Forest @forestjohnson@pixie.town

For fun hacking stream making a social sharing web app to use for demos. https://stream.sequentialread.com
Music today: 🐌🏡 Snail's House ✨💨

first garden harvest of the season, arroooooooogula and radishes

@gabek

uxbrain tip: even if you think your API requests wont take long, they eventually will for someone somewhere.

streaming some debugging and development today. music: 🐌🏡 Snail's House ✨💨

https://stream.sequentialread.com/

I have been trying to make a recognizable emoji representation of each one of my services for the readme file & the last log line after startup.

I was looking for one for Threshold (https://git.sequentialread.com/forest/threshold#threshold) but unfortunately there are no reliable emojis depicting walls, open doors, fences, open gates, or anything similar to that besides the ⛩️ shinto shrine emoji.

I read a little bit about ⛩️ on wikipedia and it seems like a conceptual fit for threshold in theory, but I don't really know much about shintoism and I'm not involved with it so I opted against using it.

So I settled on this instead, and I like it.

🏔️⛰️🛤️⛰️🏔️

@f0x

@amolith@nixnet.social

I hear you, but isn't a docker-compose.yml file plus a pile of Dockerfiles just a codified expression of "how do I set up this app" ?

The same could be said of a collection of ansible roles, .deb/.rpm packages, or raw scripts. Typically one can port/translate between those paradigms easier than one can learn how to set up the app from scratch, i.e. by reading the application code.

No matter which one a developer picks, if they pick just one, it's guaranteed to alienate a portion of their audience. I think folks go for docker because it alienates **a smaller number of people** on average.

There are a lot of reasons why docker got so popular; a lot of it has to do with how hard the "proper" system admin stuff really is. No one ever did the hard work to try to make unix admin easier and more normie-friendly until docker. Hate the game, not the player.

That said, obviously in an ideal world, all open source developers could be well-supported enough to maintain at least a couple different supported installation methods.

@fribbledom ive started just removing the cookie popups with my ad blocker 🙃

@gabek Thanks for this BTW, I put a disclaimer saying that what I did is definitely not required to be able to use Owncast, it was just for fun.

@gabek Mali-T628 MP6

I believe that the thing I'm using can in fact do real time video transcoding, even on the CPU (8 cores baybeee) But I wanted to experiment with something where it only gets encoded once -- something that would work even when transcoding on the server is not an option.

Just published a write-up on my custom Owncast fork, how I run my own stream using a 5 watt computer.

https://sequentialread.com/how-i-optimized-my-owncast-stream-to-run-on-a-raspberry-pi-part-2/

@gabek Yes, using a torrent-ish distribution mechanism would 100% introduce tons of stream latency. I think that's a price many folks would be willing to pay, however, if the alternative is what you mentioned, paying for / giving your data to a CDN.

I am interested in it precisely **because** it is hard / no good solution for it exists today, not despite that 😈

I saw the comment on the owncast issue stating that peertube already added p2p live streaming -- this is simply not true! PeerTube is using HLS just like owncast is. It is 100% handled by the server.

@gabek Thanks. Yeah I really like how HLS is simply based on HTTP & works well with existing web servers. I'm not trying to push this as a new feature for owncast really, I just wanted to upgrade my current jank solution (for streaming from a potato web server) with a slightly nicer jank solution 😀

Next up on my list would be some sort of p2p acceleration so we can stream to 1000s of viewers

@gabek

I saw you added features for controlling the latency -- have you tried to optimize for the smallest stream delay ? I'm curious if what I've been doing here is even worth anything in that department. I was hoping I could get it to be even faster, but it sounds like the way HLS works sorta limits it fundamentally.

https://git.sequentialread.com/forest/owncast/src/branch/develop/core/directhls/directhls.go

https://github.com/ForestJohnson/owncast/blob/develop/core/directhls/directhls.go

I was able to modify owncast to stream HLS segments that are outputted by OBS directly instead of re-encoding the video on the server -- got it down to 10 seconds stream delay in my experimental test!

its saaaaaturday

drinking beer and messing around with owncast code, direct HLS streaming from OBS any% attempts

https://stream.sequentialread.com/

@gabek

Have you ever looked at doing this with Owncast?

https://obsproject.com/forum/resources/how-to-do-hls-streaming-in-obs-open-broadcast-studio.945/

OBS stream direct to HLS

If you haven't given it a shot yet, I'd love to tinker with this!

Also, since I lost the context of the other part of the post and you cant edit posts on mastodon, I think I should clarify that by "needs a Kubernetes" I don't mean "needs a big complex thing that takes millions of man hours to create" or "Needs a docker-based distributed clustering/scheduling system"

If you've never worked with Kubernetes you might not know this, but Kubernetes itself isn't actually an implementation, it's just a bunch of interfaces that define standard ways that all the parts of said distributed clustering/scheduling system can work together. What people colloquially refer to as "Kubernetes" is actually the interfaces + probably the reference implementation of each one of the interchangeable parts.

But the magic is that you can swap that parts out with your own if you want. You can upgrade 1 part without breaking the others. You can have a proliferation of the "flavors" of Kubernetes similar to the proliferation of linux distributions.

I just think that we as "small web" developers should be mindful of this trend tech has followed since its inception, since the unix days -- small, simple programs that can work together.

I imagine a "small tech kubernetes" as a set of interfaces that all of our projects can conform to so they can interoperate and proliferate. So other developers can take them and adapt them to other use cases without losing as much interoperability .

explanation of what I meant by "secure-attestation-based" at the end:

The first part talked a lot about servers and ownership over processes & how its related to power, having power over other people.. About how in academic Computer Science right now, no one has figured out how to make a process that operates directly on data but does not "own" that data (can censor, falsify, or spy on it).

Also mentioned how DRM today uses an ersatz solution for this called CPU secure enclaves and secure remote attestation. And how some projects (Signal's secure contact discovery) have started using the same tech to try to liberate people, but I have a lot of doubts about how viable this ersatz solution is in the long term / how viable it is for more widespread use.