Reputation: 16831
I'm attempting to create a rudimentary "DVR" for an HTML5 video element by utilizing MediaRecorder
, MediaSource
, and SourceBuffer
. At the moment this is just a proof of concept. However since many projects like HLS.js take advantage of the HTML5 video element, I believe this would have wide-spread value.
Here's the gist of my code:
<html>
<head>
</head>
<body>
<video id="src-video" src="http://localhost:8080/video/source.mp4" autoplay></video>
<video id="dvr-video"></video>
<input id="seekbar" type="range" min="-120" max="0" value="0" />
<script>
var mr; // MediaRecorder
var ms = new MediaSource();
var srcBuf; // SourceBuffer
var srcUrl = URL.createObjectURL(ms);
var srcVid = document.getElementById("src-video");
var dvrVid = document.getElementById("dvr-video");
var dvrData = []; // array of ArrayBuffer
var queue = [];
ms.addEventListener("sourceopen", sourceOpen);
srcVid.addEventListener("playing", setupMediaRecorder);
dvrVid.src = srcUrl;
var seekBar = document.getElementById("seekbar");
seekBar.addEventListener("change", function(e) {
// Destroy the old media source and make a new one
URL.revokeObjectURL(srcUrl);
srcBuf = null;
ms = new MediaSource();
ms.addEventListener("sourceopen", sourceOpen);
srcUrl = URL.createObjectURL(ms);
body.removeChild(dvr);
dvr = document.createElement("video");
body.insertBefore(dvr, seekBar);
dvr.src = srcUrl;
});
function sourceOpen()
{
// Create the source buffer
if (!srcBuf)
{
srcBuf = src.addSourceBuffer('video/webm; codecs="opus,vp8"');
srcBuf.mode = "sequence";
}
srcBuf.addEventListener('updateend', function() {
if ( queue.length ) {
srcBuf.appendBuffer(queue.shift());
} else {
dvr.play();
}
}, false);
// Add all fragments in cache
var start = dvrData.length + parseInt(seekBar.value);
queue = [];
for( var i = start; i < dvrData.length; i++ )
{
if (dvrData[i])
queue.push(dvrData[i]);
}
if (queue.length)
srcBuf.appendBuffer(queue.shift());
}
function setupMediaRecorder()
{
var stream = srcVid.captureStream()
mr = new MediaRecorder(stream);
mr.ondataavailable = function(e) {
// Convert the Blob to an ArrayBuffer
var fileReader = new FileReader();
fileReader.onload = function() {
// Append this ArrayBuffer to our playing video
if (srcBuf)
{
if (srcBuf.updating || queue.length)
queue.push(this.result);
else
srcBuf.appendBuffer(this.result);
}
// And to our historical array (for seeking purposes)
dvrData.push(this.result);
if (dvrData.length > 120) {
// Keep only 2 minutes of data
dvrData.splice(0, 1);
}
};
fileReader.readAsArrayBuffer(e.data);
};
mr.start();
// Record 1-second chunks
setInterval(function() {
mr.requestData();
}, 1000);
}
</script>
</body>
</html>
When the page first loads, the "live" video element begins to play, and 1 second later the "dvr" element begins to play - with a 1-second delay. So it seems to be working at first.
The moment I perform a seek, the dvr element goes black and I get the following error in the console (line number may not match exactly with code above):
Uncaught DOMException: Failed to execute 'appendBuffer' on 'SourceBuffer': This SourceBuffer has been removed from the parent media source.
at SourceBuffer.<anonymous> (http://localhost:8080/video/dvr.html:80:13)
Looking at chrome://media-internals
for further details, I see the following for the DVR player:
+-------------+----------------+--------------------------------------------------------------------------------------------
| Timestamp | Property | Value
+-------------+----------------+--------------------------------------------------------------------------------------------
| 00:00:00 00 | origin_url | http://localhost:8080/
| 00:00:00 00 | frame_url | http://localhost:8080/video/dvr.html
| 00:00:00 00 | frame_title |
| 00:00:00 00 | url | blob:http://localhost:8080/cec4134a-4498-43c5-8321-3743761636ac
| 00:00:00 00 | info | ChunkDemuxer: buffering by DTS
| 00:00:00 00 | pipeline_state | kStarting
| 00:00:00 03 | error | Unexpected element ID 0xa3
| 00:00:00 03 | error | Append: stream parsing failed. Data size=112300 append_window_start=0 append_window_end=inf
| 00:00:00 08 | pipeline_error | CHUNK_DEMUXER_ERROR_APPEND_FAILED
| 00:00:00 10 | pipeline_state | kStopping
| 00:00:00 10 | pipeline_state | kStopped
+-------------+----------------+--------------------------------------------------------------------------------------------
The Unexpected element ID 0xa3
seems to be the culprit. Although for some reason this error wasn't thrown when the page first loaded (I'm appending the same ArrayBuffer
s to my SourceBuffer
, so if they didn't throw this error before I don't know why they're throwing it now)
Looking up 0xa3
with respect to the WEBM format, it sounds like this refers to a "SimpleBlock" -- https://chromium.googlesource.com/webm/libwebm/+/libwebm-1.0.0.26/webmids.hpp -- I don't know why this would throw an error?
video
element and re-creating itSourceBuffer
is somehow modifying the ArrayBuffer
s when they're played, I tried appending copies instead of the original objects (srcBuf.appendBuffer(queue.shift().slice(0));
)segment
and sequence
modes on the SourceBuffer
SourceBuffer
(before I had data) and entering a "media complete" state, I tried pausing the DVR elementSo far I haven't had any luck getting DVR to work correctly. What am I missing?
Upvotes: 4
Views: 1841
Reputation: 16831
After a lot more experimentation I finally figured out the issue.
WEBM files are effectively binary-encoded XML files. The schema looks something like this:
<EBML>
<EBMLVersion>...</EBMLVersion>
<EBMLReadVersion>...</EBMLReadVersion>
<EBMLMaxIDLength>...</EBMLMaxIDLength>
<EBMLMaxSizeLength>...</EMBLMaxSizeLength>
<DocType>...</DocType>
<DocTypeVersion>...</DocTypeVersion>
<DocTypeReadVersion>...</DocTypeReadVersion>
</EBML>
<Segment>
<SeekHead>
<Seek>...</Seek>
<Seek>...</Seek>
<Seek>...</Seek>
</SeekHead>
<Void></Void>
<Info>...</Info>
<Tracks>
<TrackEntry>
<Video>
<Colour>
<MatrixCoeffciient>...</MatrixCoefficients>
...
</Colour>
</Video>
<Audio>
...
</Audio>
</TrackEntry>
</Tracks>
<Cues>
<CuePoint>
<CueTrackPositions>...</CueTrackPositions>
</CuePoint>
</Cues>
<Cluster>
<Timecode>...</Timecode>
<SimpleBlock>...</SimpleBlock>
<SimpleBlock>...</SimpleBlock>
<SimpleBlock>...</SimpleBlock>
<SimpleBlock>...</SimpleBlock>
...
</Cluster>
<Cluster>
<Timecode>...</Timecode>
<SimpleBlock>...</SimpleBlock>
<SimpleBlock>...</SimpleBlock>
<SimpleBlock>...</SimpleBlock>
<SimpleBlock>...</SimpleBlock>
...
</Cluster>
...
</Segment>
The way I was reading the data, the first block contained all of the header information (EBML data, segment data, tracks, timecode, etc) and all later chunks were just a stream of <Cluster>
and <SimpleBlock>
tags (with an odd <Timecode>
every now and then)
Ultimately what I had to do was build a rudimentary demuxer to parse through the EBML file and extract the header information. Then whenever I performed a seek, this header information was injected into the buffer before any video data.
Don't bother. MSE is terrible and this was a 72 hour nightmare. Save yourself the headache.
Upvotes: 3