How can I distribute Pyav jobs on many servers with Spark

[vtexier]

Overview

I am using Spark (pyspark) to distribute big data jobs across servers. I would like to distribute Pyav encoding and filter features with Spark on many servers .

Expected behavior

The distributed function will execute Pyav features which need to be aware of previous and/or next frames in the GOP.

How to implement those features with GOP acknowledgement?

Actual behavior

Isolated frame in a separate job can not be used alone.

Investigation

A solution to give Spark the context by serializing the Frame object with pickle is not enough, see #652

Research

I have done the following:

• [*] Checked the PyAV documentation
• [*] Searched on Google
• [*] Searched on Stack Overflow
• [*] Looked through old GitHub issues
• [*] Asked on PyAV Gitter
• ... and waited 72 hours for a response.

https://gitter.im/mikeboers/PyAV?at=5eab0dd59f0c955d7d97bbb1

Additional context

@koenvo, may be you can help me on this as you propose in #652 (comment)

[jaredmcgrath]

Hello!

Did you ever get to the bottom of this issue? I came across the discussion. The use case described by @koenvo in this comment is exactly what I'm trying to recreate. I.e. splitting the reading of packets into a separate thread from where the decoding happens.

[vtexier]

Nope. Sorry. I have switched to another software to achieve this.

[diogommartins]

Which software?