HTML5 Safari live broadcast vs not

Can you post the headers sent by the server both with and without the PHP script? I’m wondering if the PHP script is sending a different Content-Type than serving the files normally.

It would also be a good idea to specify the type attribute on the source elements, so the browser does not have to download both clips to determine if it can play them.

I cannot reproduce your problem. I have tried to recreate the problem in Safari 4.0.4, and the current WebKit nightly, with the following test page. I am simply using mod_rewrite to dispatch to different formats based on a parameter instead of PHP, but I don’t think that should make a difference, unless somehow PHP is modifying the file.

<!DOCTYPE html>
<title>Auido test</title>
<audio controls autobuffer>
  <source src="https://stackoverflow.com/questions/1995589/gnossienne-no-1?foo=bar&format=.mp4">
  <source src="gnossienne-no-1?foo=bar&format=.ogg">
</audio>

Can you try my example out and let me know if it works for you?

edit Ah. After poking around at it a bit more, it appears that the problem is due to an odd way that the <audio> element in Safari behaves in attempting to determine the size of the content.

Here’s an excerpt from a packet capture of Safari upon encountering an <audio> element pointing to a file served directly from Apache. As you can see, it first tries to fetch the first two bytes of the media, presumably so it can get a Content-Length back, and possibly other headers. It then tries to fetch the whole thing. Then, inexplicably, it tries to fetch the first two bytes again, but passes appropriate caching headers to get a “304 Not Modified” response. And finally, still inexplicably, it fetches the last 3440 bytes of the file all over again. It does all of these in separate TCP connections, which adds considerable overhead, in addition to the overhead of fetching the data a couple of times.

GET /stackoverflow/audio-test/say-noid3?foo=bar&format=.mp3 HTTP/1.1
Host: ephemera.continuation.org
Range: bytes=0-1
Connection: close
User-Agent: Apple Mac OS X v10.6.2 CoreMedia v1.0.0.10C540
Accept: */*
Accept-Encoding: identity
Cookie: [redacted]

HTTP/1.1 206 Partial Content
Date: Tue, 05 Jan 2010 02:12:48 GMT
Server: Apache
Last-Modified: Tue, 05 Jan 2010 02:02:08 GMT
ETag: "b2a80ad-11f6-47c6139aaa800"
Accept-Ranges: bytes
Content-Length: 2
Content-Range: bytes 0-1/4598
Connection: close
Content-Type: audio/mpeg

# 2 bytes of data

GET /stackoverflow/audio-test/say-noid3?foo=bar&format=.mp3 HTTP/1.1
Host: ephemera.continuation.org
Range: bytes=0-4597
Connection: close
User-Agent: Apple Mac OS X v10.6.2 CoreMedia v1.0.0.10C540
Accept: */*
Accept-Encoding: identity
Cookie: [redacted]

HTTP/1.1 206 Partial Content
Date: Tue, 05 Jan 2010 02:12:48 GMT
Server: Apache
Last-Modified: Tue, 05 Jan 2010 02:02:08 GMT
ETag: "b2a80ad-11f6-47c6139aaa800"
Accept-Ranges: bytes
Content-Length: 4598
Content-Range: bytes 0-4597/4598
Connection: close
Content-Type: audio/mpeg

# 4598 bytes of data

GET /stackoverflow/audio-test/say-noid3?foo=bar&format=.mp3 HTTP/1.1
Host: ephemera.continuation.org
Range: bytes=0-1
Connection: close
User-Agent: Apple Mac OS X v10.6.2 CoreMedia v1.0.0.10C540
Accept: */*
Accept-Encoding: identity
Cookie: [redacted]
If-None-Match: "b2a80ad-11f6-47c6139aaa800"
If-Modified-Since: Tue, 05 Jan 2010 02:02:08 GMT

HTTP/1.1 304 Not Modified
Date: Tue, 05 Jan 2010 02:12:49 GMT
Server: Apache
Connection: close
ETag: "b2a80ad-11f6-47c6139aaa800"

# no data

GET /stackoverflow/audio-test/say-noid3?foo=bar&format=.mp3 HTTP/1.1
Host: ephemera.continuation.org
Range: bytes=1158-4597
Connection: close
User-Agent: Apple Mac OS X v10.6.2 CoreMedia v1.0.0.10C540
Accept: */*
Accept-Encoding: identity
Cookie: [redacted]

HTTP/1.1 206 Partial Content
Date: Tue, 05 Jan 2010 02:12:49 GMT
Server: Apache
Last-Modified: Tue, 05 Jan 2010 02:02:08 GMT
ETag: "b2a80ad-11f6-47c6139aaa800"
Accept-Ranges: bytes
Content-Length: 3440
Content-Range: bytes 1158-4597/4598
Connection: close
Content-Type: audio/mpeg

# 3440 bytes of data

Anyhow, on to how it deals with the output of your PHP script. Here, Safari again tries to download the first two bytes, but your script ignores the Range request and returns the whole thing. Apparently, WebKit doesn’t like that, and so it tries again, without the Range request. Again, your script sends the full contents. Safari now tries once more, adding an Icy-Metadata header, which indicates it thinks that it’s downloading a stream and wants streaming metadata to be sent. It finally accepts the output of that, and the <audio> element can play.

GET /say.php?text=this%20is%20a%20test&format=.mp3 HTTP/1.1
Host: tts.mindtrove.info
Range: bytes=0-1
Connection: close
User-Agent: Apple Mac OS X v10.6.2 CoreMedia v1.0.0.10C540
Accept: */*
Accept-Encoding: identity

HTTP/1.1 200 OK
Date: Tue, 05 Jan 2010 02:14:28 GMT
Server: Apache
X-Powered-By: PHP/5.2.10
Content-Length: 4598
Connection: close
Content-Type: audio/mpeg

# 4598 bytes of data

GET /say.php?text=this%20is%20a%20test&format=.mp3 HTTP/1.1
Host: tts.mindtrove.info
Connection: close
User-Agent: Apple Mac OS X v10.6.2 CoreMedia v1.0.0.10C540
Accept: */*

HTTP/1.1 200 OK
Date: Tue, 05 Jan 2010 02:14:28 GMT
Server: Apache
X-Powered-By: PHP/5.2.10
Content-Length: 4598
Connection: close
Content-Type: audio/mpeg

# 4598 bytes of data

GET /say.php?text=this%20is%20a%20test&format=.mp3 HTTP/1.1
Host: tts.mindtrove.info
Accept: */*
User-Agent: Apple Mac OS X v10.6.2 CoreMedia v1.0.0.10C540
Icy-Metadata: 1
Connection: close

HTTP/1.1 200 OK
Date: Tue, 05 Jan 2010 02:14:28 GMT
Server: Apache
X-Powered-By: PHP/5.2.10
Content-Length: 4598
Connection: close
Content-Type: audio/mpeg

# 4598 bytes of data

In summary, it appears that Safari (or more accurately, QuickTime, which Safari uses to handle all media and media downloading) has a completely braindamaged approach to downloading media. Something in the way you send your data back, probably the fact that you don’t respond to Range requests, makes it think that you are sending streaming media, causing it to download the content repeatedly (though even when confronted with a server that does respond to a Range request, it still does several more requests than it really needs to).

My advice would be to try to respond appropriately to Range requests; when serving up media, browsers will likely use them to try to minimize bandwidth, by only buffering as much as they need to be able to play through (although have the autobuffer attribute which indicates that you would like them to buffer the whole thing, browsers may ignore that). I would recommend using X-Sendfile to let Apache deal with serving the file, caching, and range requests, but you appear to be on Dreamhost, which doesn’t have mod_xsendfile installed, so you’re going to have to roll your own Range handling.

Leave a Comment