iscsi failover with switchless setup and 1 iscsi target

[nickdd]

Hi,

I got a dual port 10G nic reserved for my storage network in my esxi (vmnic2 and vmnic3).

=> vmnic2 is connected directly to SAN1

=> vmnic3 is connected directly to SAN2

The two SANs work in active-standby mode and share a virtual IP, the active SAN listens to this ip.

I can't use nic teaming and can only assign 1 nic to each vmkernel port (to be used for the iscsi port binding).

So I've set up 2 vmkernel ports, one for each nic.

The problem is now that when I scan the HBA I always only get 1 path.

After a failover to the other SAN my connection to the datastore is lost.  I have to trigger a rescan of the HBA and then it finds the other path.

Can someone point out how I should set things up ?

Kind regards,

Nick

[chriswahl]

It sounds like the issue is the network topology. When you scan the HBA, the iSCSI initiator logs into the storage array's target and sets up the paths. If you have only one uplink going to the active side of the storage array, then iSCSI will only create one path on that particular vmkernel port. The other vmkernel port will not detect any targets to create a path. After you fail over the array, that original path on the first vmkernel port is no longer available because the storage is now using the other vmnic which has a different vmkernel port attached.

One solution is to put physical switches in between the storage array and the vSphere host, and mesh the topology so that each vmnic has a path to both sides of the storage. When the vip moves, the host will be (mostly) unaware of this as the switch will simply update its CAM table with the new location that is advertised by the array.

[nickdd]

Thanks, unfortunately physical 10GE switches are too costly for me at the moment.

That's why I wanted to work with direct connections. Saw some blog post that this was possible, but it doesn't explain the setup in detail.

It would be cool if I could turn a vswitch with 2 physical nics into a real switch, but I don't believe that's possible.

[chriswahl]

Can you present storage to both sides of the array instead of using a vip? This should work for your topology. You can then mark the paths to the active side as preferred or try a MRU path policy.

[nickdd]

Hi,

I was able to make it work as I just discovered that I don't need to use the iscsi port binding.

Since I'm using a single 10 gbit connection to each SAN, I don't need MPIO and therefore I could leave nic teaming active with both nics in active-active.

Enabling beacon probing just did the trick.

This way the virtual ip is always available.

Thanks for the support !

Nick

[chriswahl]

Beacon probing requires 3 uplinks to be of any use.

KB: What is beacon probing?

[nickdd]

Thanks for bringing this to my attention.

The setup seems to be working now, but I'll need to do some hands-on testing in the datacenter to see what disconnecting a cable will actually do.

So now it has been set up without iscsi port binding and just one active path.