Is RDM Still a nono?

[francomartin33]

Hey im looking for alternatives to building an ESXi 2 cluster with one freenas on each server to handle storage. The thing is than I actually have only one hba card to passthru to one of the VMs so Im evaluating using RDM for the other vm. I read that cyberjock had problems a couple of years ago with data being suddenly gone hence the thread title.
Im going to be replicating the pools with each other for "HA" and I have daily backups in another both.
Im open to suggestions.

 

[kdragon75]

building an ESXi 2 cluster with one freenas on each server to handle storage.

Why are you planning to use two FreeNAS VMs? This adds complexity in addition to trying to use RDMs on RAID backed storage. If you need to have the VMs cross replicate, BUY the correct HBA. If you dont, you will likely regret it.

Side note, If using RDMs to pass individual full disks from an HBA, it is strongly recommended to use physical mode RMDs as this will pass the SCSI commands directly to the drive. With that said, I would not try this with SATA drives as I don't know how the commands would be translated.

 

[kdragon75]

To directly answer you topic question in the context of your post, Yes, RDMs are a nono on a RAID controller.

 

[francomartin33]

To directly answer you topic question in the context of your post, Yes, RDMs are a nono on a RAID controller.

What do you mean on a raid controller? I want to do a passthru of the invididual drives to the vm.

 

[HoneyBadger]

TL;DR - Yes, local RDM is still a no-no due to potentially missing or filtered SCSI commands by the ESXi stack.

Suggestion: Buy a second HBA, or re-evaluate why you want two FreeNAS VMs.

 

[francomartin33]

TL;DR - Yes, local RDM is still a no-no due to potentially missing or filtered SCSI commands by the ESXi stack.

Suggestion: Buy a second HBA, or re-evaluate why you want two FreeNAS VMs.

I want to have two freenas vms to eliminate the single point of failure i have with the current hardware.

 

[HoneyBadger]

I want to have two freenas vms to eliminate the single point of failure i have with the current hardware.

The physical hardware it's hosted on would still be a single point of failure in this configuration.

You will still have a window of downtime if your "active" FreeNAS VM fails as well - and your other system will only be as current as your replication frequency, unless you plan to configure HAST/CARP to attempt to make a fault-tolerant system, which is a different solution entirely (and probably better done on top of vanilla FreeBSD)

 

[francomartin33]

The physical hardware it's hosted on would still be a single point of failure in this configuration.

You will still have a window of downtime if your "active" FreeNAS VM fails as well - and your other system will only be as current as your replication frequency, unless you plan to configure HAST/CARP to attempt to make a fault-tolerant system, which is a different solution entirely (and probably better done on top of vanilla FreeBSD)

I know, but my aplications are designed for HA, behind a load balancer, if a host fails the other one can take over. Today if freenas fails, vms on both hosts have no storage... Thats why I had the idea to use freenas VMS

 

[HoneyBadger]

Ah, I understand now. I somehow thought you were trying to put two FreeNAS VMs on the same machine and that would give you HA.

For your use case, where HA is assured at the application level, the answer is now simply "buy a second HBA for the second physical host" and go for it.

 

[francomartin33]

Ah, I understand now. I somehow thought you were trying to put two FreeNAS VMs on the same machine and that would give you HA.

For your use case, where HA is assured at the application level, the answer is now simply "buy a second HBA for the second physical host" and go for it.

Awesome, Ill try that then.

Thank you