Down the Linux Rabbit Hole  ## Document Notes ## Highlights - **Choose Containers By Tradeoffs** - Prefer Docker for usability when you want simple, reliable container workflows in a home lab. - Use Podman when enterprise rootless security matters, but expect UID mapping and usability friction. Transcript: Alex Kretzschmar Yeah. Adam Stacoviak How do you feel about the Podman-Docker war? Do you feel like it's a war that's won? Are you a Podman person? What's your feels? I'm not. Alex Kretzschmar You know, I'm ex-Red Hat, so I really should be pro-Podman, shouldn't I? Yeah. The thing that Podman misses for me is it's a very purist implementation. So it's extremely technically sound. What it misses for me, though, is some of the user spice on top, like that last 10% that Docker closed in terms of usability. Because all the primitives for Docker pretty much were there before Docker came along. It was just the packaging of Docker with the standardized image format that really made it take off. The trouble with Podman though is when you want to do basic stuff, like even just mounting a volume, you've got to do this user ID shuffle because I'm talking specifically about rootless Podman here. You've got to do this user ID shuffle because the user IDs inside the container don't map to the IDs on the host properly unless you do this shift. And so it ends up being this world of complicated UID scripts, shuffling nonsense that I just haven't got time for. Adam Stacoviak I guess I don't, I'm not that much of a Docker user to know exactly what you're talking. Would definitely mapped uids to uids but i never really hit the issue i do know that if the if you mismatch on ownership there's an issue but that's i sort of leave it to the ai or the you know Alex Kretzschmar Stack overflows or something else that's like slightly above my pay grade well i'll break it down as best i can just really simply so in a in a rootless Podman container, you end up shifting The user IDs by roughly 100,000 or so. It depends on how you've got it configured. So on your host, you would probably have a user ID of something like 1,000. Inside the container, it would be the user ID 100,000 and 1,000. So like 101,000. So you've got to find a way to map those IDs from what the container sees to what the host can speak so that the file permissions for those bind mount volumes actually work. And that whole mess, the fact I've just had to explain that to you is exactly my point. ([Time 0:08:23](https://share.snipd.com/snip/8e59f0ee-8c73-4500-9533-c5c3ba5cba98)) - **Leaving Google Photos For Self-Hosting** - Alex migrated off Google Photos and deleted ~250,000 images after self-hosting Image and Image Frame. - He uses Tailscale to expose a self-hosted album to family photo frames and reclaimed data sovereignty. Transcript: Adam Stacoviak Yes. Tell me about this. Alex Kretzschmar Done by Alex and his team at Futo and Lewis Rossman in Austin, no less. So they're on your doorstep, dude. So I set this up and I was like, I can delete Google Photos now. And I finally did it. I've actually finally deleted my quarter of a million pictures in Google Photos. It took me all weekend. You have to go through the web UI and select a maximum of like 10,000 images at a time. It's really one million for real. Adam Stacoviak Yeah. Alex Kretzschmar It was like quarter of a million pictures. Yeah. It was quite a lot, but the Google make it really difficult to leave. But anyway, anyway, I'm off, I'm off the Google source. Finally, it's, it's about time the self-hosted guy actually did that. But I was like, well, I bought my family for Christmas when my daughter was born five years ago, some of those little Google picture frames. And so we just had this shared album where I just dropped photos in every few months and then it would update their picture frames remotely. And I'm like, Oh, oops, I guess I've just broken. So I've just broken that workflow for them. So I just came in just a favorites basically effectively yeah and so I came across this project called image frame and um this lets you turn any device because it's just a web view so you Can run it on android ios and apple tv and android tv just a web browser even and it lets you turn any of those devices into essentially a google photos frame so I can just drop pictures into An image album on my self-hosted server, connect this Android tablet that I'm going to give to my family for Christmas back over tail scale to my house. And the same functionality exists, but it's all completely self-hosted and completely, I own the whole, the whole shebang. It's, that's what I love about this stuff. ([Time 0:25:05](https://share.snipd.com/snip/cc44aec7-02db-4fb3-8506-9f2c9ea732a9)) - **Backups Move Beyond RAID** - Don’t rely on RAID alone; maintain an offsite copy of important data. - Use ZFS replication over SSH (e.g., zrepl) with Tailscale or a remote host to mirror your pool offsite. Transcript: Adam Stacoviak So how do you do your backups, Adam? Gosh, I don't have backups, Alex. I got RAID. Okay, I'm just kidding. I really don't have off-site backups, okay? I'm not practicing 3-2 I'm in a situation. It could go down right in this moment. It could catch on fire over there, and I will have to say, you know, get the thing, get the extinguisher and put it out and my stuff is gone. So I don't have backups. Alex Kretzschmar Oh, well, there goes one of my pitches for you would be an offsite backup. Well, that could be a pitch though. Adam Stacoviak I need an offsite backup. So what would you do? How would you do it? Alex Kretzschmar Well, I have a friend in Canada, so I'll tell you how I do it. And so I have a ZFS array in the basement, which just got, I think five, 20 terabyte hard drives in it. And then I use, um, it's a program called Z REPL, Z-R for Z replicate, I guess. And it, it essentially mirrors all of the ZFS blocks to Canada, which is where, friend lives, and he's got a similar sized server. He actually runs an MSP, so he just carved out a few use of rack space for me to use there. ([Time 0:31:21](https://share.snipd.com/snip/86914b6d-e244-428e-994b-90d8d804e96a)) - **Cheap Offsite ZFS Backups** - For affordable offsite ZFS backups, consider zfs.rent: ship a drive and pay per slot monthly. - Use the provided VM and IP as a ZFS replication target to keep offsite copies cheap and simple. Transcript: Alex Kretzschmar A couple of ways. I mean, the trouble with any cloud provider is that once you get over a few hundred gigs of storage, the cost becomes prohibitive pretty fast. So there's a service called, and I'm not affiliated with them in any way, but there's a service called ZFS.rent. Go take a look. What you do is you send them a hard drive and they put it into a slot in one of their servers, I think in California, but don't quote me on that. And you pay them 10 bucks a month per slot that you occupy. So if you want 20 terabytes of offsite backup, you send them a 20 terabyte hard drive. It can be direct shipped from Amazon or Best Buy or wherever. Wow. They'll rack the drive for you. And then they'll give you an IPv4 address and a VM. And then you use that as your ZFS replication target offsite for $10 a month, job done. ([Time 0:33:35](https://share.snipd.com/snip/dc4e5771-8508-4a88-b977-103d499ac3e2)) - **Weekend Learning Yields Real Infrastructure** - Learning new infrastructure quickly can be productive and fun when you commit focused time. - Alex built a three-node Kubernetes cluster over a weekend and turned it into content for work. Transcript: Alex Kretzschmar How'd that go? And I spent the whole weekend glued to my laptop and I had a great time. And I've now got a three node Kubernetes cluster in a closet and it works great. And then I was able to turn that into a piece of content for work. So, you know, for me, I'm just very privileged. ([Time 0:41:01](https://share.snipd.com/snip/c342ef58-8a47-4942-adf1-adc712d2b461)) - **Simple, Practical Home Lab Design** - Keep home-lab infrastructure simple and separate production-like services from lab experiments. - Use modest mini PCs, 10Gb trunks, and ZFS for reliable media and storage without overcomplicating your rack. Transcript: Alex Kretzschmar Yeah. Unify fanboy over here. You are okay yeah yeah it's hard not to like them right it really is they're apart from the price there there's definitely a unified tax definitely does get you yeah yeah i've got the ucg Fiber their new fiber gateway that's really nice and i replaced a i think i had a dream machine pro or something like when we're getting ready to sell the house i i just downsized everything From sort of rack gear as much as i could to the sort of small utility switches yeah nice okay so i've got like a ton of those lenovo mini pcs the kubernetes cluster i mentioned a few minutes Ago i've got a couple of MS01 Minis Forum devices doing like all the heavy lifting. I've got 100 terabytes of spinning rust in the basement, give or take, running ZFS that I talked about. That's in a 19-inch Sligo case. But yes, it's pretty minimal, dare I say, for what I do. Like I need a set of infrastructure to run tailscale stuff on for the videos. And then I need a separate set of stuff to run my home sort of pseudo prod. And yeah, I try and keep it as simple as I can, like 10 gig for all the trunks, the unified aggregation switches for like 260 bucks are great. ([Time 0:51:45](https://share.snipd.com/snip/879dc46e-259a-4f35-9ba2-05ebcd74c53b)) - **Streaming Consolidation Spurs Piracy** - Media industry consolidation and rising costs will likely push some consumers toward piracy. - Fragmentation and higher subscription prices reduce consumer choice and increase motivation to pirate. Transcript: Alex Kretzschmar I do think piracy is going to have a moment as inflation continues to bite and people just can't afford. As the media space continues to homogenize, you know, you saw the Warner Brothers acquisition by Netflix recently. Those massive corporate mergers are never good for consumers. They're just, it's always going to result in long-term higher prices, less choice, less content. More gates. Yeah. More tall booths. I do think piracy will have its moment over the next few years as a result of all that homogenization. We kind of had a nice honeymoon period for the last decade, led by Spotify, I think, where that was all you can eat. But you're starting to see famous artists now pulling out of Spotify because they don't pay their artists enough. And people like Tidal are coming along and Cobuzz, and you end up with this fragmented ecosystem. And the same thing's happening with video. Arguably, it's already happened with video because of just how lucrative a streaming service can be. ([Time 0:57:18](https://share.snipd.com/snip/ee440bb2-3d91-434c-a61e-d5514a38d82f)) - **AI Replaces Quick Research** - AI replaces many lookup tasks — people now reach for chat models instead of Google for explanations and drafting. - Alex uses AI to polish scripts and code, increasing personal productivity and output speed. Transcript: Alex Kretzschmar We have to talk about AI now. We haven't talked about AI yet. You said that. You look at the utility that people get from these things. Certainly like when I'm writing a video script, I'll run a bit that I'm stuck on through it and say, what do you understand as to what I'm saying here? Does this make sense? Like I'll do it into a voice memo and dictate it and then give it to Claude and say, polish this explanation about something or tweak this bit of code and make this do this instead of that Or whatever. Historically, if you look at industrial revolution, like we're going back a couple of hundred years here, there was a huge amount of speculation in the industrial revolution era. A lot of people made a lot of money and that's happening now in the AI space. And not every, not every idea that should have been funded, got funded. If I mean the other way around, there were a bunch of ideas that shouldn't have been funded that did get funded simply because people were hoping, speculating, right? We've seen this many times over the years, but these industrial revolutions and the ai revolution is one of those because it's a tooling thing as opposed to say 2008 which was a made-up Financial banking bubble whatever um there was nothing material you know moving humanity forward there uh so say what you like about ai but it's here to stay, I think. And the big, the big players like OpenAI, Anthropic, Google, of course, they'll be here to stay. Because every other startup is building their tooling on their model, on those big guys models. They're not, they're not doing the R&D spend to train their own models. They're just using Anthropic models for an API key behind the scenes. So I think the big guys will be okay. But it's, yeah, it's the circle of money that has me concerned, like OpenAI giving NVIDIA money, NVIDIA investing back in OpenAI and so on. Right. That's worrying. Adam Stacoviak Yeah. What is your relationship with AI? Are you a daily active user? Are you, what's your stance so far? Yeah, I'll reach for it quite often now. Alex Kretzschmar Yeah. Adam Stacoviak Okay. Do you have the itch right now in the pod to be like, can I just clod something real quick just because you have to? ([Time 1:11:01](https://share.snipd.com/snip/0d2fe577-3127-4d04-9025-c566b997c60b)) - **Plan First, Vibe-Code Later** - Use AI to bootstrap projects: plan thoroughly, then let models generate the implementation quickly. - Spend hours designing the spec, then let AI produce working code to iterate from. Transcript: Alex Kretzschmar Lab? Well, take it, take a look at this website, quicksync.ktz 100% VibeCode slop. Oh, gosh. What is this? This is a benchmarking website that I put together over the weekend that allows people to benchmark the transcoding performance of their Intel CPUs. So trying to answer the question of which CPU do I buy for my media server? Oh, my gosh this is spectacular. And this was me and Claude for about two or three evenings this weekend. So my positives around AI are that I know just enough to be dangerous as someone that's been adjacent to developer technologies for the last 15 years. I'm not very good at actually writing the code. It takes me way too long, and I get bored before I get finished. This allowed me to get something finished and out the door way faster than I ever could have done by myself. Okay, I didn't learn the intricacies of how the API calls work and what have you under the hood, but honestly, I don't need to. It's just a flat JSON file that gets stored in Cloudflare R2. This page loads that JSON file client side and then renders the webpage locally. And that's all I need to know as kind of like a normal developer adjacent person. ([Time 1:18:40](https://share.snipd.com/snip/a39e3703-7c28-4358-9585-40c2b31e8fa5)) - **Crowdsourced Transcode Benchmarks** - Crowdsource real-world hardware benchmarks and share results as JSON for transparency. - Alex's quicksync.ktz collects transcode runs to compare CPU hardware and energy tradeoffs for media servers. Transcript: Alex Kretzschmar Oh, gosh. What is this? This is a benchmarking website that I put together over the weekend that allows people to benchmark the transcoding performance of their Intel CPUs. So trying to answer the question of which CPU do I buy for my media server? Oh, my gosh this is spectacular. And this was me and Claude for about two or three evenings this weekend. So my positives around AI are that I know just enough to be dangerous as someone that's been adjacent to developer technologies for the last 15 years. I'm not very good at actually writing the code. It takes me way too long, and I get bored before I get finished. This allowed me to get something finished and out the door way faster than I ever could have done by myself. Okay, I didn't learn the intricacies of how the API calls work and what have you under the hood, but honestly, I don't need to. It's just a flat JSON file that gets stored in Cloudflare R2. This page loads that JSON file client side and then renders the webpage locally. And that's all I need to know as kind of like a normal developer adjacent person. I don't need to know every intricacy. So for me, it's like this has allowed something to exist in the world that I would have really liked to have seen 10 years ago when building my first server. And, you know, it's taken the information from being a flat text file to being something interactive that people can kind of, you know, enjoy, hopefully. Adam Stacoviak So, you know. Help me understand what I'm actually seeing here. So how are you getting all these benchmarks? It's crowdsourced. Alex Kretzschmar So you see up in the top right-hand corner, you can clone the Git repo to your server. Right. And then you could just run the benchmark script, and then it will submit the – it will pull down a jellyfin container because it just uses ffmpeg it pulls down some drone footage that I edited together a couple of years ago into like a test file and then it will just encode those files in different codecs so it will do a cpu software encode to give you an idea of the raw Horsepower of your cpu it will then jump into the quicksync engine which is the hardware transcoding thing built into your CPUs. So it does 1080p. It does 4k 8 bit HEVC and then 4k 10 bit HEVC hardware encodes. And so you can kind of see a graph. You can kind of see on the graph, the performance of the newer chips is a lot, a lot better as you would expect, but the energy usage also goes up. So it's like trying to find that sweet spot between energy usage, performance, price, et cetera. Right. So yeah, crowdsourced information. ([Time 1:18:48](https://share.snipd.com/snip/ca89db77-5182-4037-9e87-5178144e37b8))