During a couple of recent client engagements, I realised there are still quite a few people using Azure Classic Resources. The simple reason for this is that they work and we all know in IT, if it ain’t broke…etc.
Now, if you only recently began working with Azure, you may never have even seen a Classic Resource. The change to Azure Resource Manager (ARM) started back in 2016. Classic Resources do have some similar objects now that they are fully in the “new” portal (yes there was an old portal, it was…odd) but the concepts are different. This is because the focus was on services, hence it being referred to as Azure Service Management (ASM). Now, Azure is focused on resources, therefore Azure Resource Manager (ARM).
So if you discover, or own some classic resources and would like to move them to ARM, hopefully this blog post will help you make a start.
First, do you need to migrate away from ASM? So far, Microsoft has said no, they’re “not deprecating ASM in the near future”. However, if you note the use of “near” that could mean that as the Azure Roadmap progresses, ASM is eventually deemed redundant.
Updated 2020-02-28, Microsoft have announced deprecation of ASM VMs and dependent resources on March 1st, 2023.
The other reason for a migration may be that you want to avail of some of the features only supported in ARM or want to gain greater granular control of your individual resources. Whatever your reason, the approach is the same:
In this stage, you need to understand and most likely document all of the resources in scope for the migration. Once you have this, you can compare and contrast to options supported and changes required for migration to be successful. The following resources are supported for migration:
- Virtual Machines
- Availability Sets
- Cloud Services with Virtual Machines
- Storage Accounts
- Virtual Networks
- VPN Gateways
- Express Route Gateways (in the same subscription as Virtual Network only)
- Network Security Groups
- Route Tables
- Reserved IPs
The list for currently unsupported features is here.
While the migration operation for the most part takes place in the management plane, and there is therefore no downtime, you could proceed with a migration without impact to your users. However, I’d recommend either completing the migration out of business hours or planning a maintenance window to account for any unforeseen issues. I’d always rather have the window and not need it than the opposite!
It’s also important at this point to understand the order resources should be moved too. Microsoft have this flowchart, the lines are bit confusing but it’s accurate.
The most important point to take from the above is that if your VMs are all in the same vnet, well then you migrate the vnet, it will bring all dependent resources with it. Then your next step is any storage accounts.
Microsoft have provided a helpful validation operation within the portal and via Poweshell. This will give you a very quick look into your resources and their config to see if it’s compatible with a migration or not. However, it can not take into account all unsupported features, such as an issue on the ARM side, so ensure you complete your own checks!
This is a key phase. The migration process occurs on the control plane, the data plane is unaffected at all times. The preparation operation allows you to create the ARM metadata while leaving your current ASM resources in place. This is represented in the graphic below:
Once the operation begins, the management plane is locked for the duration. This means you cannot make any changes to your resources. However, your users are unaffected. With one exception, if your VM is not in a vnet ( I know, crazy in ARM!) it will be stopped and started during Prepare.
If Prepare is successful, you can continue with the migration or, abort and fix any issues. The operation creates a non-editable Resource Group for each Cloud Service, copying the name and appending “-Migrated”. Nice and neat, relatively low impact and risk. Exactly what is useful during a migration! Remember, the management plane is locked, so if all is OK complete the migration to unlock. If not, abort to unlock.
Once you’re good to go, the last step is to migrate. Again, this is only a control plane change. The operation is idempotent, so if it fails it can be retried. However, once complete you cannot rollback. Once complete, you will now be in the state represented below:
The Classic Resources are removed and only the ARM versions remain. As the models are different certain services and aspects are renamed or incorporated in a different way, a full list of these mappings is here and is very useful during the Prepare phase and for future use once migrated.
All the steps required to migrate all VMs in the same vnet are documented and I’d recommend reviewing all steps before running any.
If you take your time and follow the recommended methods, you should be able to make your move from ASM to ARM pain free.
As always, if there are any questions comment or get in touch on Twitter!