A tool to bring your existing Azure Resources under the Management of Terraform, Basically, "Reverse Terraforming which is not as such "Infrastructure as Code" but like "Coding Your Existing Azure Infrastructure" Back to your Terraform Configuration File and update/imporing State file.
Azure Terrafy imports the resources that are supported by the Terraform AzureRM provider into the Terraform state, and generates the corresponding Terraform configuration. Both the Terraform state and configuration are expected to be consistent with the resources' remote state, i.e., terraform plan
shows no diff. The user then is able to use Terraform to manage these resources.
The Terraform configurations generated by aztfy
is not meant to be comprehensive. This means it doesn't guarantee the infrastruction can be reproduced via the generated configurations. For details, please refer to the limitation.
Precompiled binaries and Window MSI are available at Releases.
go install github.com/Azure/aztfy@latest
brew install aztfy
Supported versions:
- RHEL 8 (amd64, arm64)
- RHEL 9 (amd64, arm64)
-
Import the Microsoft repository key:
rpm --import https://packages.microsoft.com/keys/microsoft.asc
-
Add
packages-microsoft-com-prod
repository:ver=8 # or 9 dnf install -y https://packages.microsoft.com/config/rhel/${ver}/packages-microsoft-prod.rpm
-
Install:
dnf install aztfy
Supported versions:
- Ubuntu 20.04 (amd64, arm64)
- Ubuntu 22.04 (amd64, arm64)
-
Import the Microsoft repository key:
curl -sSL https://packages.microsoft.com/keys/microsoft.asc > /etc/apt/trusted.gpg.d/microsoft.asc
-
Add
packages-microsoft-com-prod
repository:ver=20.04 # or 22.04 apt-add-repository https://packages.microsoft.com/ubuntu/${ver}/prod
-
Install:
apt-get install aztfy
yay -S aztfy
There is no special precondtion needed for running aztfy
, except that you have access to Azure.
Although aztfy
depends on terraform
, it is not required to have terraform
pre-installed and configured in the PATH
before running aztfy
. aztfy
will ensure a terraform
in the following order:
- If there is already a
terraform
discovered in thePATH
whose version>= v0.12
, then use it - Otherwise, if there is already a
terraform
installed at theaztfy
cache directory, then use it - Otherwise, install the latest
terraform
from Hashicorp's release to theaztfy
cache directory
(The aztfy
cache directory is at: "<UserCacheDir>/aztfy")
Follow the authentication guide from the Terraform AzureRM provider to authenticate to Azure.
Then you can go ahead and run aztfy resource <resource id>
, aztfy resource-group <resource group name>
or aztfy query <arg where predicate>
to import a single resource, a resource group including child resources, or a customized set of resources by an Azure Resource Graph query.
aztfy resource [option] <resource id>
terrafies a single resource by its Azure control plane ID.
E.g.
aztfy resource /subscriptions/0000/resourceGroups/rg1/providers/Microsoft.Compute/virtualMachines/vm1
The command will automatically identify the Terraform resource type (e.g. correctly identifies above resource as azurerm_linux_virtual_machine
), and import it into state file and generate the Terraform configuration.
❗For data plane only resources (e.g.
azurerm_key_vault_certificate
), the resource id is using a pesudo format, as is defined here.
aztfy resource-group [option] <resource group name>
terrafies a resource group and its including resources by its name. Depending on whether --batch
is used, it can work in either interactive mode or batch mode.
aztfy query [option] <arg where predicate>
terrafies a set of resources (and its including resources with --recursive
) by an Azure Resource Graph where
predicate. Depending on whether --batch
is used, it can work in either interactive mode or batch mode.
💡 Resource group mode is the same as running
aztfy query --recursive "resourceGroup =~ 'my-rg'"
, except it also add on the resource group itself.
aztfy
depends on azlist
, which uses ARG behind the scenes. azlist
will first make an ARG call with the given where predicate, then if --recursive
is specified, it will recursively call the "LIST" on the known child resource types. Since ARG only returns ARM tracked resources at this moment, but not for the RP proxy resources (e.g. subnet, network security rules, storage containers, etc). If you uses predicate like type =~ "microsoft.network/virtualnetworks/subnets"
, it returns you nothing since subnet is not an ARM tracked resource.
To workaround above, you can query with a bigger scope (e.g. type =~ "microsoft.network/virtualnetworks"
) in interactive mode, then manually remove the resources other than subnets.
In interactive mode, aztfy
list all the resources resides in the specified resource group or customized set. For each resource, user is expected to input the Terraform resource address in form of <resource type>.<resource name>
(e.g. azurerm_linux_virtual_machine.test
). Users can press r
to see the possible resource type(s) for the selected import item. In case there is exactly one resource type match for the import item, that resource type will be automatically filled in the text input for the users, with a 💡 line prefix as an indication.
In some cases, there are Azure resources that have no corresponding Terraform resources (e.g. due to lacks of Terraform support), or some resources might be created as a side effect of provisioning another resource (e.g. the OS Disk resource is created automatically when provisioning a VM). In these cases, you can skip these resources without typing anything.
💡 Option
--resource-mapping
/-m
can be used to specify a resource mapping file, either constructed manually or from other runs ofaztfy
(generated in the output directory with name: .aztfyResourceMapping.json).
After going through all the resources to be imported, users press w
to instruct aztfy
to proceed importing resources into Terraform state and generating the Terraform configuration.
As the last step, aztfy
will leverage the ARM template to inject dependencies between each resource. This makes the generated Terraform template to be useful.
In batch mode, instead of interactively specifying the mapping from Azurem resource id to the Terraform resource info, users are expected to provide that mapping via the resource mapping file, with the following format:
{
"<Azure resource id1>": {
"resource_type" : "<terraform resource type>",
"resource_name" : "<terraform resource name>",
"resource_id" : "<terraform resource id>"
},
"<Azure resource id2>": {
"resource_type" : "<terraform resource type>",
"resource_name" : "<terraform resource name>",
"resource_id" : "<terraform resource id>"
},
...
}
Example:
{
"/subscriptions/0000/resourceGroups/aztfy-vmdisk": {
"resource_id": "/subscriptions/0000/resourceGroups/aztfy-vmdisk",
"resource_type": "azurerm_resource_group",
"resource_name": "res-1"
},
"/subscriptions/0000/resourceGroups/aztfy-vmdisk/providers/Microsoft.Compute/disks/aztfy-test-test": {
"resource_id": "/subscriptions/0000/resourceGroups/aztfy-vmdisk/providers/Microsoft.Compute/disks/aztfy-test-test",
"resource_type": "azurerm_managed_disk",
"resource_name": "res-2"
},
"/subscriptions/0000/resourceGroups/aztfy-vmdisk/providers/Microsoft.Compute/virtualMachines/aztfy-test-test": {
"resource_id": "/subscriptions/0000/resourceGroups/aztfy-vmdisk/providers/Microsoft.Compute/virtualMachines/aztfy-test-test",
"resource_type": "azurerm_linux_virtual_machine",
"resource_name": "res-3"
},
...
}
Then the tool will import each specified resource in the mapping file (if exists) and skip the others.
Especially if the no resource mapping file is specified, aztfy
will only import the "recognized" resources for you, based on its knowledge on the ARM and Terraform resource mappings.
In the batch import mode, users can further specify the --continue
/-k
option to make the tool continue even on hitting import error of any resource.
By default aztfy
uses local backend to store the state file. While it is also possible to use remote backend, via the --backend-type
and --backend-config
options.
E.g. to use the azurerm
backend, users can invoke aztfy
like following:
aztfy --backend-type=azurerm --backend-config=resource_group_name=<resource group name> --backend-config=storage_account_name=<account name> --backend-config=container_name=<container name> --backend-config=key=terraform.tfstate <importing resource group name>
For local backend, aztfy
will by default ensure the output directory is empty at the very begining. This is to avoid any conflicts happen for existing user files, including the terraform configuration, provider configuration, the state file, etc. As a result, aztfy
generates a pretty new workspace for users.
One limitation of doing so is users can't import resources to existing state file via aztfy
. To support this scenario, you can use the --append
option. This option will make aztfy
skip the empty guarantee for the output directory. If the output directory is empty, then it has no effect. Otherwise, it will ensure the provider setting (create a file for it if not exists). Then it proceeds the following steps.
This means if the output directory has an active Terraform workspace, i.e. there exists a state file, any resource imported by the aztfy
will be imported into that state file. Especially, the file generated by aztfy
in this case will be named differently than normal, where each file will has .aztfy
suffix before the extension (e.g. main.aztfy.tf
), to avoid potential file name conflicts. If you run aztfy --append
multiple times, the generated config in main.aztfy.tf
will be appended in each run.
aztfy
leverage aztft
to identify the Terraform resource type on its Azure resource ID. Then it runs terraform import
under the hood to import each resource. Afterwards, it runs tfadd
to generate the Terraform template for each imported resource.
There are several limitations causing aztfy
can hardly generate reproducible Terraform configurations.
Azure resources are modeled differently in AzureRM provider.
For example, the azurerm_lb_backend_address_pool_address
is actually a property of azurerm_lb_backend_address_pool
in Azure platform. Whilst in the AzureRM provider, it has its own resource and a synthetic resource ID as /subscriptions/00000000-0000-0000-0000-000000000000/resourceGroups/group1/providers/Microsoft.Network/loadBalancers/loadBalancer1/backendAddressPools/backendAddressPool1/addresses/address1
.
Another popular case is that in the AzureRM provider, there are a bunch of "association" resources, e.g. the azurerm_network_interface_security_group_association
. These "association" resources represent the association relationship between two Terraform resources (in this case they are azurerm_network_interface
and azurerm_network_security_group
). They also have some synthetic resource ID, e.g. /subscriptions/00000000-0000-0000-0000-000000000000/resourceGroups/mygroup1/providers/microsoft.network/networkInterfaces/example|/subscriptions/00000000-0000-0000-0000-000000000000/resourceGroups/group1/providers/Microsoft.Network/networkSecurityGroups/group1
.
Currently, this tool only works on the assumption that there is 1:1 mapping between Azure resources and the Terraform resources. For those property-like Terraform resources, aztfy
will just ignore them.
When generating the Terraform configuration, not all properties of the resource are exported for different reasons.
One reason is because there are flexible cross-property constraints defined in the AzureRM Terraform provider. E.g. property_a
conflits with property_b
. This might due to the nature of the API, or might be due to some deprecation process of the provider (e.g. property_a
is deprecated in favor of property_b
, but kept for backwards compatibility). These constraints require some properties must be absent in the Terraform configuration, otherwise, the configuration is not a valid and will fail during terraform validate
.
Another reason is that an Azure resource can be a property of its parent resource (e.g. azurerm_subnet
can be its own resource, or be a property of azurerm_virtual_network
). Per Terraform's best practice, users should only use one of the forms, not both. aztfy
chooses to always generate all the resources, but omit the property in the parent resource that represents the child resource.
- The aztfy Github Page: Everything about aztfy, including comparisons with other existing import solutions.
- Kyle Ruddy's Blog about aztfy: A live use of
aztfy
, explaining the pros and cons. - aztft: A Go program and library for identifying the correct Terraform AzureRM provider resource type on the Azure resource id.
- tfadd: A Go program and library for generating Terraform configuration from Terraform state.