Windows Azure guarantees a 99.95 % uptime SLA, but this is only for multiple instance roles. So, for example, one web server by itself will not be guaranteed 99.95 % uptime, but will rather have “best effort”. Whereas two, or more, in a multi instance role will. So how do you make something multi-instance?
You use something called an availability set. An availability set is spanned across multiple fault and update domains in the datacenter. A fault domain is a collection of resources that are expected to fail together, like a server rack. An update domain is a set of resources that are updated at the same time. An availability set ensures that your instances are spanned across multiple update and fault domains so that no single failure or update can impact all the instances. The point here is that all the instances must be the same and be able to offer the same service to your users. So going back to our example with web servers you have to have at least two identical ones in the same availability set for your service to have a 99.95 % uptime guarantee. It is not correct to say that any VM or instance is guaranteed 99.95 % since it is the service that the instance offers that is guaranteed. Furthermore only Internet facing services have the SLA, since you need to use the Windows Azure load balancer to publish your instances, but more on that another day. Back to availability sets.
You add VMs to availability sets either in the Windows Azure portal, with PowerShell or the REST API. In the portal or PowerShell there is no way to explicitly create an availability set, you always create one as a side effect of configuring a VM. In the portal it looks like this:
This is an availability set with only one instance. In the drop down list I can either remove the VM from the availability set or create a new availability set and add the VM to that set.
Note: when you either add or remove a VM from an availability set, it will get a new NIC. This will have the same IP address as before but you will see that its number has been incremented.
In PowerShell you use the Set-AzureAvailabilitySet or New-AzureVMConfig cmdlets. Set-AzureAvailabilitySet updates an existing VM while New-AzureVMConfig lets you specify an availability set when you create the VM.
Something I miss is the ability to list all my availability sets. Since there is no (yet) Get-AzureAvailabilitySet cmdlet you have to extract this information from each of your VMs. Here is a quick one-liner to do that:
https://gist.github.com/morgansimonsen/8039845
This will list the VM names, cloud service names and availability set names (if any) for all your VMs. Feel free to add any kind of formatting or sorting to the output.