How do you use Node.js to stream an MP4 file with

Question

I've been trying to solve this problem for several days now and would really appreciate any help on the subject.

I'm able to successfully stream an mp4 audio file stored on a Node.js server using fluent-ffmpeg by passing the location of the file as a string and transcoding it to mp3. If I create a file stream from the same file and pass that to fluent-ffmpeg instead it works for an mp3 input file, but not a mp4 file. In the case of the mp4 file no error is thrown and it claims the stream completed successfully, but nothing is playing in the browser. I'm guessing this has to do with the meta data being stored at the end of an mp4 file, but I don't know how to code around this. This is the exact same file that works correctly when it's location is passed to ffmpeg, rather than the stream. When I try and pass a stream to the mp4 file on s3, again no error is thrown, but nothing streams to the browser. This isn't surprising as ffmpeg won't work with the file locally as stream, so expecting it to handle the stream from s3 is wishful thinking.

How can I stream the mp4 file from s3, without storing it locally as a file first? How do I get ffmpeg to do this without transcoding the file too? The following is the code I have at the moment which isn't working. Note that it attempts to pass the s3 file as a stream to ffmpeg and it's also transcoding it into an mp3, which I'd prefer not to do.

.get(function(req,res) {
    aws.s3(s3Bucket).getFile(s3Path, function (err, result) {
        if (err) {
            return next(err);
        }
        var proc = new ffmpeg(result)
            .withAudioCodec('libmp3lame')
            .format('mp3')
            .on('error', function (err, stdout, stderr) {
                console.log('an error happened: ' + err.message);
                console.log('ffmpeg stdout: ' + stdout);
                console.log('ffmpeg stderr: ' + stderr);
            })
            .on('end', function () {
                console.log('Processing finished !');
            })
            .on('progress', function (progress) {
                console.log('Processing: ' + progress.percent + '% done');
            })
            .pipe(res, {end: true});
    });
});

This is using the knox library when it calls aws.s3... I've also tried writing it using the standard aws sdk for Node.js, as shown below, but I get the same outcome as above.

var AWS = require('aws-sdk');

var s3 = new AWS.S3({
    accessKeyId: process.env.AWS_ACCESS_KEY_ID,
    secretAccessKey: process.env.AWS_SECRET_KEY,
    region: process.env.AWS_REGION_ID
});
var fileStream = s3.getObject({
        Bucket: s3Bucket,
        Key: s3Key
    }).createReadStream();
var proc = new ffmpeg(fileStream)
    .withAudioCodec('libmp3lame')
    .format('mp3')
    .on('error', function (err, stdout, stderr) {
        console.log('an error happened: ' + err.message);
        console.log('ffmpeg stdout: ' + stdout);
        console.log('ffmpeg stderr: ' + stderr);
    })
    .on('end', function () {
        console.log('Processing finished !');
    })
    .on('progress', function (progress) {
        console.log('Processing: ' + progress.percent + '% done');
    })
    .pipe(res, {end: true});

=====================================

Updated

I placed an mp3 file in the same s3 bucket and the code I have here worked and was able to stream the file through to the browser without storing a local copy. So the streaming issues I face have something to do with the mp4/aac container/encoder format.

I'm still interested in a way to bring the m4a file down from s3 to the Node.js server in it's entirety, then pass it to ffmpeg for streaming without actually storing the file in the local file system.

=====================================

Updated Again

I've managed to get the server streaming the file, as mp4, straight to the browser. This half answers my original question. My only issue now is that I have to download the file to a local store first, before I can stream it. I'd still like to find a way to stream from s3 without needing the temporary file.

aws.s3(s3Bucket).getFile(s3Path, function(err, result){
    result.pipe(fs.createWriteStream(file_location));
    result.on('end', function() {
        console.log('File Downloaded!');
        var proc = new ffmpeg(file_location)
            .outputOptions(['-movflags isml+frag_keyframe'])
            .toFormat('mp4')
            .withAudioCodec('copy')
            .seekInput(offset)
            .on('error', function(err,stdout,stderr) {
                console.log('an error happened: ' + err.message);
                console.log('ffmpeg stdout: ' + stdout);
                console.log('ffmpeg stderr: ' + stderr);
            })
            .on('end', function() {
                console.log('Processing finished !');
            })
            .on('progress', function(progress) {
                console.log('Processing: ' + progress.percent + '% done');
            })
            .pipe(res, {end: true});
    });
});

On the receiving side I just have the following javascript in an empty html page:

window.AudioContext = window.AudioContext || window.webkitAudioContext;
context = new AudioContext();

function process(Data) {
    source = context.createBufferSource(); // Create Sound Source
    context.decodeAudioData(Data, function(buffer){
        source.buffer = buffer;
        source.connect(context.destination);
        source.start(context.currentTime);
    });
};

function loadSound() {
    var request = new XMLHttpRequest();
    request.open("GET", "/stream/<audio_identifier>", true);
    request.responseType = "arraybuffer";

    request.onload = function() {
        var Data = request.response;
        process(Data);
    };

    request.send();
};

loadSound()

=====================================

The Answer

The code above under the title 'updated again' will stream an mp4 file, from s3, via a Node.js server to a browser without using flash. It does require that the file be stored temporarily on the Node.js server so that the meta data in the file is moved from the end of the file to the front. In order to stream without storing the temporary file, you need to actual modify the file on S3 first and make this meta data change. If you have changed the file in this way on S3 then you can modify the code under the title 'updated again' so that the result from S3 is piped straight into the ffmpeg constructor, rather than into a file stream on the Node.js server, then providing that file location to ffmepg, as the code does now. You can change the final 'pipe' command to 'save(location)' to get a version of the mp4 file locally with the meta data moved to the front. You can then upload that new version of the file to S3 and try out the end to end streaming. Personally I'm now going to create a task that modifies the files in this way as they are uploaded to s3 in the first place. This allows me to record and stream in mp4 without transcoding or storing a temp file on the Node.js server.

Answer 1

Blockquote How can I stream the mp4 file from s3, without storing it locally as a file first? How do I get ffmpeg to do this without transcoding the file too? The following is the code I have at the moment which isn't working. Note that it attempts to pass the s3 file as a stream to ffmpeg and it's also transcoding it into an mp3, which I'd prefer not to do.

AFAIK - if the moov atom is in the right place in media file, for S3 hosted mp4, nothing special is require for streaming because you can rely on http for that. If the client request "chunked" encoding it will get just that, a chunked stream terminated by the "END-OF" marker shown below.

0\r\n
\r\n 

By including the chunked header, the client is saying " I want a stream" . Under the covers, S3 is just nginx or apache isn't it? They both honor the headers.

test it with curl CLI as your client...

> User-Agent: curl/7.28.1-DEV
> Host: S3.domain
> Accept: */*
> Transfer-Encoding: chunked
> Content-Type: video/mp4
> Expect: 100-continue

May want to try out adding the codecs to the "Content-Type:" header. I dont know, but dont think it would be required for this type of streaming ( the atom resolves that stuff )

Answer 2

I had an issue buffering file streams from the S3 file object. The s3 filestream does not have the correct headers set and it seems it does not implement piping correctly.

I think a better solution is to use this nodejs module called s3-streams. It sets the correct headers and buffers the output so that the stream can be correctly piped to the output response socket. It saves you from saving the filestream locally first before restreaming it.

Answer 3

One of the main issues here is that you cannot seek on a piped stream. So you would have to store the file first. However, if you want to just stream from the beginning you can use a slightly different construction and pipe. Here is an example of the most straight forward way to do it.

// can just create an in-memory read stream without saving
var stream = aws.s3(s3Bucket).getObject(s3Path).createReadStream();

// fluent-ffmpeg supports a readstream as an arg in constructor
var proc = new ffmpeg(stream)
    .outputOptions(['-movflags isml+frag_keyframe'])
    .toFormat('mp4')
    .withAudioCodec('copy')
    //.seekInput(offset) this is a problem with piping
    .on('error', function(err,stdout,stderr) {
        console.log('an error happened: ' + err.message);
        console.log('ffmpeg stdout: ' + stdout);
        console.log('ffmpeg stderr: ' + stderr);
    })
    .on('end', function() {
        console.log('Processing finished !');
    })
    .on('progress', function(progress) {
        console.log('Processing: ' + progress.percent + '% done');
    })
    .pipe(res, {end: true});