Two-node esxi cluster with HA storage

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

voodooFX

Active Member
Jan 26, 2014
247
52
28
I have this idea bouncing in my mind for too long and now I really want to try something.
In short words I want my two-node esxi cluster to be completely redundant, and when I say "completely" of course I mean "storage" :D

My idea is to passthrough a Mellanox ConnectX2 IB card, RAID Controller and some drives to a VM on each host, create a storage cluster and then use a third VM (in fault tolerance?) to export the storage to the esxi hosts.

My main doubts are three
1. What I can use on the storage hosts to export the storage? iSCSI (LiO)? GLuster? Some storage appliances?
2. Directly related to the previous point: what I can use on the "export VM" to "assemble" the two "volumes" I receive from the storage hosts?
3. Will the fault-tolerance-VM work for this kind of use?

I don't want VMware vSAN and StarWind vSAN.
The first because the price and the second because of Windows :D
 

mrkrad

Well-Known Member
Oct 13, 2012
1,244
52
48
HP Lefthand will do this! Create a cluster out of two hosts local storage like VSAN,Starwind san, not many free products can do HA storage.

Yes FT/HA will work with this!
 
  • Like
Reactions: NeverDie

TeeJayHoward

Active Member
Feb 12, 2013
376
112
43
My idea is to passthrough a Mellanox ConnectX2 IB card, RAID Controller and some drives to a VM on each host, create a storage cluster and then use a third VM (in fault tolerance?) to export the storage to the esxi hosts.
This is how I'm set up. Well, minus the IB. The problem is, AFAIK, you can't have DirectPath on a FT VM. So you end up with an ESXi host that's a single point of failure. vSAN (or similar) is a superior solution.
 

voodooFX

Active Member
Jan 26, 2014
247
52
28
HP Lefthand will do this! Create a cluster out of two hosts local storage like VSAN,Starwind san, not many free products can do HA storage.
Yes FT/HA will work with this!
Thank you, I will take a look at this "HP Lefthand" but I will prefer something more custom and indipendent

This is how I'm set up. Well, minus the IB. The problem is, AFAIK, you can't have DirectPath on a FT VM. So you end up with an ESXi host that's a single point of failure. vSAN (or similar) is a superior solution.
The VM in FT mode will be the one used for exporting the storage to esxi and this one does not need any direct path hw, just needs to recover asap if the node where is located became unavailable.
The storage VMs, with all the stuff in passthrough, will not be HA-protected, of course.
 

voodooFX

Active Member
Jan 26, 2014
247
52
28
Yesterday I did some tests
First of all, in the original idea there was a design error: the third VM, the one supposed to be protected by HA+FT and exporting the FS for esxi needs to run on a shared storage, which is also what it should provide, so how do you run on something you have to provide? :D

removed this VM from the design, I created two VM, one on each esxi-node (on local, single 500GB WD RE4) and installed glusterfs.
The install process was easy and worked pretty smoothly and in 10min I was able to mount the gfs volume via NFS.
After that I installed corosync, pacemaker and pcs, I created a cluster and a virtual IP and I'm now able to mount the NFS vol through the vIP but if the first node goes down the share become unavailable.
I'm now to pacemaker so I'm not sure if I have to specify somewhere I want to monitor the NFS service?
Performance was also terrible (~10MB/s seq.) but I think it's normal since I was using a single 1G connection for everything and two single 7200rpm drives...
 

dswartz

Active Member
Jul 14, 2011
610
79
28
You don't need to use a clustered FS here. I am using ZFS. I have a resource for the zfs pool (which I got from an inactive HA solution someone did using OmniOS). Also a resource for a virtual IP. The latter depends on the former, so if the active node fails, pacemaker will forcibly import the pool on the second node and then active the virtual IP, making it visible via NFS. Works fine. I do actually have 3 nodes - the primary storage node is running centos7/ZFS bare-metal, but the backup storage node is a centos7/ZFS on the backup HA ESXi node (I did this so I could use all of the 32MB on the primary storage node for ARC). I shouldn't think you'd need to use ZFS here - pacemaker *should* work with any supported non-clustered FS.
 
  • Like
Reactions: Patrick

JonathanS

New Member
Mar 24, 2015
1
0
1
34
If you’re not happy with Windows because of licensing fees you can easily use free Hyper-V as a core OS to run StarWind on top of it ether bare metal or inside Virtual Machine. Works like a charm! Agree with mrkrad – there are not so many free products doing this.