This document is also available in this non-normative format: pdf
This document is licensed under a Creative Commons Attribution 4.0 License.
This document contains a normative standard for designing APIs in the Dutch Public Sector. The Governance of this standard is described in a separate repository and published by Logius.
Het OBDO heeft op advies van het Forum Standaardisatie deze versie vastgesteld.
This section is non-normative.
More and more governmental organizations offer REST APIs (henceforth abbreviated as APIs), in addition to existing interfaces like SOAP and WFS. These APIs aim to be developer-friendly and easy to implement. While this is a commendable aim, it does not shield a developer from a steep learning curve getting to know every new API, in particular when every individual API is designed using different patterns and conventions.
This document aims to describe a widely applicable set of design rules for the unambiguous provisioning of REST APIs. The primary goal is to offer guidance for organizations designing new APIs, with the purpose of increasing developer experience (DX) and interoperability between APIs. Hopefully, many organizations will adopt these design rules in their corporate API strategies and provide feedback about exceptions and additions to subsequently improve these design rules.
This version of the design rules has been submitted to Forum Standaardisatie for inclusion on the Comply or Explain list of mandatory standards in the Dutch Public Sector. This document originates from the document API Strategie voor de Nederlandse Overheid, which was recently split into separate sub-documents.
This document is part of the Nederlandse API Strategie.
The Nederlandse API Strategie consists of three distinct documents.
In addition to this (normative) document, a separate document has been written providing a set of informative extensions. This extensions document exists in a latest published version (Gepubliceerde versie in Dutch) and a latest editors draft (Werkversie in Dutch). The latest editor's drafts is actively being worked on and can be found on GitHub. It contains the most recent changes.
The documents can be found here:
The REST architectural style is centered around the concept of a resource. A resource is the key abstraction of information, where every piece of information is named by assigning a globally unique URI (Uniform Resource Identifier). Resources describe things, which can vary between physical objects (e.g. a building or a person) and more abstract concepts (e.g. a permit or an event).
API-05: Use nouns to name resources
Because resources describe things (and thus not actions), resources are referred to using nouns (instead of verbs) that are relevant from the perspective of the user of the API.
A few correct examples of nouns as part of a URI:
This is different than RPC-style APIs, where verbs are often used to perform certain actions:
A resource describing a single thing is called a singular resource. Resources can also be grouped into collections, which are resources in their own right and can typically be paged, sorted and filtered. Most often all collection members have the same type, but this is not necessarily the case. A resource describing multiple things is called a collection resource. Collection resources typically contain references to the underlying singular resources.
API-54: Use plural nouns to name collection resources
Because a collection resource represents multiple things, the path segment describing the name of the collection resource must be written in the plural form.
Example collection resources, describing a list of things:
Singular resources contained within a collection resource are generally named by appending a path segment for the identification of each individual resource.
Example singular resource, contained within a collection resource:
Singular resources that stand on their own, i.e. which are not contained within a collection resource, must be named with a path segment that is written in the singular form.
Example singular resource describing the profile of the currently authenticated user:
API-04: Define interfaces in Dutch unless there is an official English glossary available
Since the exact meaning of concepts is often lost in translation, resources and the underlying attributes should be defined in the Dutch language unless there is an official English glossary available. Publishing an API for an international audience might also be a reason to define interfaces in English.
Note that glossaries exist that define useful sets of attributes which should preferably be reused. Examples can be found at schema.org.
API-48: Leave off trailing slashes from URIs
According to the URI specification [rfc3986], URIs may contain a trailing slash. However, for REST APIs this is considered as a bad practice since a URI including or excluding a trailing slash might be interpreted as different resources (which is strictly speaking the correct interpretation).
To avoid confusion and ambiguity, a URI must never contain a trailing slash. When requesting a resource including a trailing slash, this must result in a 404 (not found) error response and not a redirect. This enforces API consumers to use the correct URI.
URI without a trailing slash (correct):
URI with a trailing slash (incorrect):
API-53: Hide irrelevant implementation details
An API should not expose implementation details of the underlying application. The primary motivation behind this design rule is that an API design must focus on usability for the client, regardless of the implementation details under the hood. The API, application and infrastructure need to be able to evolve independently to ease the task of maintaining backwards compatibility for APIs during an agile development process.
A few examples of implementation details:
Although the REST architectural style does not impose a specific protocol, REST APIs are typically implemented using HTTP [rfc7231].
API-03: Only apply standard HTTP methods
The HTTP specification [rfc7231] and the later introduced
PATCH method specification [rfc5789] offer a set of standard methods, where every method is designed with explicit semantics. Adhering to the HTTP specification is crucial, since HTTP clients and middleware applications rely on standardized characteristics. Therefore, resources must be retrieved or manipulated using standard HTTP methods.
||Read||Retrieve a resource representation for the given URI. Data is only retrieved and never modified.|
||Create||Create a subresource as part of a collection resource. This operation is not relevant for singular resources. This method can also be used for exceptional cases.|
||Create/update||Create a resource with the given URI or replace (full update) a resource when the resource already exists.|
||Update||Partially updates an existing resource. The request only contains the resource modifications instead of the full resource representation.|
||Delete||Remove a resource with the given URI.|
The following table shows some examples of the use of standard HTTP methods:
||Retrieves a list of national monuments.|
||Retrieves an individual national monument.|
||Creates a new national monument.|
||Modifies national monument #12 completely.|
||Modifies national monument #12 partially.|
||Deletes national monument #12.|
HTTP also defines other methods, e.g.
TRACE. For the purpose of this design rule, these operations are left out of scope.
API-01: Adhere to HTTP safety and idempotency semantics for operations
The HTTP protocol [rfc7231] specifies whether an HTTP method should be considered safe and/or idempotent. These characteristics are important for clients and middleware applications, because they should be taken into account when implementing caching and fault tolerance strategies.
Request methods are considered safe if their defined semantics are essentially read-only; i.e., the client does not request, and does not expect, any state change on the origin server as a result of applying a safe method to a target resource. A request method is considered idempotent if the intended effect on the server of multiple identical requests with that method is the same as the effect for a single such request.
The following table describes which HTTP methods must behave as safe and/or idempotent:
One of the key constraints of the REST architectural style is stateless communication between client and server. It means that every request from client to server must contain all of the information necessary to understand the request. The server cannot take advantage of any stored session context on the server as it didn’t memorize previous requests. Session state must therefore reside entirely on the client.
To properly understand this constraint, it's important to make a distinction between two different kinds of state:
It's a misconception that there should be no state at all. The stateless communication constraint should be seen from the server's point of view and states that the server should not be aware of any session state.
Stateless communication offers many advantages, including:
API-02: Do not maintain session state on the server
In the context of REST APIs, the server must not maintain or require any notion of the functionality of the client application and the corresponding end user interactions. To achieve full decoupling between client and server, and to benefit from the advantages mentioned above, no session state must reside on the server. Session state must therefore reside entirely on the client.
The client of a REST API could be a variety of applications such as a browser application, a mobile or desktop application and even another server serving as a backend component for another client. REST APIs should therefore be completely client-agnostic.
Resources are often interconnected by relationships. Relationships can be modelled in different ways depending on the cardinality, semantics and more importantly, the use cases and access patterns the REST API needs to support.
API-06: Use nested URIs for child resources
When having a child resource which can only exist in the context of a parent resource, the URI should be nested. In that case, the child resource does not necessarily have a top-level collection resource. The best way to explain this design rule is by example.
When modelling resources for a news platform including the ability for users to write comments, it might be a good strategy to model the collection resources hierarchically:
The platform might also offer a photo section, where the same commenting functionality is offered. In the same way as for articles, the corresponding sub-collection resource might be published at:
These nested sub-collection resources can be used to post a new comment (
POST method) and to retrieve a list of comments (
GET method) belonging to the parent resource, i.e. the article or photo. An important consideration is that these comments could never have existed without the existence of the parent resource.
From the consumer's perspective, this approach makes logical sense, because the most obvious use case is to show comments below the parent article or photo (e.g. on the same web page) including the possibility to paginate through the comments. The process of posting a comment is separate from the process of publishing a new article. Another client use case might also be to show a global latest comments section in the sidebar. For this use case, an additional resource could be provided:
If this would have not been a meaningful use case, this resource should not exist at all. Because it doesn't make sense to post a new comment from a global context, this resource would be read-only (only
GET method is supported) and may possibly provide a more compact representation than the parent-specific sub-collections.
The singular resources for comments, referenced from all 3 collections, could still be modelled on a higher level to avoid deep nesting of URIs (which might increase complexity or problems due to the URI length):
Although this approach might seem counterintuitive from a technical perspective (we simply could have modelled a single
/comments resource with optional filters for article and photo) and might introduce partially redundant functionality, it makes perfect sense from the perspective of the consumer, which increases developer experience.
API-10: Model resource operations as a sub-resource or dedicated resource
There are resource operations which might not seem to fit well in the CRUD interaction model. For example, approving of a submission or notifying a customer. Depending on the type of the operation, there are three possible approaches:
goedgekeurdthat can be modified by issuing a
PATCHrequest against the resource. Drawback of this approach is that the resource does not contain any metadata about the operation (when and by whom was the approval given? Was the submission declined in an earlier stage?). Furthermore, this requires a fine-grained authorization model, since approval might require a specific role.
/inzendingen/12/beoordelingenand add an approval or declination by issuing a
POSTrequest. To be able to retrieve the review history (and to consistently adhere to the REST principles), also support the
GETmethod for this resource. The
/inzendingen/12resource might still provide a
goedgekeurdboolean attribute (same as approach 1) which gets automatically updated on the background after adding a review. This attribute should however be read-only.
/_search. Depending on the operation characteristics,
POSTmethod may be supported for such a resource.
In this design rule, approach 2 and 3 are preferred.
An API is as good as the accompanying documentation. The documentation has to be easily findable, searchable and publicly accessible. Most developers will first read the documentation before they start implementing. Hiding the technical documentation in PDF documents and/or behind a login creates a barrier for both developers and search engines.
API-16: Use OpenAPI Specification for documentation
The OpenAPI Specification (OAS) [OPENAPIS] defines a standard, language-agnostic interface to RESTful APIs which allows both humans and computers to discover and understand the capabilities of the service without access to source code, documentation, or through network traffic inspection. When properly defined, a consumer can understand and interact with the remote service with a minimal amount of implementation logic.
API documentation must be provided in the form of an OpenAPI definition document which conforms to the OpenAPI Specification (from v3 onwards). As a result, a variety of tools can be used to render the documentation (e.g. Swagger UI or ReDoc) or automate tasks such as testing or code generation. The OAS document should provide clear descriptions and examples.
API-17: Publish documentation in Dutch unless there is existing documentation in English
In line with design rule API-04, the OAS document (e.g. descriptions and examples) should be written in Dutch. If relevant, you may refer to existing documentation written in English.
API-51: Publish OAS document at a standard location in JSON-format
To make the OAS document easy to find and to facilitate self-discovering clients, there should be one standard location where the OAS document is available for download. Clients (such as Swagger UI or ReDoc) must be able to retrieve the document without having to authenticate. Furthermore, the CORS policy for this URI must allow external domains to read the documentation from a browser environment.
The standard location for the OAS document is a URI called
openapi.yaml within the base path of the API. This can be convenient, because OAS document updates can easily become part of the CI/CD process.
At least the JSON format must be supported. When having multiple (major) versions of an API, every API should provide its own OAS document(s).
An API having base path
https://api.example.org/v1/ must publish the OAS document at:
Optionally, the same OAS document may be provided in YAML format:
Changes in APIs are inevitable. APIs should therefore always be versioned, facilitating the transition between changes.
API-56: Adhere to the Semantic Versioning model when releasing API changes
Version numbering must follow the Semantic Versioning [SemVer] model to prevent breaking changes when releasing new API versions. Versions are formatted using the
major.minor.patch template. When releasing a new version which contains backwards-incompatible changes, a new major version must be released. Minor and patch releases may only contain backwards compatible changes (e.g. the addition of an endpoint or an optional attribute).
API-20: Include the major version number in the URI
The URI of an API (base path) must include the major version number, prefixed by the letter
v. This allows the exploration of multiple versions of an API in the browser. The minor and patch version numbers are not part of the URI and may not have any impact on existing client implementations.
An example of a base path for an API with current version 1.0.2:
API-57: Return the full version number in a response header
Since the URI only contains the major version, it's useful to provide the full version number in the response headers for every API call. This information could then be used for logging, debugging or auditing purposes. In cases where an intermediate networking component returns an error response (e.g. a reverse proxy enforcing access policies), the version number may be omitted.
The version number must be returned in an HTTP response header named
API-Version (case-insensitive) and should not be prefixed.
An example of an API version response header:
API-55: Publish a changelog for API changes between versions
When releasing new (major, minor or patch) versions, all API changes must be documented properly in a publicly available changelog.
API-18: Include a deprecation schedule when deprecating features or versions
Managing change is important. In general, well documented and timely communicated deprecation schedules are the most important for API users. When deprecating features or versions, a deprecation schedule must be published. This document should be published on a public web page. Furthermore, active clients should be informed by e-mail once the schedule has been updated or when versions have reached end-of-life.
API-19: Schedule a fixed transition period for a new major API version
When releasing a new major API version, the old version must remain available for a limited and fixed deprecation period. Offering a deprecation period allows clients to carefully plan and execute the migration from the old to the new API version, as long as they do this prior to the end of the deprecation period. A maximum of 2 major API versions may be published concurrently.
A resource is the key abstraction of information, where every piece of information is identified by a globally unique URI.
A singular resource is a resource describing a single thing (e.g. a building, person or event).
A collection resource is a resource describing multiple things (e.g. a list of buildings).
A URI [rfc3986] (Uniform Resource Identifier) is a globally unique identifier for a resource.