restify-clients

Getting Started

Install the module with: npm install restify-clients

Usage

Client API

There are three separate clients:

• JsonClient: sends and expects application/json
• StringClient: sends url-encoded request and expects text/plain
• HttpClient: thin wrapper over node's http/https libraries

The idea being that if you want to support "typical" control-plane REST APIs, you probably want the JsonClient, or if you're using some other serialization (like XML) you'd write your own client that extends the StringClient. If you need streaming support, you'll need to do some work on top of the HttpClient, as StringClient and friends buffer requests/responses.

All clients support retry with exponential backoff for getting a TCP connection; they do not perform retries on 5xx error codes like previous versions of the restify client. You can set retry to false to disable this logic altogether. Also, all clients support a connectTimeout field, which is use on each retry. The default is not to set a connectTimeout, so you end up with the node.js socket defaults.

Here's an example of hitting the Joyent CloudAPI:

var clients = require('restify-clients');

// Creates a JSON client
var client = clients.createJsonClient({
  url: 'https://us-east-1.api.joyent.com'
});


client.basicAuth('$login', '$password');
client.get('/my/machines', function(err, req, res, obj) {
  assert.ifError(err);

  console.log(JSON.stringify(obj, null, 2));
});

As a short-hand, a client can be initialized with a string-URL rather than an options object:

var clients = require('restify-clients');

var client = clients.createJsonClient('https://us-east-1.api.joyent.com');

Note that all further documentation refers to the "short-hand" form of methods like get/put/del which take a string path. You can also pass in an object to any of those methods with extra params (notably headers):

var options = {
  path: '/foo/bar',
  headers: {
    'x-foo': 'bar'
  },
  retry: {
    'retries': 0
  },
  agent: false
};

client.get(options, function(err, req, res) { .. });

If you need to interpose additional headers in the request before it is sent on to the server, you can provide a synchronous callback function as the signRequest option when creating a client. This is particularly useful with node-http-signature, which needs to attach a cryptographic signature of selected outgoing headers. If provided, this callback will be invoked with a single parameter: the outgoing http.ClientRequest object.

JsonClient

The JSON Client is the highest-level client bundled with restify; it exports a set of methods that map directly to HTTP verbs. All callbacks look like function(err, req, res, [obj]), where obj is optional, depending on if content was returned. HTTP status codes are not interpreted, so if the server returned 4xx or something with a JSON payload, obj will be the JSON payload. err however will be set if the server returned a status code >= 400 (it will be one of the restify HTTP errors). If obj looks like a RestError:

{
  "code": "FooError",
  "message": "some foo happened"
}

then err gets "upconverted" into a RestError for you. Otherwise it will be an HttpError.

createJsonClient(options)

var client = restify.createJsonClient({
  url: 'https://api.us-east-1.joyent.com',
  version: '*'
});

API Options:

Name	Type	Description
accept	String	Accept header to send
appendPath	Boolean	Append paths provided at verb time to existing client path
audit	Boolean	Enable Audit logging
auditor	Function	Function for Audit logging
connectTimeout	Number	Amount of time to wait for a socket
contentMd5	Object	How response content-md5 headers are handled. See the contentMd5 option to StringClient
contentType	String	Content-Type header to send
requestTimeout	Number	Amount of time to wait for the request to finish
dtrace	Object	node-dtrace-provider handle
gzip	Object	Will compress data when sent using content-encoding: gzip
headers	Object	HTTP headers to set in all requests
log	Object	bunyan instance
query	Object	querystring object to be serialized via querystring module
retry	Object	options to provide to node-retry;"false" disables retry; defaults to 4 retries
safeStringify	Boolean	Safely serialize JSON objects, i.e. circular dependencies
signRequest	Function	synchronous callback for interposing headers before request is sent
url	String	Fully-qualified URL to connect to
userAgent	String	user-agent string to use; restify inserts one, but you can override it
version	String	semver string to set the accept-version
followRedirects	Boolean	Follow redirects from server
maxRedirects	Number	Maximum number of redirects to follow
proxy	String	An HTTP proxy URL string (or parsed URL object) to use for requests. If not specified, then the https_proxy or http_proxy environment variables are used. Pass proxy: false to explicitly disable using a proxy (i.e. to ensure a proxy URL is not picked up from environment variables). See the Proxy section below.
noProxy	String	A comma-separated list of hosts for which to not use a proxy. If not specified, then then NO_PROXY environment variable is used. One can pass noProxy: '' to explicitly set this empty and ensure a possible environment variable is not used. See the Proxy section below.

get(path, callback)

Performs an HTTP get; if no payload was returned, obj defaults to {} for you (so you don't get a bunch of null pointer errors).

client.get('/foo/bar', function(err, req, res, obj) {
  assert.ifError(err);
  console.log('%j', obj);
});

head(path, callback)

Just like get, but without obj:

client.head('/foo/bar', function(err, req, res) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
});

post(path, object, callback)

Takes a complete object to serialize and send to the server.

client.post('/foo', { hello: 'world' }, function(err, req, res, obj) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
  console.log('%j', obj);
});

put(path, object, callback)

Just like post:

client.put('/foo', { hello: 'world' }, function(err, req, res, obj) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
  console.log('%j', obj);
});

del(path, callback)

del doesn't take content, since you know, it should't:

client.del('/foo/bar', function(err, req, res) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
});

StringClient

StringClient is what JsonClient is built on, and provides a base for you to write other buffering/parsing clients (like say an XML client). If you need to talk to some "raw" HTTP server, then StringClient is what you want, as it by default will provide you with content uploads in application/x-www-form-url-encoded and downloads as text/plain. To extend a StringClient, take a look at the source for JsonClient. Effectively, you extend it, and set the appropriate options in the constructor and implement a write (for put/post) and parse method (for all HTTP bodies), and that's it.

contentMd5

The contentMd5 option to StringClient specifies how content-md5 response headers are handled:

var contentMd5 = {
  ignore: false,
  encodings: undefined
};

By default if a server responds with a content-md5 header, StringClient will automatically verify the response body md5 hash, and return a BadDigest error when it does not match. You can disable this behaviour by setting the contentMd5.ignore option to true when creating the StringClient instance. You can also set the contentMd5.encodings property to be an array of encoding names that StringClient will use to verify the md5. The default encoding is "utf8" for node 6 and above, and "binary" (a latin1 encoding) for node 4 and older (as node changed the default crypto hash encoding between versions 4 and 6). To support content-md5 hashes generated by both node 4 and 6 servers, you can specify contentMd5.encodings = ["utf8", "binary"] when creating the StringClient instance.

createStringClient(options)

var client = restify.createStringClient({
  url: 'https://example.com'
});

get(path, callback)

Performs an HTTP get; if no payload was returned, data defaults to '' for you (so you don't get a bunch of null pointer errors).

client.get('/foo/bar', function(err, req, res, data) {
  assert.ifError(err);
  console.log('%s', data);
});

head(path, callback)

Just like get, but without data:

client.head('/foo/bar', function(err, req, res) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
});

post(path, object, callback)

Takes a complete object to serialize and send to the server.

client.post('/foo', { hello: 'world' }, function(err, req, res, data) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
  console.log('%s', data);
});

put(path, object, callback)

Just like post:

client.put('/foo', { hello: 'world' }, function(err, req, res, data) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
  console.log('%s', data);
});

del(path, callback)

del doesn't take content, since you know, it should't:

client.del('/foo/bar', function(err, req, res) {
  assert.ifError(err);
  console.log('%d -> %j', res.statusCode, res.headers);
});

HttpClient

HttpClient is the lowest-level client shipped in restify, and is basically just some sugar over the top of node's http/https modules (with HTTP methods like the other clients). It is useful if you want to stream with restify. Note that the event below is unfortunately named result and not response (because Event 'response' is already used).

client = restify.createClient({
  url: 'http://127.0.0.1'
});

client.get('/str/mcavage', function(err, req) {
  assert.ifError(err); // connection error

  req.on('result', function(err, res) {
    assert.ifError(err); // HTTP status code >= 400

    res.body = '';
    res.setEncoding('utf8');
    res.on('data', function(chunk) {
      res.body += chunk;
    });

    res.on('end', function() {
      console.log(res.body);
    });
  });
});

Or a write:

client.post(opts, function(err, req) {
  assert.ifError(connectErr);

  req.on('result', function(err, res) {
    assert.ifError(err);
    res.body = '';
    res.setEncoding('utf8');
    res.on('data', function(chunk) {
      res.body += chunk;
    });

    res.on('end', function() {
      console.log(res.body);
    });
  });

  req.write('hello world');
  req.end();
});

Note that get/head/del all call req.end() for you, so you can't write data over those. Otherwise, all the same methods exist as JsonClient/StringClient.

One wishing to extend the HttpClient should look at the internals and note that read and write probably need to be overridden.

Proxy

A restify client can use an HTTP proxy, either via options to createClient or via the http_proxy, https_proxy, and NO_PROXY environment variables common in many tools (e.g., curl).

restify.createClient({
  proxy: <proxy url string or object>,
  noProxy: <boolean>
});

The proxy option to createClient specifies the proxy URL, for example:

proxy: 'http://user:[email protected]:4321'

Or a proxy object can be given. (Warning: the proxyAuth field is not what a simple require('url').parse() will produce if your proxy URL has auth info.)

proxy: {
  protocol: 'http:',
  host: 'example.com',
  port: 4321,
  proxyAuth: 'user:password'
}

Or proxy: false can be given to explicitly disable using a proxy -- i.e. to ensure a proxy URL is not picked up from environment variables.

If not specified, then the following environment variables (in the given order) are used to pick up a proxy URL:

HTTPS_PROXY
https_proxy
HTTP_PROXY
http_proxy

Note: A future major version of restify(-clients) might change this environment variable behaviour. See the discussion on this issue.

The noProxy option can be used to exclude some hosts from using a given proxy. If it is not specified, then the NO_PROXY or no_proxy environment variable is used. Use noProxy: '' to override a possible environment variable, but not match any hosts.

The value is a string giving a comma-separated set of host, host-part suffix, or the special '*' to indicate all hosts. (Its definition is intended to match curl's NO_PROXY environment variable.) Some examples:

$ export NO_PROXY='*'               # don't proxy requests to any urls
$ export NO_PROXY='127.0.0.1'       # don't proxy requests the localhost IP
$ export NO_PROXY='localhost:8000'  # ... 'localhost' hostname and port 8000
$ export NO_PROXY='google.com'      # ... "google.com" and "*.google.com"
$ export NO_PROXY='www.google.com'  # ... "www.google.com"
$ export NO_PROXY='127.0.0.1, google.com'   # multiple hosts

Note: The url being requested must match the full hostname or hostname part to a '.': NO_PROXY=oogle.com does not match "google.com". DNS lookups are not performed to determine the IP address of a hostname.

basicAuth(username, password)

Since it hasn't been mentioned yet, this convenience method (available on all clients), just sets the Authorization header for all HTTP requests:

client.basicAuth('mark', 'mysupersecretpassword');

Upgrades

If you successfully negotiate an Upgrade with the HTTP server, an upgradeResult event will be emitted with the arguments err, res, socket and head. You can use this functionality to establish a WebSockets connection with a server. For example, using the watershed library:

var ws = new Watershed();
var wskey = ws.generateKey();
var options = {
  path: '/websockets/attach',
  headers: {
    connection: 'upgrade',
    upgrade: 'websocket',
    'sec-websocket-key': wskey,
  }
};
client.get(options, function(err, res, socket, head) {
  res.once('upgradeResult', function(err2, res2, socket2, head2) {
    var shed = ws.connect(res2, socket2, head2, wskey);
    shed.on('text', function(msg) {
      console.log('message from server: ' + msg);
      shed.end();
    });
    shed.send('greetings program');
  });
});

Common APIs

The following methods are shared by all versions of the client:

inflightRequests()

Returns the number of inflight requests the client is currently handling. The request count is incremented as soon as a verb method is called, which means that an inflight request is defined as a request that is in any of the following states:

• after the verb function is called but before any I/O
• waiting on I/O (dns resolution, socket connection, server response, etc.)
• request serialization (uploading of req bodies, etc.)
• response marshalling/consumption

The counter is decremented when the response's end event is emitted, or when the request's error event is emitted.

Events

The client emits the following events:

client.on('timings', function(timings) {... });

Per request timings are available under the req.getTimings() or emitted from the client via a timings event in milliseconds:

{
  dnsLookup: 34,
  tcpConnection: 52,
  tlsHandshake: 112,
  firstByte: 66,
  contentTransfer: 2,
  total: 266
}

All timings except total can be null under various circumstances like keep-alive connection, missing https etc.

client.on('metrics', function(metrics) {... });

Per request metrics are available under the req.getMetrics() method, and are also emitted from the client via a metrics event:

{
  statusCode: 200,
  method: 'POST',
  success: true,
  path: '/foo/bar',
  url: 'http://netflix.com',
  timings: {
    dnsLookup: 34,
    tcpConnection: 52,
    tlsHandshake: 112,
    firstByte: 66,
    contentTransfer: 2,
    total: 266
  }
}

client.on('after', function(req, res, err) {... });

The client will emit an after event after each completed request (successful or not). This can be useful for attaching after handlers to report metrics or other useful information about the client. When using this event, the timings and metrics information is available via the request getters, req.getTimings() and req.getMetrics().

For the StringClient and JSONClient, the after event is fired after the response has been received and parsed. For the HttpClient, usage is a bit more nuanced, which will be covered below.

Using the after event with the HttpClient

As the low level HttpClient's callback interface exposes the associated request and response streams directly to the consumer, there are a few caveats around using after with the HttpClient. Consumers must consume the response stream (by consuming the data event on the response) in order for the after event to be fired. For example, here's how an HttpClient instance consumes the response's data to trigger the after event:

httpClient.get('/200', function (err, req) {
  req.on('result', function (err, res) {
    let body = '';
    // must consume the stream first
    res.on('data', function (chunk) {
      body += chunk;
    });
  });
});

httpClient.once('after', function (req, res, err) {
  // this is fired after response's `end` event.
});

Another effect due to the low level nature of HttpClient means that any errors created in the consuming callback will not be available automatically in the after event. For example, if you get a 500, but then determine that your payload has some bad or malformed data, or you encounter an error while consuming your stream, it's your responsibility to propagate that back to after if you want to consume it later:

httpClient.get('/500', function (err, req) {
  req.on('result', function (err, res) {
    // err is an InternalServerError here. but a 500 could still return
    // some data like an error message, so let's parse the response now.
    res.on('data', function (chunk) {
      // parse the chunks here...
      // oh no, bad data! but how do I get this error to the after event?
      var invalidDataError = new Error('invalid data format!');
      // attach it to the req or res object
      req.httpClientError = invalidDataError;
    });
  });
});

httpClient.once('after', function (req, res, err) {
  // err will be an InternalServerError, and req.httpClientError will be your
  // custom error.
});

In timeout scenarios (connection timeout or request timeout), there is no response to consume, in which case the client's after event is fired at the same time as the req's result event.

Contributing

Add unit tests for any new or changed functionality. Ensure that lint and style checks pass. Ensure that your commit messages follow the conventional commits specification.

To start contributing, install the git pre-push hooks:

make githooks

Before committing, run the prepush hook:

make prepush

If you have style errors, you can auto fix whitespace issues by running:

make codestyle-fix

Cutting a release

In order to release a new version of the restify-clients module:

make release

This will run the release target, which will:

1. Automatically update the changelog.
2. Automatically bump the version number appropriate, using the conventional commit messages found in the git history since the last release.
3. Push the git tags up to origin
4. Call npm publish

License

Copyright (c) 2015 Alex Liu

Licensed under the MIT license.