Following the video transcoding work included in v0.10.0, uploading an image and
restarting Celery resulted in the image being marked as failed, even after it
had been initially successfully processed.
The issue was that the initial processing task was not being acknowledged by the task queue following the introduction of the
`CELERY_ACKS_LATE` setting. It's not clear why this is the case, but reverting
the setting fixes this issue and doesn't negatively impact video processing.
Issue is that Werkzeug > 1.0.0 has removed werkzeug.contrib.atom.AtomFeed,
making it difficult to use a distribution-packaged version of werkzeug. To solve
this, I've replaced use of werkzeug.contrib.atom.AtomFeed with
feedgenerator.Atom1Feed.
After the change, the only major difference between the feeds before and after is
that they use <summary> instead of <content>. Minor differences include no longer
adding 'type="text/html"' on some <link> elements and no "xml:base" attribute on
<entry> elements. I don't think these differences will have any noticable
effect.
Tested on Liferea feed reader.
Previously had partial docs for Fedora 31. This updates to Fedora 33, adds
support for audio and video and adds dependencies to allow the test suite to run
to completion.
The `audiotospectrogram` module is a complete rewrite of the existing spectrogram
code with support for Python 3. This allows us to drop the bundled `freesound`
library and Python 2-only `audioprocessing` and `spectrogram` modules.
Signed-off-by: Ben Sturmfels <ben@sturm.com.au>
By increasing the limit from 8m to 100m, we should immediately fix initial
problems for people trying to upload audio or video. From there, they can read
the documentation more closely when they try to upload larger files.
This change starts Celery only after RabbitMQ is available. It also returns the
unnecessarily low BROKER_HEARTBEAT setting to default of 120.0 to help prevent
connections being reset on machines under load.
