This document describes the Resource Provider Definition Schema which is a meta-schema that extends draft-07 of JSON Schema to define a validating document against which resource schemas can be authored.
Numerous examples exist in this repository to help you understand various shape and semantic definition models you can apply to your own resource definitions.
The meta-schema which controls and validates your resource type definition is called the Resource Provider Definition Schema. It is fully compliant with draft-07 of JSON Schema and many IDEs including IntelliJ, PyCharm and Visual Studio Code come with built-in or plugin-based support for code-completion and syntax validation while editing documents for JSON Schema compliance. Comprehensive documentation for JSON Schema exists and can answer many questions around correct usage.
To get started, you will author a specification for your resource type in a JSON document, which must be compliant with this meta-schema. To make authoring resource specifications simpler, we have constrained the scope of the full JSON Schema standard to apply opinions around how certain validations can be expressed and encourage consistent modelling for all resource schemas. These opinions are codified in the meta-schema and described in this document.
All resources MUST specify a typeName
which adheres to the Regular Expression ^[a-zA-Z0-9]{2,64}::[a-zA-Z0-9]{2,64}::[a-zA-Z0-9]{2,64}$
. This expression defines a 3-part namespace for your resource, with a suggested shape of Organization::Service::Resource
. For example AWS::EC2::Instance
or Initech::TPS::Report
. This typeName
is how you will address your resources for use in CloudFormation and other provisioning tools.
The shape of your resource defines the properties for that resource and how they should be applied. This includes the type of each property, validation patterns or enums, and additional descriptive metadata such a documentation and example usage. Refer to the #/definitions/properties
section of the meta-schema for the full set of supported properties you can use to describe your resource shape.
Certain properties of a resource are semantic and have special meaning when used in different contexts. For example, a property of a resource may be readOnly
when read back for state changes - but can be specified in a settable context when used as the target of a $ref
from a related resource. Because of this semantic difference in how this property metadata should be interpreted, certain aspects of the resource definition are applied to the parent resource definition, rather than at a property level. Those elements are;
primaryIdentifier
: Must be either a single property, or a set of properties which can be used to uniquely identify the resource. If multiple properties are specified, these are treated as a composite key and combined into a single logical identifier. You would use this modelling to express contained identity (such as a named service within a container). This property can be independently provided as keys to a READ or DELETE request and MUST be supported as the only input to those operations. These properties are usually also marked asreadOnlyProperties
and MUST be returned from READ and LIST operations.additionalIdentifiers
: Each property listed in theadditionalIdentifiers
section must be able to be used to uniquely identify the resource. These properties can be independently provided as keys to a READ or DELETE request and MUST be supported as the only input to those operations. These properties are usually also marked asreadOnlyProperties
and MUST be returned from READ and LIST operations. A provider is not required to supportadditionalIdentifiers
; doing so allows for other unique keys to be used to READ resources.readOnlyProperties
: A property in thereadOnlyProperties
list cannot be specified by the customer.writeOnlyProperties
: A property in thewriteOnlyProperties
cannot be returned in a READ or LIST request, and can be used to express things like passwords, secrets or other sensitive data.createOnlyProperties
: A property in thecreateOnlyProperties
cannot be specified in an UPDATE request, and can only be specified in a CREATE request. Another way to think about this - these are properties which are 'write-once', such as theEngine
property for anAWS::RDS::DBInstance
and if you wish to change such a property on a live resource, you should replace that resource by creating a new instance of the resource and terminating the old one. This is the behaviour CloudFormation follows for all properties documented as 'Update Requires: Replacement'. An attempt to supply these properties to an UPDATE request will produce a runtime error from the handler.deprecatedProperties
: A property in thedeprecatedProperties
is not guaranteed to be present in the response from a READ request. These fields will still be accepted as input to CREATE and UPDATE requests however they may be ignored, or converted to new API forms when outbound service calls are made.
When defining resource semantics like createOnlyProperties
, primaryIdentifier
you are expected to use a JSON Pointer to a property definition in the same resource document. Schemas you author can be checked with the CFN CLI validate
command.
The following (truncated) example shows some of the semantic definitions for an AWS::S3::Bucket
resource type;
{
"$id": "aws-s3-bucket.json",
"typeName": "AWS::S3::Bucket",
"resourceLink": {
"templateUri": "/s3/home?region=${awsRegion}&bucket=${BucketName}",
"mappings": {
"BucketName": "/BucketName"
}
},
"definitions": { ... },
"properties": {
"Arn": {
"$ref": "aws.common.types.v1.json#/definitions/Arn"
},
"BucketName": {
"type": "string"
}
},
"createOnlyProperties": [
"/properties/BucketName"
],
"readOnlyProperties": [
"/properties/Arn"
],
"primaryIdentifier": [
"/properties/BucketName"
],
"additionalIdentifiers": [
"/properties/Arn",
"/properties/WebsiteURL"
]
}
Relationships between resources can be expressed through the use of the $ref
keyword when defining a property schema. The use of the $ref
keyword to establish relationships is described in JSON Schema documentation.
The following example shows a property relationship between an AWS::EC2::Subnet.VpcId
and an AWS::EC2::VPC.Id
. The schema for the 'remote' type (AWS::EC2::VPC
) is used to validate the content of the 'local' type (AWS::EC2::Subnet
) and can be inferred as a dependency from the local to the remote type.
Setting the $id property to a remote location will make validation framework to pull dependencies expressed using relative $ref
URIs from the remote hosts. In this example, VpcId
property will be verified against the schema for AWS::EC2::VPC.Id
hosted at https://schema.cloudformation.us-east-1.amazonaws.com/aws-ec2-vpc.json
{
"$id": "https://schema.cloudformation.us-east-1.amazonaws.com/aws-ec2-subnet.json",
"typeName": "AWS::EC2::Subnet",
"definitions": { ... },
"properties": {
{ ... }
"VpcId": {
"$ref": "aws-ec2-vpc.json#/properties/Id"
}
}
}
{
"$id": "https://schema.cloudformation.us-east-1.amazonaws.com/aws-ec2-vpc.json",
"typeName": "AWS::EC2::VPC",
"definitions": { ... },
"properties": {
"Id": {
"type": "string",
"pattern": "$vpc-[0-9]{8,10}^"
}
}
}
We have taken an opinion on certain aspects of the core JSON Schema and introduced certain constrains and changes from the core schema. In the context of this project, we are not building arbitrary documents, but rather, defining a very specific shape and semantic for cloud resources.
readOnly
: the readOnly field as defined in JSON Schema does not align with our determination that this is actually a restriction with semantic meaning. A property may be readOnly when specified for a particular resource (for example it'sArn
), but when that same property is referenced (using$ref
tokens) from a dependency, the dependency must be allowed to specify an input for that property, and as such, it is no longerreadOnly
in that context. The AWS CloudFormation Resource Schema uses the concept ofreadOnlyProperties
for this mechanic.writeOnly
: see above
Array types can define a boolean insertionOrder
, which specifies whether the order in which elements are specified should be honored when processing a diff between two sets of properties. If insertionOrder
is true, then a change in order of the elements will constitute a diff. The default for insertionOrder
is true.
Together with the uniqueItems
property (which is native to JSON Schema), complex array types can be defined, as in the following table:
insertionOrder | uniqueItems | result |
---|---|---|
true | false | list |
false | false | multiset |
true | true | ordered set |
false | true | set |
$id
: an$id
property is not valid for a resource property.$schema
: a$schema
property is not valid for a resource property.if
,then
,else
,not
: these imperative constructs can lead to confusion both in authoring a resource definition, and for customers authoring a resource description against your schema. Also this construct is not widely supported by validation tools and is disallowed here.propertyNames
: use ofpropertyNames
implies a set of properties without a defined shape and is disallowed. To constrain property names, usepatternProperties
statements with defined shapes.additionalProperties
use ofadditionalProperties
is not valid for a resource property. UsepatternProperties
instead to define the shape and allowed values of extraneous keys.properties
andpatternProperties
it is not valid to use both properties and patternProperties together in the same shape, as a shape should not contain both defined and undefined values. In order to implement this, the set of undefined values should itself be a subshape.items
andadditionalItems
theitems
in an array may only have one schema and may not use a list of schemas, as an ordered tuple of different objects is confusing for both developers and customers. This should be expressed as key:value object pairs. Similarly,additionalItems
is not allowed.
The handlers
section of the schema allows you to specify which CRUDL operations (create, read, update, delete, list) are available for your resource, as well as some additional metadata about each handler.
For each handler, you should define a list of API permissions
required to perform the operation. Currently, this is used to generate IAM policy templates and is assumed to be AWS API permissions, but you may list 3rd party APIs here as well.
For each handler, you may define a timeoutInMinutes
property, which defines the maximum timeout of the operation. This timeout is used by the invoker of the handler (such as CloudFormation) to stop listening and cancel the operation. Note that the handler may of course decide to timeout and return a failure prior to this max timeout period. Currently, this value is only used for Create
, Update
, and Delete
handlers, while Read
and List
handlers are expected to return synchronously within 30 seconds.
This library is licensed under the Apache 2.0 License.