… and that’s why I like Storage Spaces Direct
I’d like to just leave it at that “…. and that’s why I like Storage Spaces Direct”. It’s such a positive statement. I’m concerned that once I publicly state this, then my Storage Spaces Direct experience will take a dramatic downturn and a future post will be titled “Why Storage Spaces Direct….. WHY!”
So firstly what is Storage Spaces Direct – if you don’t know then stop reading right now and go here https://docs.microsoft.com/en-us/windows-server/storage/storage-spaces/storage-spaces-direct-overview
OK – you’re back.
So in 2016 I built from some O-L-D hardware a Storage Spaces Direct (let’s call it S2D to save us some time) Hyper-V cluster on Windows Server 2016 technical preview. I ran a few VM’s on the cluster, did some fail tests and thought ‘this is pretty cool’. At the time I didn’t know there was a ‘time bomb’ bug and I came back to my cluster after a few months to find the hosts off – broken. I mean really broken – I tried for a bit then though ‘whatever. It’s a test right?’.
By this stage Windows Server 2016 release had come out so I installed the OS on the host drives, formed the cluster and would you believe it – the storage from S2D was there, added it to the cluster, imported my VM”s and voila – back up and running.
Worked well for a while and my hardware started to die. One of the host OS disks (in a mirror had been marked as failed and would you believe it? No spares. Oh well, it’s a test right? 3 months later after coming back from leave the host is off and wouldn’t you know it – the other disk had died. So now I have one of my hosts with no OS.
Had a little bit of trouble at this point because the remaining node didn’t have the majority vote of my two node cluster (Maybe there is some credence to an independent disk witness in a two node cluster?) but once I’d done some vote changing and node eviction, there was my single node S2D cluster still running – albeit complaining about missing disks in the other half of it’s S2D storage pool.
By this time I’d found a spare disk and decided to install the OS onto it as a single disk, no redundancy because it’s a test – right? Installed the OS, roles, joined it to the cluster and hey presto – a redundant S2D (Hyper-Converged – may have forgotten to mention that) Hyper-V cluster up and running again with my VM’s ticking along nicely.
The commitment in the coding to ensure the S2D configuration is retained independently of the Operating System is nothing short of amazing and on some very old hardware too! Without really trying I’ve put S2D through the paces of some really show stopping hardware failures on systems that have no physically shared storage fabric and it’s continued along just nicely.
While there isn’t a fantastic managed layer (seriously have a look at Nutanix Prism), this stuff is cool and pretty damn unbreakable.