Automattic / knox

S3 Lib
MIT License
1.74k stars 285 forks source link

knox

Node Amazon S3 Client.

Features

Examples

The following examples demonstrate some capabilities of knox and the S3 REST API. First things first, create an S3 client:

var client = knox.createClient({
    key: '<api-key-here>'
  , secret: '<secret-here>'
  , bucket: 'learnboost'
});

More options are documented below for features like other endpoints or regions.

PUT

If you want to directly upload some strings to S3, you can use the Client#put method with a string or buffer, just like you would for any http.Client request. You pass in the filename as the first parameter, some headers for the second, and then listen for a 'response' event on the request. Then send the request using req.end(). If we get a 200 response, great!

If you send a string, set Content-Length to the length of the buffer of your string, rather than of the string itself.

var object = { foo: "bar" };
var string = JSON.stringify(object);
var req = client.put('/test/obj.json', {
    'Content-Length': Buffer.byteLength(string)
  , 'Content-Type': 'application/json'
});
req.on('response', function(res){
  if (200 == res.statusCode) {
    console.log('saved to %s', req.url);
  }
});
req.end(string);

By default the x-amz-acl header is private. To alter this simply pass this header to the client request method.

client.put('/test/obj.json', { 'x-amz-acl': 'public-read' });

Each HTTP verb has an alternate method with the "File" suffix, for example put() also has a higher level method named putFile(), accepting a source filename and performing the dirty work shown above for you. Here is an example usage:

client.putFile('my.json', '/user.json', function(err, res){
  // Always either do something with `res` or at least call `res.resume()`.
});

Another alternative is to stream via Client#putStream(), for example:

http.get('http://google.com/doodle.png', function(res){
  var headers = {
      'Content-Length': res.headers['content-length']
    , 'Content-Type': res.headers['content-type']
  };
  client.putStream(res, '/doodle.png', headers, function(err, res){
    // check `err`, then do `res.pipe(..)` or `res.resume()` or whatever.
  });
});

You can also use your stream's pipe method to pipe to the PUT request, but you'll still have to set the 'Content-Length' header. For example:

fs.stat('./Readme.md', function(err, stat){
  // Be sure to handle `err`.

  var req = client.put('/Readme.md', {
      'Content-Length': stat.size
    , 'Content-Type': 'text/plain'
  });

  fs.createReadStream('./Readme.md').pipe(req);

  req.on('response', function(res){
    // ...
  });
});

Finally, if you want a nice interface for putting a buffer or a string of data, use Client#putBuffer():

var buffer = new Buffer('a string of data');
var headers = {
  'Content-Type': 'text/plain'
};
client.putBuffer(buffer, '/string.txt', headers, function(err, res){
  // ...
});

Note that both putFile and putStream will stream to S3 instead of reading into memory, which is great. And they return objects that emit 'progress' events too, so you can monitor how the streaming goes! The progress events have fields written, total, and percent.

GET

Below is an example GET request on the file we just shoved at S3. It simply outputs the response status code, headers, and body.

client.get('/test/Readme.md').on('response', function(res){
  console.log(res.statusCode);
  console.log(res.headers);
  res.setEncoding('utf8');
  res.on('data', function(chunk){
    console.log(chunk);
  });
}).end();

There is also Client#getFile() which uses a callback pattern instead of giving you the raw request:

client.getFile('/test/Readme.md', function(err, res){
  // check `err`, then do `res.pipe(..)` or `res.resume()` or whatever.
});

DELETE

Delete our file:

client.del('/test/Readme.md').on('response', function(res){
  console.log(res.statusCode);
  console.log(res.headers);
}).end();

Likewise we also have Client#deleteFile() as a more concise (yet less flexible) solution:

client.deleteFile('/test/Readme.md', function(err, res){
  // check `err`, then do `res.pipe(..)` or `res.resume()` or whatever.
});

HEAD

As you might expect we have Client#head and Client#headFile, following the same pattern as above.

Advanced Operations

Knox supports a few advanced operations. Like copying files:

client.copy('/test/source.txt', '/test/dest.txt').on('response', function(res){
  console.log(res.statusCode);
  console.log(res.headers);
}).end();

// or

client.copyFile('/source.txt', '/dest.txt', function(err, res){
  // ...
});

even between buckets:

client.copyTo('/source.txt', 'dest-bucket', '/dest.txt').on('response', function(res){
  // ...
}).end();

and even between buckets in different regions:

var destOptions = { region: 'us-west-2', bucket: 'dest-bucket' };
client.copyTo('/source.txt', destOptions, '/dest.txt', function(res){
  // ...
}).end();

or deleting multiple files at once:

client.deleteMultiple(['/test/Readme.md', '/test/Readme.markdown'], function(err, res){
  // ...
});

or listing all the files in your bucket:

client.list({ prefix: 'my-prefix' }, function(err, data){
  /* `data` will look roughly like:

  {
    Prefix: 'my-prefix',
    IsTruncated: true,
    MaxKeys: 1000,
    Contents: [
      {
        Key: 'whatever'
        LastModified: new Date(2012, 11, 25, 0, 0, 0),
        ETag: 'whatever',
        Size: 123,
        Owner: 'you',
        StorageClass: 'whatever'
      },
      ⋮
    ]
  }

  */
});

And you can always issue ad-hoc requests, e.g. the following to get an object's ACL:

client.request('GET', '/test/Readme.md?acl').on('response', function(res){
  // Read and parse the XML response.
  // Everyone loves XML parsing.
}).end();

Finally, you can construct HTTP or HTTPS URLs for a file like so:

var readmeUrl = client.http('/test/Readme.md');
var userDataUrl = client.https('/user.json');

Client Creation Options

Besides the required key, secret, and bucket options, you can supply any of the following:

endpoint

By default knox will send all requests to the global endpoint (s3.amazonaws.com). This works regardless of the region where the bucket is. But if you want to manually set the endpoint, e.g. for performance or testing reasons, or because you are using a S3-compatible service that isn't hosted by Amazon, you can do it with the endpoint option.

region

For your convenience when using buckets not in the US Standard region, you can specify the region option. When you do so, the endpoint is automatically assembled.

As of this writing, valid values for the region option are:

If new regions are added later, their subdomain names will also work when passed as the region option. See the AWS endpoint documentation for the latest list.

Convenience APIs such as putFile and putStream currently do not work as expected with buckets in regions other than US Standard without explicitly specify the region option. This will eventually be addressed by resolving issue #66; however, for performance reasons, it is always best to specify the region option anyway.

secure and port

By default, knox uses HTTPS to connect to S3 on port 443. You can override either of these with the secure and port options. Note that if you specify a custom port option, the default for secure switches to false, although you can override it manually if you want to run HTTPS against a specific port.

token

If you are using the AWS Security Token Service APIs, you can construct the client with a token parameter containing the temporary security credentials token. This simply sets the x-amz-security-token header on every request made by the client.

style

By default, knox tries to use the "virtual hosted style" URLs for accessing S3, e.g. bucket.s3.amazonaws.com. If you pass in "path" as the style option, or pass in a bucket value that cannot be used with virtual hosted style URLs, knox will use "path style" URLs, e.g. s3.amazonaws.com/bucket. There are tradeoffs you should be aware of:

For more information on the differences between these two types of URLs, and limitations related to them, see the following S3 documentation pages:

agent

Knox disables the default HTTP agent, because it leads to lots of "socket hang up" errors when doing more than 5 requests at once. See #116 for details. If you want to get the default agent back, you can specify agent: require("https").globalAgent, or use your own.

Beyond Knox

Multipart Upload

S3's multipart upload is their rather-complicated way of uploading large files. In particular, it is the only way of streaming files without knowing their Content-Length ahead of time.

Adding the complexity of multipart upload directly to knox is not a great idea. For example, it requires buffering at least 5 MiB of data at a time in memory, which you want to avoid if possible. Fortunately, @nathanoehlman has created the excellent knox-mpu package to let you use multipart upload with knox if you need it!

Easy Download/Upload

@superjoe30 has created a nice library, called simply s3, that makes it very easy to upload local files directly to S3, and download them back to your filesystem. For simple cases this is often exactly what you want!

Uploading With Retries and Exponential Backoff

@jergason created intimidate, a library wrapping Knox to automatically retry failed uploads with exponential backoff. This helps your app deal with intermittent connectivity to S3 without bringing it to a ginding halt.

Listing and Copying Large Buckets

@goodeggs created knox-copy to easily copy and stream keys of buckets beyond Amazon's 1000 key page size limit.

@segmentio created s3-lister to stream a list of bucket keys using the new streams2 interface.

@drob created s3-deleter, a writable stream that batch-deletes bucket keys.

Running Tests

To run the test suite you must first have an S3 account. Then create a file named ./test/auth.json, which contains your credentials as JSON, for example:

{
  "key": "<api-key-here>",
  "secret": "<secret-here>",
  "bucket": "<your-bucket-name>",
  "bucket2": "<another-bucket-name>",
  "bucketUsWest2": "<bucket-in-us-west-2-region-here>"
}

Then install the dev dependencies and execute the test suite:

$ npm install
$ npm test