OpenShift cluster manager API model

Introduction

This project contains the specification of the OpenShift cluster manager API, also known as the model.

Concepts

The specification of the API is written using a DSL (the model language) similar to Go.

The specification is split into multiple model files with extension .model inside the model directory of this project. The location of the model files within sub-directories is important as it indicates which service and version a type or resource belongs to. For example, the clusters_mgmt/v1 sub-directory corresponds to version 1 of the clusters management service.

Data types are represented by class, struct and enum blocks.

For example, the Cluster type defining the cluster concept should be located in a file within the clusters_mgmt/v1 directory, should be named cluster_type.model and should contain something like this:

// Definition of an _OpenShift_ cluster.
class Cluster {
	// Name of the cluster.
	Name String

	// Link to the _flavour_ that was used to create the cluster.
	link Flavour Flavour

	...
}

Classes and structs contain attributes defined by an attribute name followed by the type of that attribute. In the above example there are two attributes defined: Name of type string and Flavour of type Flavour.

In the model language attributes are written using CamelCase because that is what is typically used in Go, which is the language that the model language tries to be close to. But the actual JSON representation uses snake_case. These are some examples of what is the correspondence between attribute names in the model and in JSON:

Model	JSON
ID	id
HREF	href
Cluster	cluster
AccessKeyID	access_key_id

The main difference between classes and structs is that classes are intended to represent objects that have an identity. In practice that means that classes have built-in ID and HREF attributes. The ID attribute will contain the unique identifier of the object, and the HREF attribute will contain the location of the object in the server. For example, when retrieving a particular cluster from the server the returned JSON document will be like this:

{
    "id": "123",
    "href": "/api/clusters_mgmt/v1/clusters/123",
    ...
}

Resources are represented by resource blocks.

For example, the Clusters resource defining the collection of clusters should be located in a file within the clusters_mgmt/v1 directory, should be named clusters_resource.model and should contain something like this:

// Manages the collection of clusters.
resource Clusters {
	// Retrieves the list of clusters.
	method List {
		...
	}

	// Provision a new cluster and add it to the collection of clusters.
	method Add {
		...
	}

	...
}

Resource methods are represented as nested method blocks.

For example, the method that retrieves the list of clusters is defined in a nested List method block inside the resource block for the clusters resource:

// Retrieves the list of clusters.
method List {
	// Index of the requested page, where one corresponds to the first page.
	in out Page Integer

	// Maximum number of items that will be contained in the returned page.
	in out Size Integer

	// Search criteria.
	in Search String

	// Order criteria.
	in Order String

	// Total number of items of the collection that match the search criteria,
	// regardless of the size of the page.
	out Total Integer

	// Retrieved list of clusters.
	out Items []Cluster
}

Methods have parameters defined by their direction (in or out), their name and their type. In the above example there are four input parameters (named Page, Size, Search and Order) and four output parameter (named Page, Size, Total and Items).

• "List" conceptual method is implemented by HTTP GET method, with in parameters represented in the URL as HTTP query parameters (with names converted from CamelCase to snake_case).

  cluster_mgmt API supports an alternative way to List — as HTTP POST method, with a method=get query parameter and in paramaters sent as fields of a JSON request body (again, with snake_case names).

  Out parameters become top-level fields in the JSON response body (again, with snake_case names). An additional kind field is added automatically (set to the type’s name + a "List" suffix), it should not be included in the method block.

• "Get" is regular HTTP GET method, declared with single out parameter representing the JSON response body.

• "Add" is performed by HTTP POST method, declared with single in out parameter representing the JSON request body — as well as the response body.

• "Delete" is performed by HTTP DELETE method, with no parameters.

In addition to methods resources also have locators, defined by nested locator blocks. Locators represent the relationships between resources. For example, the resource that manages the collection of clusters knows how to locate the resource that manages a specific cluster. That is represented in the model language with a locator block like this:

// Returns a reference to the service that manages an specific cluster.
locator Cluster {
	target Cluster
	variable ID
}

All resource locators have a name and a target. The target is defined using the target keyword and the name of the resource.

There are two kinds of resource locators: with and without variable.

Locators with variable are intended for collections, where location a sub-resource resource requires specifying the identifier of that object that is managed by that sub-resource. For example, to locate the sub-resource that manages a specific cluster it is necessary to provide the identifier of that cluster. That identifier is the variable. These kind of locators are defined using the variable keyword and the name of the variable, like in the previous example.

Locators without variable are intended for cases where no additional information is needed to identify the sub-resource. For example, the locator for the credentials sub-resource of a cluster can be defined like this:

// Reference to the resource that manages the credentials of the cluster.
locator Credentials {
	target Credentials
}

Locators also define the URL structure of the API: the path component of the URL of a particular resource is constructed concatenating the names/variables of the locators that are in the chain of locators from the root to that resource. For example, to get to the Credentials resource of cluster with identifier 123 the chain starts at the root resource of the clusters management service, it continues with the Clusters locator to find the clusters collection, then the Cluster locator with variable 123 to get the cluster resource and finally the Credentials locator to get to the credentials resource:

Root -> Clusters -> Cluster(123) -> Credentials

Each link in that chain of locators is translated into an URL path segment using the following rules:

• The root resource corresponds to the root of the service/version. For example, for version 1 of the clusters management service that would be /api/clusters_mgmt/v1.

• Locators without variables correspond to URL segments named like the locator, but using snake_case instead of CamelCase. For example, for the first link in the above example the URL path segment would be clusters.

• Locators with variables correspond to URL segments that contain the value of the variable. For example, for the second link in the above example the URL path segment would be 123.

Taking these rules into account the complete URL path for the above example would be the following:

/api/clusters_mgmt/v1/clusters/123/credentials

Documentation

The Go language supports adding documentation in the code itself, using the documentation comments. These comments start with // and appear immediately before the documented item. The model language uses the same kind of documentation comments. For example, the Cluster type can be documented like this:

// Definition of an _OpenShift_ cluster.
//
// The `cloud_provider` attribute is a reference to the cloud provider. When a
// cluster is retrieved it will be a link to the cloud provider, containing only
// the kind, id and href attributes:
//
// [source,json]
// ----
// {
//   "cloud_provider": {
//     "kind": "CloudProviderLink",
//     "id": "123",
//     "href": "/api/clusters_mgmt/v1/cloud_providers/123"
//   }
// }
// ----
//
// When a cluster is created this is optional, and if used it should contain the
// identifier of the cloud provider to use:
//
// [source,json]
// ----
// {
//   "cloud_provider": {
//     "id": "123",
//   }
// }
// ----
//
// If not included, then the cluster will be created using the default cloud
// provider, which is currently Amazon Web Services.
//
// The region attribute is mandatory when a cluster is created.
//
// The `aws.access_key_id`, `aws.secret_access_key` and `dns.base_domain`
// attributes are mandatory when creation a cluster with your own Amazon Web
// Services account.
class Cluster {
	...
}

Unlike Go the format of this documentation isn’t plain text, but Asciidoc.

Attributes of types, methods of resources and parameters of methods can all be documented in a similar way, just placing documentation comment before the definition of the item. For example, to document the Search parameter of the List method of the Clusters resource the following documentation comment could be used:

// Search criteria.
//
// The syntax of this parameter is similar to the syntax of the _where_ clause of a
// SQL statement, but using the names of the attributes of the cluster instead of
// the names of the columns of a table. For example, in order to retrieve all the
// clusters with a name starting with `my` in the `us-east-1` region the value
// should be:
//
// [source,sql]
// ----
// name like 'my%' and region.id = 'us-east-1'
// ----
//
// If the parameter isn't provided, or if the value is empty, then all the
// clusters that the user has permission to see will be returned.
in Search String

This documentation is used to automatically generate OpenAPI reference documentation (with some constructs converted to markdown).