This library creates a JSON API-compliant REST API from your Node app and automatically generates API documentation.
It currently integrates with Express or Koa apps that use Mongoose models, but it can easily be integrated with other frameworks and databases. If you want to see an integration with another stack, just open an issue!
This library implements all the required portions of the 1.0 spec, which is more than enough for basic CRUD. It does not yet implement some of the smaller, optional pieces, like related resource URIs.
$ npm install json-api
Check out the full, working v3 example repo for all the details on building an API with this library. Or, take a look at the basic example below:
var app = require('express')()
, API = require('json-api');
var models = {
"Person": require('./models/person'),
"Place": require('./models/place')
};
var adapter = new API.dbAdapters.Mongoose(models);
var registry = new API.ResourceTypeRegistry({
"people": {
beforeRender: function(resource, req, res) {
if(!userIsAdmin(req)) resource.removeAttr("password");
return resource;
}
},
"places": {}
}, {
"dbAdapter": adapter,
"urlTemplates": {
"self": "/{type}/{id}"
}
});
// Tell the lib the host name our API is served from; needed for security.
const opts = { host: 'example.com' };
// Set up a front controller, passing it controllers that'll be used
// to handle requests for API resources and for the auto-generated docs.
var Front = new API.httpStrategies.Express(
new API.controllers.API(registry),
new API.controllers.Documentation(registry, {name: 'Example API'})
);
// Render the docs at /
app.get("/", Front.docsRequest);
// Add routes for basic list, read, create, update, delete operations
app.get("/:type(people|places)", Front.apiRequest);
app.get("/:type(people|places)/:id", Front.apiRequest);
app.post("/:type(people|places)", Front.apiRequest);
app.patch("/:type(people|places)/:id", Front.apiRequest);
app.delete("/:type(people|places)/:id", Front.apiRequest);
// Add routes for adding to, removing from, or updating resource relationships
app.post("/:type(people|places)/:id/relationships/:relationship", Front.apiRequest);
app.patch("/:type(people|places)/:id/relationships/:relationship", Front.apiRequest);
app.delete("/:type(people|places)/:id/relationships/:relationship", Front.apiRequest);
app.listen(3000);
The JSON-API spec is built around the idea of typed resource collections. For example, you can have a "people"
collection and a "companies"
collection. (By convention, type names are plural, lowercase, and dasherized.)
To use this library, you describe the special behavior (if any) that resources of each type should have, and then register those descriptions with a central ResourceTypeRegistry
. Then the library takes care of the rest. Each resource type description is simply an object with the following properties:
-
urlTemplates
: an object containing url templates used to output the json for resources of the type being described. Currently, the supported keys are:"self"
, which defines the template used for building resource urls of that type, and"relationship"
, which defines the template that will be used for relationship urls. -
dbAdapter
: the database adapter to use to find and update these resources. By specifying this for each resource type, different resource types can live in different kinds of databases. -
beforeRender
(optional): a function called on each resource after it's found by the adapter but before it's sent to the client. This lets you do things like hide fields that some users aren't authorized to see. Usage details. -
beforeSave
(optional): a function called on each resource provided by the client (i.e. in aPOST
orPATCH
request) before it's sent to the adapter for saving. You can transform the data here as necessary or pre-emptively reject the request. Usage details. -
parentType
(optional): this allows you to designate one resource type being a sub-type of another (itsparentType
). This is often used when you have two resource types that live in the same database table/collection, and their type is determined with a discriminator key. See theschools
type in the example repository. If a resource has aparentType
, it inherits that type's configuration (i.e. itsurlTemplates
,beforeSave
/beforeRender
functions,info
, etc), but can override those configuration settings too. -
info
(optional): this allows you to provide extra information about the resource that will be included in the documentation. Available properties are"description"
(a string describing what resources of this type are) and"fields"
."fields"
holds an object in which you can describe each field in the resource (e.g. listing validation rules). See the example implemenation for more details.
When a request comes in, the json-api library extracts various parameters from it to build a query that will be used to fulfill the user's request.
However, to support advanced use cases, you might want to override how the library generates this query in order to select/update different data, or to modify how the query's result (data or error) is placed into the JSON:API response. To do this, you can just pass in your own function (a "query factory") for constructing this query. See an example here.
One simple thing you can do with query factories is to create urls (or, in REST terminology, resources) that map to different database items over time. For example, you could have an /events/upcoming
resource or a /users/me
resource. To do that, your query factory function would call the library's built-in function to get its auto-generated query, and then modifiy that query (which would likely be for all the events and users respectively) to add a filter constraint that only returns the appropriate resources.
Query factory functions are also a good place to do permissions checks that rely on access to the parsed request body. If the user doesn't have permission, you can throw an error, and the library will pass it along gracefully in the response. See error handling.
This library supports filtering out of the box, using a syntax that's designed to be easier to read, easier to write, and more expressive than the square bracket syntax used by libraries like qs.
For example, to include only items where the zip code is either 90210 or 10012, you'd write:
?filter=(zip,:in,[90210,10012])
.
By contrast, with the square-bracket syntax, you'd have to write something like:
?filter[zip][in][]=90210&filter[zip][in][]=10012
.
Also, the square-bracket syntax can't represent empty arrays or distinguish between non-string literals (e.g. true
) and strings, while this library's format can. See details below.
In this library's default format, the value of the filter
parameter is one or more "filter constraints" listed next to each other. These constraints narrow the results to only include those that match. The format of a filter constraint is: (fieldName,:operator,value)
. For example:
(name,:eq,`Bob`)
: only include items where the name equals "Bob"(salary,:gte,150000)
: only include items where the salary is greater than or equal to 150,000.
The value can be a number, null
, true
, or false
, or a backtick-delimited string (like `Bob`
). To define a list of values, surround the values in square brackets and separate them with commas (e.g. [90210,10012]
is a list of values used with the in
operator above).
The valid operators (for the buit-in Mongoose adapter) are: eq
, neq
, in
, nin
, lt
, gt
, lte
, and gte
.
If you have multiple constraints, you can choose whether to combine them with an AND
or an OR
. To do that, instead of providing three items in your field constraint (i.e., the field name, operator, and value), provide and
or or
as the opertor, followed by the applicable constraints. E.g.:
GET /people?filter=(:or,(name,:eq,`Bob`),(zip,:eq,90210))
Will find all the people who are named Bob or who live in the 90210 zip code.
Filter constraints listed next to each other at the top-level are combined with an "AND". E.g., GET /people?filter=(name,`Bob`)(zip,90210)
will give only the people named Bob who live in the 90210 zip code.
The (name,`Bob`)
constraint above is euivlent to (name,:eq,`Bob`)
. This is a shorthand. Whenever you don't provide an operator, eq
will be inferred.
Putting it all together, you could do:
GET /people?filter=(:or,(email,`[email protected]`),(name,`Test`))(dob,:gte,1963)
This will find everyone born after 1963 who also has either the name "Test" or the email "[email protected]".
Note: your API can use a totally different filter query parameter format if you so desire, by providing a custom parser (see example). Also, each adapter can indicate support for a custom set of operators.
When sending filter constraints, make sure you don't URL encode characters that the syntax above uses as delimiters (namely, commas, parentheses, backticks, square brackets, and the exclamation point), unless you mean for these characters to be interpreted as part of your data (e.g., part of a field name or value) rather than as a separator.
Be aware that some clients, by default, automatically percent-encode certain characters in url query strings -- especially the backtick and square brackets. This will cause the server to error, because it won't see the appropriate (unescaped) delimiter characters. Whether the client is correct to do this automatic encoding is a nuanced question, as there are competing and conflicting standards governing URLs today (namely, RFC 3986 and WHATWG Url). If you encounter this problem, check your client for a way to make requests without any automatic encoding. If your client absolutely insists on URL encoding backticks (the trickiest character), you can delimit your strings with exclamation points (!
) instead, and then encode exclamation points within the string.
Pagination limit and offset parameters are accepted in the following form:
?page[offset]=<value>&page[limit]=<value>
For example, to get 25 people starting at the 51st person:
GET /people?page[offset]=50&page[limit]=25
Code that you provide the library (e.g., a query factory function) can throw an error or, in some cases, return a promise that rejects with an error.
Any time an Error
is encountered, the library responds with a generic "An error has ocurred" response. The library doesn't pass any information from the error object (e.g., its message or stack trace) back to the client by default, as that could leak details about the server's implementation, which is not great for security.
If you want to pass information about the error back to the user, you need to explicitly mark it as safe to expose to the user, which you can do in two ways. First, you can throw an instance of the library's APIError
class directly, instead of a more general Error
. (See the APIError constructor and the json:api spec on errors for details about what properties you can expose.) Second, if your error is coming from some existing code, you can attach a special, Symbol-named property to it with a truthy value, and that'll trigger the framework to know that it's safe to display.
Many errors are also serialized by the library with standard URIs in their code
field, which serves as a reliable way to identify each type of error. The library exports an Errors object, which contains a collection of functions for making errors of known types with these codes already set in them. If you throw errors, you should reuse these functions/codes where possible. To detect an APIError's type in your code, call .toJSON()
and read the resulting .code
property.
This library gives you a Front controller (shown in the example) that can handle requests for API results or for the documentation. But the library doesn't prescribe how requests get to this controller. This allows you to use any url scheme, routing layer, or authentication system you already have in place.
You just need to make sure that: req.params.type
reflects the requested resource type; req.params.id
reflects the requested id, if any; and req.params.relationship
reflects the relationship name, in the event that the user is requesting a relationship url. The library will read these values to help it construct the query needed to fulfill the user's request.
The library may, in the future, also read req.params.related
for related resource urls, so make sure you're not using that if you don't want the library to pick up it's value and use it in the query.
In the example above, routing is handled with Express's built-in app[VERB]
methods, and the three parameters are set properly using express's built-in :param
syntax. If you're looking for something more robust, you might be interested in Express Simple Router. For authentication, check out Express Simple Firewall.
An adapter handles all the interaction with the database. It is responsible for turning requests into standard Resource
or Collection
objects that the rest of the library will use. See the built-in MongooseAdapter for an example.