@Viss @foone this is more of a thing than you might...

@Viss @foone this is more of a thing than you might expect. I've seen a few high-volume realtime media distribution backends that use Redis as a rolling video stream cache.

I'm particularly unsurprised to see it here because there are published tools for realtime ML media analytics using redis:

https://github.com/RedisGears/EdgeRealtimeVideoAnalytics

Like 19 August at 20:56 | Wall-to-wall | Open on chaos.social

4 comments

Viss replied to Graham

@gsuberland @foone my brain just fell out

19 August at 20:56 | Open on mastodon.social

Graham Spookyland🎃/Polynomial replied to Viss

@Viss @foone there is some method in the madness. if you've got a lot of transient video data, and you need access to a rolling window of it (either for buffering/stability purposes or for realtime analytics), storing it to disk ends up costing a fortune because you'll end up running headlong into DWPD limits on drives and having to swap them out constantly. but with RAM there's no such wear. for a few hundred concurrent clients you can do it on a single consumer desktop PC worth of RAM.

19 August at 21:01 | Open on chaos.social

Ganonmaster replied to Graham

@gsuberland @Viss @foone I first learned of this when I read that FFMPEG supports AMQP. https://ffmpeg.org/ffmpeg-protocols.html#amqp

19 August at 21:05 | Open on open3dlab.social

Graham Spookyland🎃/Polynomial replied to Graham

@Viss @foone Redis isn't required to flush to disk (the whole dataset is in-memory) and it supports blob storage, so it's not the worst option. my guess is someone ran the `save` command at some point (or they were doing periodic saves because they didn't separate out nonvolatile data into a separate redis instance) and foone's seeing that last snapshot.

19 August at 21:05 | Open on chaos.social