Each Virtual Machine Scale Set has an upgrade policy that determines how virtual machines are brought up-to-date with the latest scale set model. The upgrade policy you choose can impact the overall service uptime of your Virtual Machine Scale Set.
Automatic upgrade policy
With an automatic upgrade policy, the scale set makes no guarantees about the order of virtual machines being brought down. The scale set might take down all virtual machines at the same time to perform upgrades.
Automatic upgrade policy is best suited for DevTest scenarios where you aren’t concerned about the uptime of your instances while making changes to configurations and settings.
Manual upgrade policy
With a manual upgrade policy, you choose when to initiate an update to the scale set instances. Nothing happens automatically to the existing virtual machines when changes occur to the scale set model. New instances added to the scale set use the most update-to-date model available.
Manual upgrade policy is best suited for workloads where the instances in the scale set are composed of different configurations and each configuration might require different updates and changes.
Rolling upgrade policy
With a rolling upgrade policy, the scale set performs updates in batches with an optional pause time in between. You also get more control over the upgrades with settings like batch size, max healthy percentage, prioritizing unhealthy instances and enabling upgrades across multiple availability zones at the same time. Additionally, when using a rolling upgrade policy, you can enable max surge which will replace VMs running the old model with new VMs using the latest model.
Rolling upgrade policy is best suited for production workloads that require a set number of instances always be available. Rolling upgrades is safest way to upgrade instances to the latest model without compromising availability and uptime.
To learn more about upgrade policies, see upgrade policies for Virtual Machine Scale Sets
