Accessing the raw file stream from a node-formidable file upload

Question

I am creating an application that takes some file uploads and send them straight up to S3. I would prefer not to even have the tmp file on my server, so I am using the Knox module and would like to take the raw stream from Formidable and send it over Knox to S3. I have done something similar using Knox to download a file using this code:

knox.downloads.get(widget.download).on('response',function(sres){
    res.writeHead(200, {
        'Content-Type':'application/zip',
        'Content-Length': sres.headers['content-length'],
        'Content-Disposition':'attachment; filename=' + widget.download
    });
    util.pump(sres, res);
}).end();

Now I would like to do something similar in the oposite direction (File upload from the browser to S3).

So far I have written an event handler to capture each piece of data from the file as it's being uploaded:

var form = new formidable.IncomingForm();
form.onPart = function(part){
    if(!part.filename){
        form.handlePart(part);
    }else{
        if(part.name == 'download'){
            // Upload to download bucket
            controller.putDownload(part);
        }else{
            // Upload to the image bucket
            controller.putImage(part);
        }
        //res.send(sys.inspect(part));
    }
}
form.parse(req, function(err, fields, files){
    if(err){
        res.json(err);
    }else{
        res.send(sys.inspect({fields:fields, files:files}), {'content-type':'text/plain'});
        //controller.createWidget(res,fields,files);            
    }
});


controller.putDownload = function(part){
    part.addListener('data', function(buffer){
        knox.download.putStream(data,part.filename, function(err,s3res){
            if(err)throwError(err);
            else{
                console.log(s3res);
            }
        });
    })
    knox.downloads.putStream(part, part.filename, function(err,s3res){

        if(err)throwError(err);
        else{
            console.log(s3res);
        }
    });
}

But the data event only give me the buffer. So is it possible to capture the stream itself and push it to S3?

Answer 1

What you want to do is override the Form.onPart method:

IncomingForm.prototype.onPart = function(part) {
  // this method can be overwritten by the user
  this.handlePart(part);
};

Formidable's default behavior is to write the part to a file. You don't want that. You want to handle the 'part' events to write to the knox download. Start with this:

form.onPart = function(part) {
    if (!part.filename) {
        // let formidable handle all non-file parts
        form.handlePart(part);
        return;
    }

Then open the knox request and handle the raw part events yourself:

part.on('data', function(data) {
    req.write(data);
});
part.on('end', function() {
    req.end();
});
part.on('error', function(err) {
    // handle this too
});

As a bonus, if the req.write(data) return false that means the send buffer is full. You should pause the Formidable parser. When you get a drain event from the Knox stream you should resume Formidable.

Answer 2

Use multiparty instead. It supports this kind of streaming like you want. It even has an example of streaming directly to s3: https://github.com/superjoe30/node-multiparty/blob/master/examples/s3.js

Answer 3

In an Express middleware, I use formidable together with PassThrough to stream-upload a file to S3 (in my case, to Minio which is S3 compatible through Minio SDK; and I believe it works for AWS S3 too with the same Minio SDK)

Here is the sample code.

const formidable = require('formidable')
const { PassThrough } = require('stream')

const form = new formidable.IncomingForm()
const pass = new PassThrough()

const fileMeta = {}
form.onPart = part => {
  if (!part.filename) {
    form.handlePart(part)
    return
  }
  fileMeta.name = part.filename
  fileMeta.type = part.mime
  part.on('data', function (buffer) {
    pass.write(buffer)
  })
  part.on('end', function () {
    pass.end()
  })
}
form.parse(req, err => {
  if (err) {
    req.minio = { error: err }
    next()
  } else {
    handlePostStream(req, next, fileMeta, pass)
  }
})

And handlePostStream looks like below, for your reference:

const uuidv1 = require('uuid/v1')

const handlePostStream = async (req, next, fileMeta, fileStream) => {
  let filename = uuidv1()

  try {
    const metaData = {
      'content-type': fileMeta.type,
      'file-name': Buffer.from(fileMeta.name).toString('base64')
    }

    const minioClient = /* Get Minio Client*/
    await minioClient.putObject(MINIO_BUCKET, filename, fileStream, metaData)

    req.minio = { post: { filename: `${filename}` } }
  } catch (error) {
    req.minio = { error }
  }
  next()
}

You can find the source code on GitHub, and its unit tests too.