want FT in vSphere 6 Enterprise Plus but that's VM based replication and you can only replicate 8 per host IIRC.

 

[lopoetve]

Then you want FT in vSphere 6 Enterprise Plus but that's VM based replication and you can only replicate 8 per host IIRC.

And can have some NASTY penalties as well, and some very tight requirements for performance (IIRC)

 

[MySongRanHills]

Thanks everyone, think I'll just do things manually. When the NAS1 needs a reboot I'll migrate all of its VMs to NAS2 first.

 

[Eulogy]

Couldn't SRM be leveraged for this? Of course, that'd require a 2nd vCenter, 2nd SRM, SSO and so on... but hey, I have no idea your ability, budget or anything .

 

[lopoetve]

Couldn't SRM be leveraged for this? Of course, that'd require a 2nd vCenter, 2nd SRM, SSO and so on... but hey, I have no idea your ability, budget or anything .

Sure, fail over to DR, do what you need at prod, fail back. Happens all the time.

But that's a LOTOMONEY for a home lab it sounds like.

 

[KapsZ28]

Cool. Didn't realize that FT vSphere 6.0 also had storage redundancy. Can you use it to protect the vCenter appliance? I still haven't looked into version 6.0 much. I used vCenter Heartbeat in the past, but wasn't overly impressed and ended up switching to VCSA. My only regret was losing linked mode.

 

[Vader]

My only regret was losing linked mode.

FYI..Linked Mode is supported in VCSA with vSphere 6....

 

[KapsZ28]

FYI..Linked Mode is supported in VCSA with vSphere 6....

Nice! Now I have a reason to upgrade!

 

[Olga-SAN]

you have to have some sort of a storage virtualization software that is going to take both of your nas boxes and replicate them between each other hiding housekeeping process from you and esxi

datacore does this for sure but is going to cost like golden gate bridge + taxes

starwind does this for sure and for free but for local disks and you need to check do they support nas/san as a source

cheap nas units start to support replication but active-passive only and in a very ugly way like synology does

enterprise storage does as a standard or as a paid option

I cant seem to google the correct terms for I would like to accomplish. Can someone point me in the right direction on setting up redundant datastores for a Vmware HA Cluster?

Say I have a NAS that is providing VM storage via iSCSI for 2 hosts in a HA cluster. What would I need to do if I'd like to add another datastore on a different NAS so that if I rebooted the original NAS, VMs would be able to continue to run from the 2nd NAS without interruption?

edit: I think I'm looking to form a datastore cluster, but how does it work - is it like RAID 1, where all data is updated to both datastores in sync?

 

[Deleted member 16178]

Someone should have asked earlier in the thread, but why would you want to do this?

Generally, enterprises will have a SAN with redundant components and replication (either synchronous or async) to a DR site. Nonetheless, failover is usually difficult to automate.

Another solution is ensuring that your application is HA. You can use a load-balancer in front of your redundant web/application layer, and multiple databases running in replica/mirror mode (like MSSQL AlwaysOn). You can place half of these components on different storage arrays or vendors.

 

[lopoetve]

Someone should have asked earlier in the thread, but why would you want to do this?

Generally, enterprises will have a SAN with redundant components and replication (either synchronous or async) to a DR site. Nonetheless, failover is usually difficult to automate.

Another solution is ensuring that your application is HA. You can use a load-balancer in front of your redundant web/application layer, and multiple databases running in replica/mirror mode (like MSSQL AlwaysOn). You can place half of these components on different storage arrays or vendors.

Because done properly, assuming you have the budget, your recovery time is literally measured in minutes. Stretched clusters in VMware, for instance, rely on VMware HA for the failover, which means you're starting to boot in 30 seconds after failure.