Skip to content

Latest commit

 

History

History
352 lines (285 loc) · 20.6 KB

README.md

File metadata and controls

352 lines (285 loc) · 20.6 KB
Raw

Browser Emulator documentation

Worker service implemented in NodeJS and controlled with a REST protocol that capable of connecting to an OpenVidu session and sending and receiving WebRTC media using openvidu-browser library. The browser-emulator service provides several connection modes:

  • Traditional mode: It is able to start and launch containerized Chrome browsers using Docker and Selenium emulating a fully real user.

  • Emulated mode: It is capable to emulate a user connection without use a browser:

    • Overriding a WebRTC API using node-webrtc library and getting the media from a canvas publishing a moving image.
    • Overriding a WebRTC API using node-webrtc library and getting the media from a video file using ffmpeg.
    • Overriding the peerConnection objects using Kurento and getting the media from a video file.

This app provides a simple REST API that will be used by Load Test application and it allows:

This services also is listening for a WebSocket communication on ws:browser-emulator-addres:5001/events. It will send information from openvidu-browser to the loadtest-controller.

Running QoE Analysis

BrowserEmulator can run analysis on the Quality of Experience on your video and audio. For this, the following requisites have to be met:

  • The browserMode when adding a participant has to be REAL (at the moment this option doesn't work with EMULATED users).
  • AWS public and secret keys or Minio keys have to be set so the recorded video fragments can be uploaded to S3 or Minio.
  • The original video and audio (separate files expected) need to be cut and concrete paddings added to them. This can be done with the generate-mediafiles-qoe.sh script. The mandatory options are:
    • -d: Duration of the cut video, the video will be cut from the beginning to the second passed in this argument.
    • -p: Duration of the padding to be added (in seconds)
    • -v: Video location URL, the script will try to download the video from this URL using wget.
    • -a: Audio location URL, the script will try to download the audio from this URL using wget.
    • -w: Video width.
    • -h: Video height.
    • -f: Video framerate.The video and audio have to be saved in src/assets/mediafiles in y4m and wav fromat respectively so browser-emulator can see them.
      • The video file has to have the following name: fakevideo_[framerate]fps_[width]x[height].y4m, for example: fakevideo_30fps_640x480.y4m.
      • The audio file has to have the following name: fakeaudio.wav. Alternatively you can use our already preprocessed videos that can be downloaded using the download_mediafiles.sh script. These have 1 secvond of padding and 5 seconds of cut video.

For the purpose of running the QoE Analysis, for each user pair 2 videos will be recorded during the run, one recording what the first user is receiving from the second and viceversa. These videos will be uploaded to S3 or Minio when Delete all participant is run. These recordings will be called individual recordings.

When running Initialize instance, you can add the property "qoeAnalysis" to the body, an object with the following options: - enabled: defaults false, setting this to true will enable the individual recordings. - fragment_duration: The duration of the cut video without the paddings chosen when running generate-mediafiles-qoe.sh - padding_duration: The duration of the padding chosen when running generate-mediafiles-qoe.sh

When Delete all participant is run, all individual recorded videos will be uploaded to S3 with the following name structure: QOE_[Session]_[UserFrom]_[UserTo].webm, from here you have 2 options.

  • Running the analysis in the same worker machine that recorded the videos, for that send a request to Start QoE Analysis
  • Running the analysis on another machine, for that pull this repository in that machine, cd to browser-emulator and follow the next steps:
    • Install all the needed dependencies on the machine, which are:
      • Python3 and pip3
      • VMAF, set a VMAF_PATH environment variable with the path to the vmaf binary and move the vmaf_v0.6.1.json file found in the model directory to /usr/local/share/vmaf/models/vmaf_v0.6.1.json
      • VQMT, set a VQMT_PATH environment variable with the path to the vqmt binary
      • PESQ, set a PESQ_PATH environment variable with the path to the pesq binary
      • ViSQOL, set a VISQOL_PATH environment variable with the path to the directory that contains banzel-bin/visqol
      • Tesseract OCR, it is highly recommended to build it disabling multithreading as explained here to improve performance.
        • You will probably need to save the necessary model found here in /usr/local/share/tessdata/eng.traineddata
      • npm dependencies 
        npm install
      • qoe script dependencies 
        pip3 install -r qoe_scripts/requirements.txt
    • Save the individual recordings you want to analyze in ./recordings/qoe directory.
    • Create a qoe-results-processing-config.json file (there is an example file with the same name in this repo), with the following structure:
      • elasticsearch_hostname: ELK hostname
      • elasticsearch_username: ELK username
      • elasticsearch_password: ELK password
      • index: ELK index to save the data
      • fragment_duration: The duration of the cut video without the paddings chosen when running generate-mediafiles-qoe.sh
      • padding_duration: The duration of the padding chosen when running generate-mediafiles-qoe.sh
      • width: Video width
      • height: Video height
      • framerate: Video framerate
      • presenter_video_file_location: Location of the original video, usually located in src/assets/mediafiles/fakevideo_[framerate]fps_[width]x[height].y4m
      • presenter_audio_file_location: Location of the original audio, usually located in src/assets/mediafiles/fakeaudio.wav
      • timestamps: Optional. An array of objects with info about when a user has been added to a session, used to make a timeline. If not added, this info will be searched in the index indicated in ELK. The objects have the following structure:
        • new_participant_id: username of the user
        • new_participant_session: session
        • @timestamp: timestamp when the user was added to the session (in ISO format)
    • Run: 
      npm run qoe
    • You can also put a limit on the number of virtual cores used by the QoE scripts by adding the number as argument, for example with max 4 vcpus: 
      npm run qoe -- --cpus=4
    • If you have the results of running the qoe scripts but have not uploaded them to ELK (*_cuts.json files) you can upload them by running: 
      npm run qoe -- --process
    • If you don't want to automatically upload the data to ELK when a test finishes, use: 
      npm run qoe -- --onlyfiles

After the analysis is done, the results will be uploaded to the selected index in ELK. A dashboard can be imported to Kibana importing the loadtest.ndjson file. For more information about the imported data, check their respective pages: VMAF, VQMT, PESQ, ViSQOL. Note: The QoE results are normalized in the range 0-1 before importing them to ELK.

REST API

PING INSTANCE

INITIALIZE INSTANCE

{
	"elasticSearchHost": "your-elasticsearch-hostname",
	"elasticSearchUserName": "your-username",
	"elasticSearchPassword": "your-password",
	"elasticSearchIndex": "your-optional-index",
	"awsAccessKey": "your-AWS-access-key",
	"awsSecretAccessKey": "your-AWS-secret-key",
	"s3BucketName": "your-s3-bucket-name",
	"minioAccessKey": "your-Minio-access-key",
	"minioSecretKey": "your-Minio-secret-key",
	"minioHost": "your-Minio-endpoint",
	"minioBucket": "your-Minio-bucket",
	"minioPort": "your-optional-Minio-port (default 443)",
	"browserVideo": {
		"videoType": "bunny",
		"videoInfo":
		{
			"width": 640,
			"height": 480,
			"fps": 30
		}
	},
	"qoeAnalysis": {
		"enabled": true,
		"fragment_duration": 5,
		"padding_duration": 1
	}

}

  • RESPONSE

Instance has been initialized

Note: the Minio variables (minioAccessKey, minioSecretAccessKey, minioHost, minioBucket) and the AWS variables (awsAccessKey, awsSecretAccessKey, s3BucketName) are exclusive, in the case the request has both types of variables the Minio ones have priority, and files will try to be uploaded to Minio.

browserVideo is only needed if using browserMode REAL. This object indicates what video to use, which will be downloaded. There are 3 types of video by default, having to be selected in the videoType property inside browserVideo. If using a default video type, the resolution and fps of the video to use have to be selected in the videoInfo property. All default videos have padding for possible usage in QoE Analysis. The default video types and available resolutions and framerates are the following:

  • bunny: blender animated demo video
    • available resolutions and framerates:
      • Width: 640, height: 480, fps: 30
      • Width: 640, height: 480, fps: 60
      • Width: 1280, height: 720, fps: 30
      • Width: 1280, height: 720, fps: 60
      • Width: 1920, height: 1080, fps: 30
      • Width: 1920, height: 1080, fps: 60
  • interview: interview video
    • available resolutions and framerates:
      • Width: 640, height: 480, fps: 30
  • game: gameplay video with commentary
    • available resolutions and framerates:
      • Width: 640, height: 480, fps: 30

In Qoe Analysis object:

  • fragment_duration indicates the duration of the video fragment between the paddings
  • padding_duration indicates the duration of the padding

The default videos have been already processed by adding padding, the following durations are:

  • bunny:
    • fragment_duration: 5
    • padding_duration: 1
  • interview:
    • fragment_duration: 15
    • padding_duration: 1
  • game:
    • fragment_duration: 30
    • padding_duration: 1

Also, custom videos can be used by providing an url, example of using a custom video (Note: it is highly recommended that the width and height of the videos are a multiple of 16 if using the QoE Analysis features):

{
	...,
	"browserVideo": {
		"videoType": "custom",
    "customVideo": {
			"audioUrl": "https://fakeurl.com/fakeaudio.wav",
			"video": 
				{
					"url": "https://fakeurl.com/fakevideo480.y4m",
					"width": 640,
					"height": 480,
					"fps": 30
				}
		}
	}
}

CREATE PARTICIPANT

This endpoint provides a lot of configuration that you should take into account. As said before, you can make a request to create a new Stream Manger using your own token or make a request letting the browser-emulator create a new one.

CREATE PARTICIPANT (using your custom token)

Create a new participant with a specified token

  • METHOD: POST

  • URL: https://localhost:5000/openvidu-browser/streamManager

  • BODY

    {
	"browserMode": "EMULATE",
	"properties": {
		"token": "*****",
		"userId": "User1",
		"role": "PUBLISHER",
		"audio": true,
		"video": true,
	}
}

  • RESPONSE

    {
	"connectionId": "con_FmtswrvbkT",
	"workerCpuUsage": 10.00
}

CREATE PARTICIPANT (Creating a token)

Create a new participant with a specified role and connect it into a specified sessionName

  • METHOD: POST

  • URL: https://localhost:5000/openvidu-browser/streamManager

  • BODY

    {
	"openviduUrl": "https://localhost:4443",
	"openviduSecret": "MY_SECRET",
	"browserMode": "EMULATE",
	"properties": {
		"userId": "User1",
		"sessionName": "LoadTestSession",
		"role": "PUBLISHER",
		"audio": true,
		"video": true,
	}
}

  • RESPONSE

{
	"connectionId": "con_FmtswrvbkT",
	"workerCpuUsage": 10.00
}

Moreover, you can customize the request with many of the parameters that we can found in OpenVidu Browser library.

Create Stream Manager: Body Parameters

To make the load test completely functional, the browser-emulator service also accept others extra body parameters.

{
	"openviduUrl": "your OpenVidu hostname",
 	"openviduSecret": "your OpenVidu Secret",
    "browserMode": "'EMULATE' or 'REAL'",
	"properties": Properties JSON object << See properties list >>
}
Body parameters
Properties Type Description
openviduUrl String OpenVidu URL.
openviduSecret String OpenVidu secret.
browserMode String If EMULATE the service will emulate a browser. If REAL, the service will launch a Chrome browser docker container. Default EMULATE Choosing EMULATE, you must ensure that OpenVidu aren't forcing H264 coded
properties JSON Object See properties object
Properties JSON object
Properties Type Description
userId * String Participant name
sessionName * String Session name
audio * Boolean If role is PUBLISHER and you want to initially publish to the session with the audio unmuted or muted See publisher property
video * Boolean If role is PUBLISHER and you want to initially publish to the session with the video enabled or disabled. See publisher property
token String Your custom token. If it is empty the service will create one.
role String Stream Manager role:PUBLISHER or SUBSCRIBER. Default PUBLISHER
resolution String Resolution of the video. Default 640x480. See publisher property
recordingOutputMode String COMPOSED or INDIVIDUAL
frameRate Number (0-30) Desired framerate of the video in frames per second. Default 30
recording Boolean If browserMode is REAL and you want record the Chrome browser using ffmpeg. Otherwise, If browserMode is EMULATE and you have started browser.emulator with KMS user type (see worker running options) Default false.
showVideoElements Boolean If browserMode is REAL and you want show videos elements into the app running in Chrome. Default true
headless Boolean If browserMode is REAL and you want launch a headless Chrome. Default false. See Headless Chromium

DISCONNECT ALL PARTICIPANTS

Disconnect all participants

DELETE PARTICIPANT (by connectionId)

Delete a single participant by its connectionId

DELETE PARTICIPANTS (by ROLE: PUBLISHER or SUBSCRIBER)

Delete all participants with the specified ROLE

DELETE PARTICIPANTS

Delete all participants

START QOE ANALYSIS

Start QoE Analysis if qoeAnalysis has been set to true when instance was initialized. Use after deleting all participants. Results will be uploaded to ELK when finished.