http-api-guidelines
http-api-guidelines copied to clipboard
Rest HTTP API Guidelines that I have gathered while working on number of APIs through the years.
HTTP API Guidelines
RESTful consistency
I love consistency and one of the things that frustrate me is when number of devs working on the project make up their own standards. It is especially noticeable in the APIs. Many of the APIs start as internal project and at some point get exposed to the world. Once API is exposed, it is very difficult to change; its important to get a good start.
I have built number of HTTP APIs over the years and in fact its one of my favourite things to do. I have used various formats to represent data and have gathered number of guidelines that can help develop clean, easy to consume RESTful api.
API (Application Programming Interface)
My philosophy of the API is that it should be very easy to consume and must be human readable. API is there for computers to communicate, but in the end it will be built and maintained by developers.
REST
REST (Representational state transfer) is yet another buzzword that has taken over IT world. I think of REST API as pure HTTP API that allows programmatic navigation through hyperlinks.
HTTP Headers
Use HTTP Headers to represent API metadata. Headers should be used to specify Media Type, Caching Strategy, Accepted formats and API version, Resource Location, Encoding. Please, do not use headers to represent the data. There is payload for that.
HTTP Payload
There is where data goes. XML is widely used, but my preference is to use JSON to represent the data. JSON is very light weight and very human readable. JSON can also be very easily consumed by clients running in the browser. Please avoid any kind of envelopes, like SOAP for example. Use headers to represent metadata.
API Version
This is very touchy subject as there are number of preferred ways to deal with versions of the API and backwards compatibility.
Couple of common ways to version APIs:
- version in the URI /v1/resource/123
- version as request param /resource/123?v=1
- use Media Type:
- in the media type name
- added as qualifier
Use Media Type headers to represent resource version
I think using Media type is the most appropriate way where version is added as qualifier. URL should contain resource location info only, while header represents the version.
Accept: application/vnd.hostname.userprofile+json;v=1
Content-Type: application/vnd.hostname.userprofile+json;v=1
Accept: application/vnd.hostname.user+json;v=1.2
Content-Type: application/vnd.hostname.user+json;v=1.2
As you can see, client can specify the version required in the Accept header and server specifies the version served in content type header.
Note that Media Type is specified as custom vnd
(vendor) type with format
in json
.
This version method is sometimes difficult to implement as it may require networking equipment to operate on and parse HTTP headers instead of URL if different versions are served from different servers or network segments.
Hostname
Use hostname (incl sub name) to represent the domain identifier.
{domain}-api.hostname.com
user-api.hostname.com
profile-api.hostname.com
user-preference.api.hostname.com
Note: domain word identifies your domain concept, use - to separate if domain concept consists of more then 1 word.
Resource names
Resource names must be nouns (with some exceptions, like search, for example). Resource names should represent your business domain model.
Use plural to represent collections.
{domain}.api.hostname.com/resource-collection/{resource-id}/resource-collection/{resource-id}
user.api.hostname.com/users/1234/profiles/52342
user.api.hostname.com/users/1234/account
URI (Unique Resource Identifier)
Do not include query params as part of resource identifier as they are optional as per HTTP specification.
{domain}.api.hostname.com/resource-collection/{resource-id}/resource-collection/{resource-id}
{domain}.api.hostname.com/resource-collection/{resource-id}
Format
Type | Standard | Format | Examples |
---|---|---|---|
Resource Name | hyphen-separated, lower case | /users, /user-preferences | |
Query Parameters | camelCase string (low 1st) | ?sortBy=name&order=desc | |
Headers | HTTP 1.1 | Snake-Case (hyphen, 1st upper | Content-Type, If-Match |
Media Type | HTTP 1.1 | application/vnd.hostname.{resourcetype}+json;v=version | application/vnd.hostname.user+json;v=2.1 |
Date | ISO 8601 | YYYY-MM-DD | 2015-11-14 |
Date Time | ISO 8601 (UTC) | YYYY-MM-DDThh:mm:ssZ (Z for UTC, no other timezones) | 2015-11-14T21:48:13Z |
Duration | ISO 8601 | PnYnMnDTnHnMnS | "P23DT23H" and "P4Y" are both acceptable duration representations as all values are optional and can be omitted if value is 0 |
Currency Code | ISO 4217 | 3 alpha (Caps) | USD, EUR, CAD, AUD |
Currency Amount | Number | 45.99, 2003.56, 5000 | |
Country Code | ISO 3166-1 | 2 alpha (Caps) | US, CA, CH, KZ, UA |
Region Code | ISO 3166-2 | CountryCode-RegionCode (2 alpha) | CA-ON, CA-AB, UA-71, US-CA |
Language Code | ISO 6391 + ISO 6393 (for dialects not covered by 6391) | 2 alpha or 3 alpha | en, es, it, hi, zh, cmn (chinese mandarin), cjy (chinese jin) |
Airport Code | IATA | 3 alpha (Caps) | YYZ,LAX,AMS |
Request
HTTP Methods
- POST - create, can only be done on collection (/users)
POST /users
POST /users/123/accounts
POST /users/123/orders
- PUT/PATCH - update/partial update, can only be done on existing resource URI, idempotent
PUT /users/123
PUT /users/123/orders/456
- GET - idempotent
GET /users/123
GET /users
GET /users?lastName=Smith&yearOfBirth=1970
GET /users/123/orders/456
GET /users/123/orders?createdStartDate=2014-05-14T13:00:00Z&createdEndDate=2014-05-14T18:00:00Z
GET /users/123/orders
- DELETE - idempotent
DELETE /users/123
DELETE /users/123/orders
DELETE /users/123/orders/456
- OPTIONS - to see what methods are available on the resouces
Response
Status Codes
- Use one of 200s to represent success
- Use 201 for successful POST with Location header
- Use 202 for asynchronous interactions
- Use others appropriately as per spec
- Use one of 300s to represent redirections
- Use others appropriately as per spec
- Use one of 400s to represent user error. User MUST make changes to the request and try again.
- Use 400 for malformed payload (malformed JSON)
- Use 401 to communicate that client is unauthorized
- Use 403 for invalid permissions (read vs write)
- Use 422 for valid JSON/XML but invalid semantic errors, like invalid airport code, currency code or invalid date precedence.
- Use others appropriately as per HTTP spec
- Use one of 500s to represent server error. Users should not retry requests with 500 responses.
- 502 - to represent failure in the upstream response (if our service is consuming another service that is down).
- Use other appropriately as per spec
Error Response
Header:
Content-Type: application/vnd.hostname.errors+json;v=1
Payload Body
{
"errors": [
{
"code": "mandatory|error code |snake case, all caps|USER_NOT_FOUND",
"description": "mandatory|verbose description of the error that developer will use for troubleshooting, user will not see",
"field": "optional|field name, shown during validation errors for specific fields"
}
]
}
Example
{
"errors": [
{
"code": "USER_NOT_FOUND",
"description": "user with registrationDate=2014-05-14T15:22:11Z and lastLoginDate=2014-05-16T15:22:11Z cannot be found."
}
]
}
“code” field in the error response will be used to for specifying exactly what happened and may be used for driving UI logic. Code field should be self descriptive with all capital letters separated by the underscore (USER_NOT_FOUND, INVALID_LAST_NAME)
There are 2 types of codes:
- HTTP Status codes, integers
- client can get these from HTTP request
- Custom codes that go in addition to HTTP Status codes - Error object, Code field
- These are Enum values, constants.
- Client will get these from Error object in the response
- error.code is generally required in addition to HTTP status code to precisely identify the issue.
- These are also checked in addition to HTTP status codes against to determine Error message/Error screen to display for the user.
So in example where POST /users request is made to the server and payload contains invalid email (johngmail.com as an example)
From API perspective you will return:
- HTTP 422 - as it is semantic error where serve fully understands the request (don't use 400 in this case) JSON payload with header Content-Type: application/vnd.hostname.errors+json;v=1
[
{
"code": "INVALID_EMAIL",
"description": "email johngmail.com is invalid, its missing @ character",
"field": "email"
}
]
As you can see, from HTTP 422 it's impossible to determine that there is a problem with the email, this is when the code and field combination is used to determine the action on the UI side.
Media Type
use Accept
and Content-Type
http headers to specify custom media type.
Encoding
- Use UTF-8
- JSON is UTF-8 by default
Retrieval and Search
specific resource by id
/users/12345
collection of resources by attributes
/users?countryCode=US
collection of resources by range
use startXXX and endXXX query parameters
/users?registrationStartDate=2014-05-14T15:22:11Z®istrationEndDate=2014-05-14T18:30:00Z
Pagination
Offset and limit
It has it's drawbacks: http://blog.vermorel.com/journal/2015/5/8/nearly-all-web-apis-get-paging-wrong.html):
use offset and limit query parameters
- offset = 0, limit = 10 is 0 - 9
- offset = 10, limit = 10 is 10 - 19
/users?offset=10&limit=10
Continuation token
use continuation token query parameter
/users?continuationToken=token
token would be a reference for the last record that was returned (this requires either a 2 way hash for the token or some state on the server side).
Sorting
use sortBy and order query parameters
- sortBy - sort criteria with values comma separated
- order - specify sort order
?sortBy=X,Y,Z&order=asc|desc
?sortBy=FirstName,LastName&order=asc
Hyperlinking and HATEOAS
As per rfc5988 Link HTTP header should be used to define links.
According to Roy Fielding (principal author of HTTP):
What needs to be done to make the REST architectural style clear on the notion that hypertext is a constraint? In other words, if the engine of application state (and hence the API) is not being driven by hypertext, then it cannot be RESTful and cannot be a REST API. Period. Is there some broken manual somewhere that needs to be fixed?
Note: there are other common ways to do hyperlinking
- HAL
- ATOM
Resource Versioning
Use Etag headers as per HTTP to represent Resource version.
Cache Control
Use HTTP built in Cache mechanisms and Cache Control Headers as these are supported by wide varieties of servers, switches and other network devices. When we talk about Cache, I am talking about Caching HTTP Resources (like we cache all other HTTP resources: Images, CSS, JS, etc)
Optimistic Concurrency Control
Etags in combination with HTTP Cache Headers (e.g If-Match) should be used for Optimistic Concurrency Control
Security
- APIs exposed to the world require Authentication and Authorization
- APIs that contain security credentials and must be exposed over HTTPS
HTTP Basic
Http Header: Authorization
Http Header Value: Basic {base64 encoded username and password}
As per wiki page HTTP header constructed as follows:
- Username and password are combined into a string "username:password"
- The resulting string literal is then encoded using the RFC2045-MIME variant of Base64, except not limited to 76 char/line[9]
- The authorization method and a space i.e. "Basic " is then put before the encoded string.
HTTP Header Example:
Authorization: Basic QWxhZGRpbjpvcGVuIHNlc2FtZQ==
Gzip Compression
For the performance reason gzip compression is required to ensure payloads are transferred faster over the wire.