feat: replace native ffmpeg with bundled wasm version

[miraclx]

Closes #294.

This patches #297 a bit to avoid depending on modified versions of @ffmpeg/{core,ffmpeg}.

All credit to @hard-shutdown for investigating this.

[github-actions]

---

🐋 🤖

A docker image for this PR has been built!

docker pull freyrcli/freyrjs-git:pr-305

Base Branch (master)
This Patch

---

What's this?

This docker image is a self-contained sandbox that includes all the patches made in this PR. Allowing others to easily use your patches without waiting for it to get merged and released officially.

For more context, see https://github.com/miraclx/freyr-js#docker-development.

[bluefoxy009]

Hello, may I ask what node version this uses, as well as how you are able to run the embedding queue without running into errors? Thanks!

EDIT: That was what the new ffmpeg instances were used for, see the logs from the tests.

[miraclx]

Yeah, so I found out that ffmpeg-wasm works only for node < v18 because node v18 introduces the fetch() function – https://nodejs.org/en/blog/announcements/v18-release-announce/#fetch-experimental.

The hack that worked was to remove that function from the global scope. So when ffmpeg-core checks for it, it just defaults to reading it as a file:

freyr-js/cli.js

Line 974 in c500eaf

delete globalThis.fetch;

embedding queue without running into errors?

Didn't run into any embedding errors. The only issue I see now is not being able to run one command at a time.

I saw your approach to fix that was to load ffmpeg afresh every time a track needs to be transcoded, my concern with that approach is memory usage.

Because for a 100-track playlist, it would load the 8 MB WASM blob, allocate resources to assign context etc. 100 times.

[bluefoxy009]

That was an extremely smart trick regarding the fetch api :) Then again, I believe that most computers running this will have enough RAM to run the 8 MB contexts. Maybe another trick would be to wipe MEMFS to help clear memory usage because running ffmpeg.exit() seems to make freyr exit with code 1

[miraclx]

Resolved. Made it so that there's only N number of ffmpeg instances at any given point in time. Defined by the concurrency.encoder key in the conf.json.

By default, this is 6 concurrent processes. Spawned on demand and reused for more tracks, if need be. Can be less than the defined value, but never more than.

[bluefoxy009]

Excellent idea, thanks for the ideas!

[miraclx]

Finally, found 6 hours to investigate the issue with this. 😪
Fixed in #335.