Audio dropouts on lossless/uncompressed formats

[Sean M. Pappalardo]

When playing a FLAC, WAV, or AIFF file, it seems like CachingReader is sometimes waiting until it's out of data before requesting it from the disk, causing an audible dropout with a number (usually 6) of lines printing to the console like:

Debug []: Couldn't get chunk 1019 in read() of [ 16706494 , 16706964 ] chunks 1019 - 1019

It happens every so often while just playing through a track. More lines print the faster the track is played. (One particularly bad case was playing at +80%, it missed 6 chunks in a row.) This console output coincides with the access light on my disk.

This doesn't happen every time it needs data, unless the disk just sometimes happens to be fast enough to meet the deadline. If I play through the track again, the problem doesn't occur. (Suspect system cache helping out here.)

Tested on an HP EliteBook 8440p:
- Intel Core i5 M520 2.4GHz
- 2GB RAM, 95MB free with Mixxx, JACK, and FIrefox running
- Debian Squeeze
- Mixxx trunk r3166 (pre-1.11.0,) happens in 1.10 branch as well
- Music files on external 500GB RAID-1 drive connected via eSATA with XFS file system, just under 73MB/s read speed according to hdparm. (Cached reads are ~2500MB/s.)

[mutil (mutil)]

Is this anyhow related to lp:925616 ?
I get tons of these "Couldn't get chunk" debug messages from caching reader (for every file format), before it crashes.

[Sven Killig (sven-killig)]

Importance: "Medium"? A show stopper, I'd say.
Can I help somehow by compiling and testing it with different compile time #define-s ?

[Sven Killig (sven-killig)]

#define CHUNK_LENGTH 65536
//#define CHUNK_LENGTH 524288

looks promising. Is it worth trying?

[RJ Skerry-Ryan (rryan)]

Hi Sven,

It's only marked medium because it's very unclear how broadly this affects people. I can't reproduce it myself.

That CHUNK_LENGTH define changes the granularity at which Mixxx can cache decoded sections of audio. I don't think it will have much of an effect on the problem but give it a try if you're curious!

FYI it's often very normal to see 'couldn't get chunk' in your log. If I had to guess this bug has very little to do with CachingReader and more to do with the decoding code since CachingReader doesn't do anything special based on the format of the audio.

If it's simply the case that the decoding library is taking too long to decode then that would make sense for FLAC since it has to do more work decompressing the audio but it doesn't make sense at all for WAV since that's stored uncompressed on disk so there's very little work to do.

Honestly, I just assumed this is Sean's RAID array being super flaky but now that other people can reproduce it I guess that's not the case ;).

Sven, mutil : Are you absolutely sure this only happens when using FLAC/WAV/AIFF and not MP3? If you can ever reproduce this with MP3 then this is not the bug you're seeing and it's likely your latency is just slightly too low or something.

[mutil (mutil)]

Hey Ryan,
No, for me this did not happen only with FLAC/WAV, I think it glitched with every file type I used, but maybe a little more with these formats. Probably a different bug, related to the CPU shortage of a netbook.

[Uwe Klotz (uklotzde-deactivatedaccount)]

I remember that I've seen similar log messages in Mixxx versions prior to 1.12. After refactoring both the SoundSource API and CachingReader(Worker) in branch master I don't experience any of those messages. Should be fixed for future releases >1.12.

Still relevant for 1.12?

[Owen Williams (ywwg)]

It's not feasible to make soundsource fixes for 1.12, unfortunately. Better to push 1.12 out and follow up with the next release to get these edge cases addressed.

[Uwe Klotz (uklotzde-deactivatedaccount)]

- Won't fix for 1.12
- Fixed on master = current development branch

[Uwe Klotz (uklotzde-deactivatedaccount)]

I will mark the 1.11 target of this bug as Invalid, because Won't Fix is not available.

[Swiftb0y (swiftb0y)]

Mixxx now uses GitHub for bug tracking. This bug has been migrated to:
https://github.com/mixxxdj/mixxx/issues/6459