Cluster node VM size options

Question

There appears to be some churn in what VM sizes are available for nodes in a SF cluster. Not long after SF went GA I created a cluster using a mix of A0 and A1 nodes. I was ecstatic at the time to see this was supported as it's awesome for dev/qa scenarios.

Today I went to create a new cluster and find my options for VM size are severely limited. D1v2, D2v2 or D3v2 for Bronze durability and D15v2 for Gold. Hugely disappointing to say the least. And a significant backpedal from just a few weeks ago.

What is the backstory here?

1. Was my original cluster configuration never supposed to be allowed and was a bug in the Portal?
2. Were there problems seen with these sizes and the SF team decided they are unusable?
3. Something else entirely?
4. And is this a permanent decision?

I'd really like to see as many VM size options as possible be supported.

Answer 1

You're looking at the recommended list of VM sizes. You can still use any VM size you want, including A0 and A1, just click the "View All" button.

We generally recommend VMs with SSDs for stateful services so that your applications aren't bottlenecked on old spinning disks.

Answer 2

The list of recommended SKUs were rolled out in response to customer feedback. All the VM options are still available under "View all" button. The intent was to make sure that the customers choose the recommended VMs with SSDs (with enough SSD space), unless they were specifically looking for a particular SKU. This was done mainly in response to a good number of customers wrongly choosing the DS SKUs when they really were looking for D series VMs. (Choosing of the DS SKUs resulted in the VMs quickly running out of disk space).

Although I realize that A0 SKU is very attractive in terms of price, and may be ideal for a test cluster, for a production Cluster it is strongly recommended that you do not choose A0 as the SKU for the primary node type. The primary node type is where majority of the system services live. For more considerations on cluster capacity planning see - https://azure.microsoft.com/en-us/documentation/articles/service-fabric-cluster-capacity/