Include CloudFormation templates in the CDK
This module contains a set of classes whose goal is to facilitate working
with existing CloudFormation templates in the CDK.
It can be thought of as an extension of the capabilities of the
CfnInclude
class.
Basic usage
Assume we have a file with an existing template.
It could be in JSON format, in a file my-template.json
:
{
"Resources": {
"Bucket": {
"Type": "AWS::S3::Bucket",
"Properties": {
"BucketName": "some-bucket-name"
}
}
}
}
Or it could by in YAML format, in a file my-template.yaml
:
Resources:
Bucket:
Type: AWS::S3::Bucket
Properties:
BucketName: some-bucket-name
It can be included in a CDK application with the following code:
cfn_template = cfn_inc.CfnInclude(self, "Template",
template_file="my-template.json"
)
Or, if your template uses YAML:
cfn_template = cfn_inc.CfnInclude(self, "Template",
template_file="my-template.yaml"
)
Note: different YAML parsers sometimes don’t agree on what exactly constitutes valid YAML. If you get a YAML exception when including your template, try converting it to JSON, and including that file instead. If you’re downloading your template from the CloudFormation AWS Console, you can easily get it in JSON format by clicking the ‘View in Designer’ button on the ‘Template’ tab - once in Designer, select JSON in the “Choose template language” radio buttons on the bottom pane.
This will add all resources from my-template.json
/ my-template.yaml
into the CDK application,
preserving their original logical IDs from the template file.
Any resource from the included template can be retrieved by referring to it by its logical ID from the template. If you know the class of the CDK object that corresponds to that resource, you can cast the returned object to the correct type:
# cfn_template: cfn_inc.CfnInclude
cfn_bucket = cfn_template.get_resource("Bucket")
Note that any resources not present in the latest version of the CloudFormation schema
at the time of publishing the version of this module that you depend on,
including Custom Resources,
will be returned as instances of the class CfnResource
,
and so cannot be cast to a different resource type.
Any modifications made to that resource will be reflected in the resulting CDK template; for example, the name of the bucket can be changed:
# cfn_template: cfn_inc.CfnInclude
cfn_bucket = cfn_template.get_resource("Bucket")
cfn_bucket.bucket_name = "my-bucket-name"
You can also refer to the resource when defining other constructs,
including the higher-level ones
(those whose name does not start with Cfn
),
for example:
# cfn_template: cfn_inc.CfnInclude
cfn_bucket = cfn_template.get_resource("Bucket")
role = iam.Role(self, "Role",
assumed_by=iam.AnyPrincipal()
)
role.add_to_policy(iam.PolicyStatement(
actions=["s3:*"],
resources=[cfn_bucket.attr_arn]
))
Migrating templates that use Transforms
You can use this module to migrate templates that use CloudFormation transforms - including the Serverless transform.
The CDK including process does not execute Transforms,
and the cdk diff
command by default compares against the original
(meaning, unprocessed) template.
So, if you’re downloading the template to include from the CloudFormation AWS Console,
make sure to download the unprocessed template
(the “View processed template” checkbox is left unchecked, which is the default):
However, certain unprocessed templates can fail when used with the CfnInclude
class.
The most common reason for the failure is that the unprocessed template can contain cycles between resources,
which get removed after the Transform is processed,
but is not allowed when being included (as pure CloudFormation does not permit cycles). To enable cycle processing behavior similar
to cloudformation, set allowCyclicalReferences
of CfnIncludeProps to true.
When that happens, you should instead download the processed template from the CloudFormation AWS Console (make sure the “View processed template” checkbox is checked in that case):
When you include that processed template in your CDK application,
running cdk diff
will now show a lot of differences with the deployed Stack,
because cdk diff
uses the unprocessed template by default.
To alleviate that problem, you can pass the --processed
switch to cdk diff
,
which will make the diff command compare against the processed template of the deployed Stack,
which will give more precise results in this case.
Converting L1 resources to L2
The resources the getResource
method returns are what the CDK calls
Layer 1 resources
(like CfnBucket
).
However, in many places in the Construct Library,
the CDK requires so-called Layer 2 resources, like IBucket
.
There are two ways of going from an L1 to an L2 resource.
UsingfromCfn*()
methods
This is the preferred method of converting an L1 resource to an L2.
It works by invoking a static method of the class of the L2 resource
whose name starts with fromCfn
-
for example, for KMS Keys, that would be the Kms.fromCfnKey()
method -
and passing the L1 instance as an argument:
# cfn_template: cfn_inc.CfnInclude
cfn_key = cfn_template.get_resource("Key")
key = kms.Key.from_cfn_key(cfn_key)
This returns an instance of the kms.IKey
type that can be passed anywhere in the CDK an IKey
is expected.
What is more, that IKey
instance will be mutable -
which means calling any mutating methods on it,
like addToResourcePolicy()
,
will be reflected in the resulting template.
Note that, in some cases, the fromCfn*()
method might not be able to create an L2 from the underlying L1.
This can happen when the underlying L1 heavily uses CloudFormation functions.
For example, if you tried to create an L2 IKey
from an L1 represented as this CloudFormation template:
{
"Resources": {
"Key": {
"Type": "AWS::KMS::Key",
"Properties": {
"KeyPolicy": {
"Statement": [
{
"Fn::If": [
"Condition",
{
"Action": "kms:if-action",
"Resource": "*",
"Principal": "*",
"Effect": "Allow"
},
{
"Action": "kms:else-action",
"Resource": "*",
"Principal": "*",
"Effect": "Allow"
}
]
}
],
"Version": "2012-10-17"
}
}
}
}
}
The Key.fromCfnKey()
method does not know how to translate that into CDK L2 concepts,
and would throw an exception.
In those cases, you need the use the second method of converting an L1 to an L2.
Using from*Name/Arn/Attributes()
methods
If the resource you need does not have a fromCfn*()
method,
or if it does, but it throws an exception for your particular L1,
you need to use the second method of converting an L1 resource to L2.
Each L2 class has static factory methods with names like from*Name()
,
from*Arn()
, and/or from*Attributes()
.
You can obtain an L2 resource from an L1 by passing the correct properties of the L1 as the arguments to those methods:
# cfn_template: cfn_inc.CfnInclude
# using from*Attributes()
# private_cfn_subnet1: ec2.CfnSubnet
# private_cfn_subnet2: ec2.CfnSubnet
# using from*Name()
cfn_bucket = cfn_template.get_resource("Bucket")
bucket = s3.Bucket.from_bucket_name(self, "L2Bucket", cfn_bucket.ref)
# using from*Arn()
cfn_key = cfn_template.get_resource("Key")
key = kms.Key.from_key_arn(self, "L2Key", cfn_key.attr_arn)
cfn_vpc = cfn_template.get_resource("Vpc")
vpc = ec2.Vpc.from_vpc_attributes(self, "L2Vpc",
vpc_id=cfn_vpc.ref,
availability_zones=core.Fn.get_azs(),
private_subnet_ids=[private_cfn_subnet1.ref, private_cfn_subnet2.ref]
)
As long as they just need to be referenced,
and not changed in any way, everything should work;
however, note that resources returned from those methods,
unlike those returned by fromCfn*()
methods,
are immutable, which means calling any mutating methods on them will have no effect.
You will have to mutate the underlying L1 in order to change them.
Non-resource template elements
In addition to resources, you can also retrieve and mutate all other template elements:
-
# cfn_template: cfn_inc.CfnInclude param = cfn_template.get_parameter("MyParameter") # mutating the parameter param.default = "MyDefault"
-
# cfn_template: cfn_inc.CfnInclude condition = cfn_template.get_condition("MyCondition") # mutating the condition condition.expression = core.Fn.condition_equals(1, 2)
-
# cfn_template: cfn_inc.CfnInclude mapping = cfn_template.get_mapping("MyMapping") # mutating the mapping mapping.set_value("my-region", "AMI", "ami-04681a1dbd79675a5")
Service Catalog template Rules:
# cfn_template: cfn_inc.CfnInclude # mutating the rule # my_parameter: core.CfnParameter rule = cfn_template.get_rule("MyRule") rule.add_assertion(core.Fn.condition_contains(["m1.small"], my_parameter.value_as_string), "MyParameter has to be m1.small")
-
# cfn_template: cfn_inc.CfnInclude # mutating the output # cfn_bucket: s3.CfnBucket output = cfn_template.get_output("MyOutput") output.value = cfn_bucket.attr_arn
Hooks for blue-green deployments:
# cfn_template: cfn_inc.CfnInclude # mutating the hook # my_role: iam.Role hook = cfn_template.get_hook("MyOutput") code_deploy_hook = hook code_deploy_hook.service_role = my_role.role_arn
Parameter replacement
If your existing template uses CloudFormation Parameters,
you may want to remove them in favor of build-time values.
You can do that using the parameters
property:
cfn_inc.CfnInclude(self, "includeTemplate",
template_file="path/to/my/template",
parameters={
"MyParam": "my-value"
}
)
This will replace all references to MyParam
with the string 'my-value'
,
and MyParam
will be removed from the ‘Parameters’ section of the resulting template.
Nested Stacks
This module also supports templates that use nested stacks.
For example, if you have the following parent template:
{
"Resources": {
"ChildStack": {
"Type": "AWS::CloudFormation::Stack",
"Properties": {
"TemplateURL": "https://my-s3-template-source.s3.amazonaws.com/child-stack.json"
}
}
}
}
where the child template pointed to by https://my-s3-template-source.s3.amazonaws.com/child-stack.json
is:
{
"Resources": {
"MyBucket": {
"Type": "AWS::S3::Bucket"
}
}
}
You can include both the parent stack, and the nested stack in your CDK application as follows:
parent_template = cfn_inc.CfnInclude(self, "ParentStack",
template_file="path/to/my-parent-template.json",
load_nested_stacks={
"ChildStack": cfn_inc.CfnIncludeProps(
template_file="path/to/my-nested-template.json"
)
}
)
Here, path/to/my-nested-template.json
represents the path on disk to the downloaded template file from the original template URL of the nested stack
(https://my-s3-template-source.s3.amazonaws.com/child-stack.json
).
In the CDK application,
this file will be turned into an Asset,
and the TemplateURL
property of the nested stack resource
will be modified to point to that asset.
The included nested stack can be accessed with the getNestedStack
method:
# parent_template: cfn_inc.CfnInclude
included_child_stack = parent_template.get_nested_stack("ChildStack")
child_stack = included_child_stack.stack
child_template = included_child_stack.included_template
Now you can reference resources from ChildStack
,
and modify them like any other included template:
# child_template: cfn_inc.CfnInclude
cfn_bucket = child_template.get_resource("MyBucket")
cfn_bucket.bucket_name = "my-new-bucket-name"
role = iam.Role(self, "MyRole",
assumed_by=iam.AccountRootPrincipal()
)
role.add_to_policy(iam.PolicyStatement(
actions=["s3:GetObject*", "s3:GetBucket*", "s3:List*"
],
resources=[cfn_bucket.attr_arn]
))
You can also include the nested stack after the CfnInclude
object was created,
instead of doing it on construction:
# parent_template: cfn_inc.CfnInclude
included_child_stack = parent_template.load_nested_stack("ChildTemplate",
template_file="path/to/my-nested-template.json"
)
Vending CloudFormation templates as Constructs
In many cases, there are existing CloudFormation templates that are not entire applications,
but more like specialized fragments, implementing a particular pattern or best practice.
If you have templates like that,
you can use the CfnInclude
class to vend them as CDK Constructs:
from constructs import Construct
import aws_cdk.cloudformation_include as cfn_inc
import path as path
class MyConstruct(Construct):
def __init__(self, scope, id):
super().__init__(scope, id)
# include a template inside the Construct
cfn_inc.CfnInclude(self, "MyConstruct",
template_file=path.join(__dirname, "my-template.json"),
preserve_logical_ids=False
)
Notice the preserveLogicalIds
parameter -
it makes sure the logical IDs of all the included template elements are re-named using CDK’s algorithm,
guaranteeing they are unique within your application.
Without that parameter passed,
instantiating MyConstruct
twice in the same Stack would result in duplicated logical IDs.