Non-attachable datastore options for AppStacks

[VirtuallyMikeB]

Based on what I've read so far, it seems if I want my AppStacks to be available at multiple sites, I need to actually mount the non-attachable datastore that lives at, say, Site 1, to at least one host in Site 2, assuming I'm using Storage Groups. Is that right?

If that's the case, and if primary storage at each site is not able to be shared across sites (maybe it's only fibre channel within a data center), then maybe one option is an Ubuntu NFS datastore. Is that right?

[Ray_handels]

I'm trying to get my head around what you are stating there.

Long story short is that if you do want to attach an appstack or writable the VDI machines should be able to reach the datastore the appstack resides on.

If you look at Appvolumes it detects all datastores available if you connect to a vSphere server. So if you do want to attach appstacks on a different datastore then you either need to add an extra VSphere server or add the datastore to your current vsphere.

With the creating of Storage Groups the benefit of this is that it will sync all appstacks between this storage group (if you do create the appstack within the storage group!)

[VirtuallyMikeB]

Thanks.  I have two sites. I plan on using clustered databases across sites for App Volumes. I'd like to make the same AppStacks available at each site via Storage Group replication. I'm trying to understand the idea of the non-attachable AppStacks datastore.  From what I've read, it looks like this datastore needs to be mounted at both sites. Is that correct?

If so, I'm in a situation where this isn't possible because the storage is FC within a data center, not across data centers. I'm sure there are a million other ways to accomplish a shared or swing datastore, but I'm leaning toward something simple like an Ubuntu NFS server VM.

[Ray_handels]

Maybe I'm misthinking this entirely but the moment you add a datastore to a storage groups it will actively sync appstacks to all of those data stores but also attach from all of those data stores, it will not make them non attachable and not use them, that's not how the storage group works.

What we do if we do want to make those non attachable is just copy them from an active datastore to a non active datastore. These are simple VMDK that you can copy around as much as you would like to. If you need to restore it is just copy the files back and import and your good to go.

[VirtuallyMikeB]

Since App Volumes 2.10, a new feature was added to Storage Grouos that allows you to identify an AppStacks datastore as non-attachable so the AppStacks on that datastore will not be attachable to VMs. The feature allows for multi-site replication by using this non-attachable datastore as a "swing" or intermediate datastore.

I'm just trying to validate that the non-attachable datastore needs to be mounted from both sites. I'm thinking it does.

[sjesse]

How it work if it wouldn't be mounted? The storage groups feature just replicates files between data stores, the data store needs to be there to be part of the storage group to replicate the data. The swing data store is the common one between both environments.

[sjesse]

don't take my response as sacarasm,  this is how I thought it out.

[VirtuallyMikeB]

Agreed. I've never worked with App Volumes before, so I'm just validating what I get from reading - just looking for affirmation.  I like the idea of Storage Groups and non-attachable datastores to allow for replication, but not the idea of mounting datastores across sites. But it's probably not so bad since we know it's nothing more than, essentially, file copies, and no systems will be trying to read or write from the non-attachable datastore.  I'd like to understand the data path and implementation of the file copy process, though.  Is it using NFC between ESXi hosts?

[Ray_handels]

I never even knew that option existed.. Oh well, we'll leanr every day.

What we saw happening in regards to datastores is that Appvolumes will import all datastores attached to ESX hosts in the vcenter you connect to. So if you for example have a local datastore on all ESX hosts it will show those stores as an option to create an appstack on.

Looking at that option that would mean that in every site you would at least need to have 1 ESX host that has that datastore connected to it, otherwise it cannot use that datastore to replicate between sites.

We just have 1 site and thus recplicate and attach between all datastores as it is also much better for performance reasons. This only starts to count when you have a lot of users attaching 1 specific appstack (in our case a Base appstack for few hundred users).

How files are being copied I have no idea to be honest. My guess is you would need to ask VMWare, but does it really matter how it is being copied for you? Maybe due to networking restraints? Or any other reasons?

When we need to copy files from production ot test we just open up the datastore browser and copy and paste the disks. They are being copied thin provisoned.

[VirtuallyMikeB]

does it really matter how it is being copied for you?

Yes, it does. I like to understand what's happening on the network and what traffic I can expect. For instance, does storage group replication happen over the ESXi management VMkernel by default? Is it configurable? If it uses NFC, can we use the NFC-specific VMkernel port or is that just for vSphere Replication?

All the best,

Mike

[sjesse]

It would be NFS, I don't think you could do a iscsi or a fiber channel datastore between two different vmware environments without any issues. I don't think you can bind nfs traffic to a vmkernel port, but if your nfs network was on its own network you could but a vmkernal adapter on that network and the traffic would be routed out the network adapter.

[VirtuallyMikeB]

Yeah, I'm thinking this is the best way given the constraints of the environment. In principle, I don't like the general idea of using an Linux box to host datastores, but it's understood this is really just to support replication. Thanks!

Edit: Oh, you're saying AppStack / writable volume storage group replication is storage traffic?

Edit Edit: Eh, NFS by default doesn't make sense. If NFS isn't the configured storage protocol, there's no way it'd use it. So in my case, if there are FC and NFS datastores in a storage group, how's the data copied and over what interfaces?