Building a Debugger is part of the Humble Books Bundle for the next couple weeks!
www.humblebundle.com/books/linux-...
Posts by Tim Misiak
Yeah, if I was writing helm by hand I would agree. Claude is great at writing them or converting from compose, so it hasn't felt like a chore for me yet.
I'm sure I'll find pain points later that I will grumble about, but so far it's hard to imagine a reason I would switch back.
Third, k3s setup is easy. I avoided kuberenetes for a long time for my own server because I thought it would be a pain to set up. Using k3s makes it easy. The only trouble I had at all was figuring out how to get it working with IPv6-only setups.
Secondly, the way traefik works for controlling ingress with annotations on containers is really convenient. It puts the configuration in the right place... Close to the app, not the proxy. Not specific to k3s, but it's the default setup and it makes sense
For one, I actually really like managing the infra and containers with terraform. It would be annoying to write everything by hand... But that's what claude is for. Using tf with k3s is very nice, especially compared to what it would look like with docker
This might be a hot take but I think anyone running a homelab should take the time to learn enough kuberenetes to run their stuff inside k3s. It has a slightly steeper learning curve compared to just using docker compose, but you can get a lot out of it.
My cell connection is 63x faster than this hotel wifi.
Oh that's great. I'll definitely give this a try if I come back to give oracle another chance. Their free tier instance sounds pretty nice if I can get it to work.
In contrast, I spent several hours trying to set up an ARM64 linux VM on Oracle cloud and just came away confused and sad.
I set up a minimal ARM64 linux VM with IPv6-only networking on Azure (with a cloudflare tunnel for SSH access) and it took a FRACTION of the time compared to setting up on AWS. I haven't touched Azure at all since working at Microsoft and I'm honestly pretty impressed.
Anyone ride the 2 line today? Couldn't make a trip today but looking forward to seeing how the ride is sometime soon.
I love your optimism, thinking we will have a 2030
Any pain points that would have existed before are now fairly trivial to address with some LLM coding, because the sorts of things that frameworks solve are EXTREMELY well understood and are the sort of thing that Claude is going to get right on its first try.
For most small to medium applications, I think you get a better inner loop, better performance, and easier to read code by simply using modern vanilla JS with Web Components. And the only thing I would change for larger apps is switching to typescript (maybe).
Is there really any point to JS frameworks anymore? I've spent some time experimenting with frameworks by using claude to sketch out similar ideas with different frameworks, and they pretty much universally make things worse compared to Web Components.
Yeah I really hated that my user profile was usually "timmi" unless I jumped through hoops
Oh it’s called shell scripting
Screenshot of text from claude.ai: Docker Compose integration: There's no native Nix support in compose, but the cleanest approach short of a script is using compose's build with a custom dockerfile that just calls out... actually that defeats the purpose.
Started using Claude exclusively instead of ChatGPT (for obvious reasons), and laughed out loud when I saw claude say "oh wait no my bad" in the middle of an explanation.
Anyone who made fun of the windows file copy progress dialog owes someone an apology. My rsync has been sitting at 95% for 12 hours and is still making solid progress copying to my new ZFS pool.
Had a "tiramisu danish" this morning in London and it was honestly one of the best pastries I've had.
If you still have a Ring camera, you may want to think about replacing it with something a little less Continental Panopticon.
www.404media.co/with-ring-am...
The funny thing is that these drives look externally identical to another set that i bought from a different seller on ebay that worked perfectly. And they kinda sorta work when you plug them in, except you can't actually touch any data on them.
Today I learned that not only could an enterprise drive be formatted with 520 byte sectors instead of 512 byte sectors (causing linux to shrug and give up) but also it could have special vendor firmware. It's been an ordeal, but it looks like openSeaChest and sg_format are going to save this drive.
This OpenClaw thing is amazing. It helped me corrupt my database in a fraction of the time it would've taken me to do it by hand.
4TB drives are such a steal compared to what a 12TB+ drive sells for right now, but it only makes sense if you have a server that can fit a ton of drives.
If you've never played with server class machines, you'd never know some of the crazy stuff they can do. I learned today (from some drives I got on ebay) that multipath SAS is a thing. One drive connected to a single backplane talking to two HBAs. Like a little network for hard drives. Crazy
Incidentally, I think reasonable people can still differ about whether this counts as thinking or understanding, but frankly either possibility is weird. Either a machine can think and reason really well, or you can somehow solve very hard open math problems without thinking/reasoning/understanding.
I feel like cdas went out of style but they are so good.
Did the same. Then ordered one of each from the dessert menu