/video-quality-tools

The set of tools to determine quality of livestream video.

Primary LanguageJavaScriptMIT LicenseMIT

Video Quality Tools module - helps to measure live stream characteristics by RTMP/HLS/DASH streams

Features:

  • fetching live video stream on demand, displaying info on video/audio characteristics;
  • monitoring live video streams in real time, running quality checks: fps and bitrate drops, GOP structure changes and more.

video-quality-tools requires ffmpeg and ffprobe.

NPM version Release Status Build Status Coverage Status

Installation

Using npm:

$ npm install --save video-quality-tools

Using yarn:

$ yarn add video-quality-tools

Error Handling

There are a lot of methods in module that may throw an error. All errors are subclasses of a basic javascript Error object and may be distinguished by prototype. To do this, just import the Error object from the module.

const {Errors} = require('video-quality-tools');

Errors object contains a set of error classes that may be thrown.

For example: StreamInfo constructor throws Errors.ConfigError for incorrect options.

const {StreamsInfo, Errors} = require('video-quality-tools');

try {
    const streamsInfo = new StreamsInfo(null, 'rtmp://host:port/appInstance/name');
} catch (err) {
    if (err instanceof Errors.ConfigError) {
        console.error('Invalid options:', err);
    }
}

Exit Reasons

StreamsInfo and FramesMonitor use ffprobe to fetch the data. The underlying ffprobe process may be killed by someone, it may fail due to spawn issues, it may exit normally or with error code or it may be killed by the module itself if frames have invalid format.

To distinguish exit reasons you may import ExitReasons object.

const {ExitReasons} = require('video-quality-tools');

There are such available classes:

  • ExitReasons.StartError
  • ExitReasons.ExternalSignal
  • ExitReasons.NormalExit
  • ExitReasons.AbnormalExit
  • ExitReasons.ProcessingError

Description of a specific reason class may be found in further chapters.

One-time Live Stream Info

To fetch one-time info you need to create StreamsInfo class instance

const {StreamsInfo} = require('video-quality-tools');

const streamsInfoOptions = {
    ffprobePath: '/usr/local/bin/ffprobe',
    timeoutInMs: 2000
};
const streamsInfo = new StreamsInfo(streamsInfoOptions, 'rtmp://host:port/appInstance/name');

Constructor throws:

  • Errors.ConfigError if options have invalid type or value;
  • Errors.ExecutablePathError if options.ffprobePath is not found or it's not an executable.

After that you may run fetch method to retrieve video and audio info. Method can be called as many times as you want.

// async-await style
const streamInfo = await streamsInfo.fetch();

// or using old-school promise style

streamsInfo.fetch()
    .then(info => {
        console.log('Video info:');
        console.log(info.videos);
        console.log('Audio info:');
        console.log(info.audios);
    })
    .catch(err => console.error(err));

Method may throw the Errors.StreamsInfoError if it can't receive stream, stream is invalid, ffprobe exits with error or returns unexpected output.

The videos and audios fields of the returned info object are arrays. Usually there is only one video or audio stream for RTMP streams. Each element of the videos or audios array has almost the same structure as the ffprobe -show_streams output has. You may find a typical output of fetch command below.

videos and audios may be an empty array if there are no appropriate streams in the live stream.

{ videos:
   [ { index: 1,
       codec_name: 'h264',
       codec_long_name: 'H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10',
       profile: 'Main',
       codec_type: 'video',
       codec_time_base: '33/2000',
       codec_tag_string: '[0][0][0][0]',
       codec_tag: '0x0000',
       width: 854,
       height: 480,
       coded_width: 854,
       coded_height: 480,
       has_b_frames: 0,
       sample_aspect_ratio: '1280:1281',
       display_aspect_ratio: '16:9',
       pix_fmt: 'yuv420p',
       level: 31,
       chroma_location: 'left',
       field_order: 'progressive',
       refs: 1,
       is_avc: 'true',
       nal_length_size: '4',
       r_frame_rate: '30/1',
       avg_frame_rate: '1000/33',
       time_base: '1/1000',
       start_pts: 2062046,
       start_time: '2062.046000',
       bits_per_raw_sample: '8',
       disposition: [Object] } ],
  audios:
   [ { index: 0,
       codec_name: 'aac',
       codec_long_name: 'AAC (Advanced Audio Coding)',
       profile: 'LC',
       codec_type: 'audio',
       codec_time_base: '1/44100',
       codec_tag_string: '[0][0][0][0]',
       codec_tag: '0x0000',
       sample_fmt: 'fltp',
       sample_rate: '44100',
       channels: 2,
       channel_layout: 'stereo',
       bits_per_sample: 0,
       r_frame_rate: '0/0',
       avg_frame_rate: '0/0',
       time_base: '1/1000',
       start_pts: 2061964,
       start_time: '2061.964000',
       disposition: [Object] } ] }

Live Frames Monitor

To measure live stream info you need to create instance of FramesMonitor class

const {FramesMonitor} = require('video-quality-tools');

const framesMonitorOptions = {
    ffprobePath: '/usr/local/bin/ffprobe',
    timeoutInMs: 2000,
    bufferMaxLengthInBytes: 100000,
    errorLevel: 'error',
    exitProcessGuardTimeoutInMs: 1000,
    analyzeDurationInMs: 9000
};

const framesMonitor = new FramesMonitor(framesMonitorOptions, 'rtmp://host:port/appInstance/name');

Constructor throws:

  • Errors.ConfigError or TypeError if options have invalid type or value;
  • Errors.ExecutablePathError if options.ffprobePath is not found or it's not an executable.

Frames Monitor Config

The first argument of FramesMonitor must be an options object. All options object's fields are mandatory:

  • ffprobePath - string, path to ffprobe executable;
  • timeoutInMs - integer, greater than 0, specifies the maximum time to wait for (network) read/write operations to complete;
  • bufferMaxLengthInBytes - integer, greater than 0, specifies the buffer length for ffprobe frames. This setting prevents from hanging and receiving incorrect data from the stream, usually 1-2 KB is enough;
  • errorLevel - specifies log level for debugging purposes, must be equal to ffprobe's -loglevel option. May be one of the following values: trace, debug, verbose, info, warning, error, fatal, panic, quiet. For most cases error level is enough;
  • exitProcessGuardTimeoutInMs - integer, greater than 0, specifies the amount of time after which the monitoring process will be hard killed if the attempt of soft stop fails. When you try to stop a monitor with stopListen() method the FramesMonitor sends SIGTERM signal to ffprobe process. ffprobe may ignore this signal (some versions do it pretty often). If ffprobe doesn't exit after exitProcessGuardTimeoutInMs milliseconds, FramesMonitor sends SIGKILL signal and forces underlying ffprobe process to exit.
  • analyzeDurationInMs - integer, greater than 0, specifies the maximum analyzing time of the input.

Listening of Frames

After creation of the FramesMonitor instance, you may start listening live stream data. To do so, just run framesMonitor.listen() method. After that framesMonitor starts emitting frame event as soon as ffprobe decodes frame from the stream. It emits video and audio frames.

const {FramesMonitor, processFrames, ExitReasons} = require('video-quality-tools');

const framesMonitor = new FramesMonitor(options, 'rtmp://host:port/appInstance/name');

framesMonitor.on('frame', frameInfo => {
    console.log(frameInfo);
});

framesMonitor.listen();

listen() method doesn't return anything but may throw Errors.AlreadyListeningError.

Stop Listening of Frames

To stop listening, call framesMonitor.stopListen() method. It returns promise. Rejection of that promise means that the underlying ffprobe process can't be killed with a signal. Method tries to send SIGTERM signal first and wait for exitProcessGuardTimeoutInMs milliseconds (ref. Frames Monitor Config section). If the process doesn't exit after that timeout, method sends SIGKILL and forces it to exit. Resolved promise means that the process was successfully killed.

try {
    const {code, signal} = await framesMonitor.stopListen();
    console.log(`Monitor was stopped successfully, code=${code}, signal=${signal}`);
} catch (err) {
    // instance of Errors.ProcessExitError
    console.log(`Error listening url "${err.payload.url}": ${err.payload.error.message}`);
}

frame event

This event is generated on each video and audio frame decoded by ffprobe. The structure of the frame object is the following:

{ media_type: 'video',
  key_frame: 0,
  pkt_pts_time: 3530.279,
  pkt_size: 3332,
  width: 640,
  height: 480,
  pict_type: 'P' }

or

{ media_type: 'audio',
  key_frame: 1,
  pkt_pts_time: 'N/A',
  pkt_size: 20 }

exit event

Underlying process may not start at all, it may fail after some time or it may be killed with signal. In such situations FramesMonitor class instance emits exit event and passes one of the ExitReasons instances. Each instance has its own reason-specific payload field. There is a list of reasons:

  • ExitReasons.StartError - ffprobe can't be spawned, the error object is stored in payload.error field;
  • ExitReasons.NormalExit - ffprobe has exited with code = 0, payload.code is provided;
  • ExitReasons.AbnormalExit - ffprobe has exited with non-zero exit code, payload.code contains the exit code of the ffprobe process and payload.stderrOutput contains the last 5 lines from ffprobe's stderr output;
  • ExitReasons.ProcessingError - monitor has detected a logical issue and forces ffprobe to exit, this exit reason contains error object in payload.error field that may be either Errors.ProcessStreamError or Errors.InvalidFrameError;
  • ExitReasons.ExternalSignal - ffprobe process was killed by someone or by another process with the signal, the signal name can be found in payload.signal field;
framesMonitor.on('exit', reason => {
    switch(reason.constructor) {
        case ExitReasons.AbnormalExit:
            assert(reason.payload.code);
            assert(reason.payload.stderrOutput); // stderrOutput may be empty
            break;
        case ExitReasons.NormalExit:
            assert(reason.payload.code);
            break;
        case ExitReasons.ExternalSignal:
            assert(reason.payload.signal);
            break;
        case ExitReasons.StartError:
            assert.instanceOf(reason.payload.error, Error);
            break;
        case ExitReasons.ProcessingError:
            assert.instanceOf(reason.payload.error, Error);
            break;
    }
});

error event

May be emitted only once and only in case the framesMonitor.stopListen() method receives error event on killing of an underlying ffprobe process.

framesMonitor.on('error', err => {
    // indicates error during the kill process
    // when ProcessingError occurs we may encounter that can not kill process
    // in this case this error event would be emitted
    
    assert.instanceOf(err, Error);
});

Video Quality Info

video-quality-tools ships with functions that help determining live stream info based on the set of frames collected from FramesMonitor:

  • processFrames.networkStats
  • processFrames.encoderStats

processFrames.networkStats(frames, durationInMsec)

Receives an array of frames collected for a given time interval durationInMsec.

This method doesn't analyze GOP structure and isn't dependant on fullness of GOP between runs. Method shows only frame rate of audio and video streams received, bitrate of audio and video. Instead of processFrames.networkStats this method allows to control quality of network link between sender and receiver (like RTMP server).

Remember that this module must be located not far away from receiver server (that is under analysis). If link between receiver and module affects delivery of RTMP packages this module indicates incorrect values. It's better to run this module near the receiver.

const {processFrames} = require('video-quality-tools');

const INTERVAL_TO_ANALYZE_FRAMES = 5000; // in milliseconds

let frames = [];

framesMonitor.on('frame', frame => {
    frames.push(frame);
});

setInterval(() => {
    try {
        const info = processFrames.networkStats(frames, INTERVAL_TO_ANALYZE_FRAMES);
    
        console.log(info);

        frames = [];
    } catch(err) {
        // only if arguments are invalid
        console.log(err);
        process.exit(1);
    }
}, INTERVAL_TO_ANALYZE_FRAMES);

There is an output for the example above:

{
  videoFrameRate: 29,
  audioFrameRate: 50,
  videoBitrate: 1403.5421875,
  audioBitrate: 39.846875
}

Check examples/networkStats.js to see an example code.

processFrames.encoderStats(frames)

It relies on GOP structure of the stream.

The following example shows how to gather frames and pass them to the function that analyzes encoder statistic.

const {processFrames} = require('video-quality-tools');

const AMOUNT_OF_FRAMES_TO_GATHER = 300;

let frames = [];

framesMonitor.on('frame', frame => {
    frames.push(frame);

    if (AMOUNT_OF_FRAMES_TO_GATHER > frames.length) {
        return;
    }

    try {
        const info = processFrames.encoderStats(frames);
        frames = info.remainedFrames;
    
        console.log(info.payload);
    } catch(err) {
        // processing error
        console.log(err);
        process.exit(1);
    }
});

There is an output for the example above:

{ areAllGopsIdentical: true,
  bitrate:
   { mean: 1494.9075520833333,
     min: 1440.27734375,
     max: 1525.95703125 },
  fps: { 
     mean: 30,
     min: 30, 
     max: 30 },
  gopDuration: {
     mean: 2,
     min: 1.9, 
     max: 2.1 },
  displayAspectRatio: '16:9',
  width: 1280,
  height: 720,
  hasAudioStream: true
}

In given example the frames are collected in frames array and than use processFrames.encoderStats function for sets of 300 frames (AMOUNT_OF_FRAMES_TO_GATHER). The function searches the key frames and measures the distance between them.

It's impossible to detect GOP structure for a set of frames with only one key frame, so processFrames.encoderStats returns back all passed frames as an array in remainedFrames field.

If there are more than 2 key frames, processFrames.encoderStats uses full GOPs to track fps and bitrate and returns all frames back in the last GOP that was not finished. It's important to remember the remainedFrames output and push a new frame to the remainedFrames array when it arrives.

For the full GOPs processFrames.encoderStats calculates min/max/mean values of bitrates (in kbit/s), framerates and GOP duration (in seconds) and returns them in payload field. The result of the check for the similarity of GOP structures for the collected GOPs is returned in areAllGopsIdentical field. Fields width, height and displayAspectRatio are taken from data from first frame of the first collected GOP. Value of hasAudioStream reflects presence of audio frames.

To calculate display aspect ratio method processFrames::calculateDisplayAspectRatio uses list of video aspect ratio standards with approximation of frames width and height ratio. If ratio can't be found in list of known standards, even in delta neighbourhood, then GCD algorithm is used to simplify returned value.

processFrames.encoderStats may throw Errors.GopNotFoundError.

Also, you may extend the metrics. Check src/processFrames.js to find common functions.