apiman
v1.0.1
Published
Generic API methods manager
Downloads
25
Readme
ApiMan
ApiMan is the API methods manager that is exportable to multiple protocols, including REST via Express.
The Motivation
When your app needs a REST API - Express is a great choice, but imagine you need to support multiple protocols at the same time and want to have the code organized. Faking requests for Express is a tricky thing that is not guaranteed to function as it progresses...
ApiMan steps in: you define a tree of resources with named methods bound to them, and now just bind it to Express as a middleware. Wait, some methods should also be available through socket.io? No problem.
Now, we want some middleware for data preparation and authentication? Yes, we support that.
Enjoy it, guys :)
Core Components
Resource, Root
A resource is a collection of methods and sub-resources identified by path. It also keeps the related information: parameters info, middleware etc.
You create a sub-resource by calling the Resource.resource(path)
method of
a parent Resource
or the Root
container:
var root = new apiman.Root();
var user = root.resource('/user');
var user_profile = user.resource('/profile');
The Root
is actually a resource with empty path.
Although we follow the HTTP-style slash-separated paths, you're free to use any convention you're comfortable with.
Method
After you have a hierarchy of resources, you can define methods on each, including the root container.
A Method
is defined with the Resource.method(verbs, ...callbacks)
method of
a Resource
.
verbs
is the name of the method, or, optionally, an array of them.
After the verb
, you specify a callback to be executed when the method matches
the request:
user_profile.method('set', function(req, res){
save_to_db(
req.args['user'],
function(err, id){
if (err)
res.error(err);
else
res.ok({saved: true, id: id});
}
);
});
The method callback accepts two arguments: the Request
and Response
objects.
Request
The Request
object has the following useful properties:
req.path
is the full path to the current resource:'/user/profile'
req.verb
is the current verb than made the method match:'set'
req.args
is an object of method arguments:{ user: {login: 'kolypto', ...} }
req.path_array
is an array of path components split on a resource match:['/user', '/profile']
req.params
is an object of parameters from RegExps on path (see below).{ uid: 10 }
And also some internal informational fields:
req.middleware
is an array of middleware assigned to this very request.req.response
is theResponse
object shortcut used internally
Response
The Response
object is a naive wrapper for a NodeJS-style
function(err,result)
callback and has the following methods:
Response.send(err, result)
is the generic callback with both optionsResponse.error(err)
is the callback for errors that wrapsResponse.send(err)
Response.ok(result)
is the callback for results that wrapsResponse.send(undefined, result)
Middleware
Method middleware
Like in Express, each method can use an arbitrary list of middleware callbacks before the method function:
// middleware to check the permissions
var accessCheck = function(req, res, next){
if (req.args['uid'] != 10) // stupid access check
next(new Error('Access denied')); // error
else
next(); // proceed
};
user_profile.method('get', accessCheck, function(req, res){
load_from_db(function(err, user){
res.send(err, user); // delegate both arguments to the response handler
});
});
Now, the method function is only executed once all preceding middleware
callbacks have called next()
with no arguments, which indicates success.
Resource middleware
Additionally, a middleware can be attached to a Resource
: it will be executed
for all requests to its methods or methods of the sub-resources:
user_profile.use(function(req, res, next){
if (req.args['uid'] === undefined)
next(new Error('Missing required argument: uid'));
else
next();
});
Parameters
Resource paths can be specified as regular expressions, just don't forget to anchor them to the start of the string. As RegExps can capture parts of the input, I could't resist to not add the parameters support:
var device_commands = root.resource(new RegExp('/device/(\w+)/command/(\w+)'))
.param(1, 'device_type')
.param(2, 'command', function(req, res, next, value){
if (['start', 'stop'].indexOf(value) == -1)
next(new Error('Unsupported command'));
else {
req.params['command'] = value;
next();
}
})
.method('invoke', function(req, res){
});
Parameters are defined as simple capture groups in a RegExp. To have named
params, you use the Resource.param(index, [, callback])
Resource method
which maps a group to a middleware invocation:
index
is the positional index of the capture groupcallback
is the middleware that alters theRequest
object using the parameter value:function(req, res, next, value)
.
To have named parameters, you typically place them in the Request.params
object designed for that.
Merging Resources
For modularity, you might want to distribute your resources across different
files and then merge with with the Resource.merge(resource, ...)
method:
- Adds all methods from the resources to the current one
- Adds all middleware from the resources
- Adds all sub-resources to the current one
- If a resource would have been overwritten, it's merged.
// Module
var module = new apiman.Root();
module.resource('/user')
.method('get', function(req, res){ /* ... */})
// Extension
var extension = new apiman.Root();
module.resource('/user')
.method('command', function(req, res){ /* ... */})
// Index file
var api = new apiman.Root();
api.merge(module, extension);
The example above results in a tree with a single /user
Resource which has
two methods defined: get
and command
.
Executing methods
To execute a method of your API root, use the
Resource.request(path, verb, args[, req], callback)
method:
path
is the path to some resource within the treeverb
is the name of the method to executeargs
is the arguments object for the method. Optional.req
is an object with extended request fields. Optional. Useful to populate additionalRequest
fields at the invocation time: say, user session.callback
accepts the method output:function(err, result)
.
Resource.request()
does the following:
- Creates the
Request
andResponse
object - Traverses the tree using a prefix match technique and gets down to the matching Resource
- All middleware added to resources down the path are scheduled for the request
- Any parameter callbacks down the path are also scheduled
- Picks a method by
verb
- Executes all collected middleware
- Executes the method middleware
- Executes the method
- Fires the callback
If a resource or method is not found, the function returns false
.
Matching
In the examples above we follow the REST naming conventions for clarity, but again, that is not required.
Given a path, ApiMan performs a case-sensitive exact prefix matching. For instance, given the following resources chain:
var root = new apiman.Root();
root.resource('/user')
.resource('/device/commands')
.resource('/private');
path '/user/device/commands/private'
recursively matches each resource by
prefix: '/user'
, '/device/commands'
, '/private'
.
Don't expect ApiMan to forgive extra or missing slashes: it's protocol-agnostic by design and, potentially, all special characters might have a meaning.
Anyway, nothing prevents you from making a preprocessor which tunes the input to your taste:
// Ensure a leading slash, no trailing slash, and collapse duplicate slashes
path = ('/' + path).replace(/\/+/g, '/').replace(/\/$/, '');
Exporting the API
socket.io
Piece of cake: as socket.io can exchange json objects, you just need a handy convention for sending requests and getting responses.
The only difficulty is that socket.io does not support the request-response protocol out of the box, but we can easily overcome that by numbering the packets.
Given the above, let's use the following data exchange protocol:
- Request:
{{ id: Number, path: String, verb: String, args: Object }}
- Response:
{{ id: Number, data: [ undefined, Object ] }}
- Error:
{{ id: Number, data: [ String|Error, undefined ] }}
On the server:
io.sockets.on('connection', function (socket) {
socket.on('api', function (data) {
root.request(data.path, data.verb, data.args, function(err, result){
// Emit the result using the same method id
socket.emit('api.result', {
id: data.id,
ret: [err, result]
});
}) ||
socket.emit('api.result', {
id: data.id,
ret: ['unknown method', undefined]
});
});
});
And on the client:
io_method = function(path, verb, args, callback){
var request = {
id: io_method._id++, // packet id
path: path,
verb: verb,
args: args
};
io_method._wait[request.id] = callback;
socket.emit('api', request);
};
io_method._id=0;
io_method._wait = {};
// Listen for responses
socket.on('api.result', function(data){
io_method._wait[data.id].apply(null, data.ret);
});
This approach, however, has 2 weak points:
- On reconnect, the response can't be received transparently
- The exposed error objects can potentially contain sensitive data like stack traces
Express
Assume you already have your API defined under the root
variable, and now it's
time to export it to Express. There are a couple of things to take care of:
- Map your resources and methods to paths
- Format the output for responses
- Decide on the HTTP status code for errors
If your resources & methods (expecially their verbs) are directly exportable to Express and compatible with REST, you're lucky:
app.use('/api', function(req, res){
var path = req.path,
args = _(req.body).extend(req.query), // combine
verb = req.method,
apireq = {} // additional fields for Request
;
// Pass the request to ApiMan
var found = root.request(path, verb, args, apireq, function(err, result){
// Format the output
if (err)
res.type('json').send(err.httpCode || 500, { error: err.message });
else
res.type('json').send(result);
});
// Method not found
if (!found)
res.type('json').send(404, { error: 'Unknown API method' });
});
The only issue that remains is that all error codes are 400
: we don't
differentiate server errors, client errors and stuff. To overcome that, you'd
need a convention:
- Always return an error object with a custom HTTP status code set. Default to 500 for other cases (all other errors)
- Create a hierarchy of custom
Error
objects with an http status code defined on each, and return them.
Complex mappings
ApiMan supports a richer methods collection interface which's not limited to
HTTP methods: as an example, imagine a /user
resource with methods
load
, save
, del
, block
, list
. While for CRUD methods you can just
map the HTTP verbs (GET
-> load
), the block
and list
method would have
required sub-resources and/or query strings.
That's what you need the mappers for.
First, change your Express middleware a little to enable mappers for 'express' on the request:
// Tell ApiMan we're from Express
root.requestFrom('express', path, verb, args, req, function(err, result){
/* ...*/
});
In order for the magic to work for us, we need to declare mappers on non-exportable resources which routes the REST requests to ApiMan methods.
Observe the example:
var user = root.resource('/user');
user.method('load', function(req, res){/*...*/});
user.method('save', function(req, res){/*...*/});
user.method('del', function(req, res){/*...*/});
user.method('block', function(req, res){/*...*/});
user.method('list', function(req, res){/*...*/});
user.map('express', function(path, verb){
// Trick the incoming (path,verb)
switch (path){
case '': // endpoint
return [
path,
// Change the verb
{GET: 'load', POST: 'save', DELETE: 'del'}[verb]
];
case '/list': // fake path
return ['', 'list']; // route to the method
case '/block':
return ['', 'block'];
}
return undefined; // unchanged
});
The mapper function can be defined on any resource and is invoked when the
resource tree is traversed. It accepts the (path,verb)
pair, where path
is
the current path remainder with all matched prefixes already truncated. It's
expected to return an altered [path,verb]
pair sufficient for the subsequent
resource/method lookup to succeed.
As usually simple path/verb mapping is enough, you can save a callback and give a mapping instead:
user.map('express', {
'': ['', {GET: 'load', POST: 'save', DELETE: 'del'}]
'/list': ['', 'list'],
'/block': ['', 'block'],
});
The mapper will search for the path remainder in the object keys. If the value
is an array - it's taken as a [path,verb]
pair, where the verb can be
specified as a mapping.