pl4nty
pl4nty on Hacker News
yaml definitions compiled to sqlite sounds pretty similar to winget, which is scaling very well. vector search is a cool idea - how much storage does it use per package?
I had to scroll too far for someone to mention third-party CI. GitHub Actions free runners have always sucked, but the third-party runner ecosystem is really strong for those who can afford it. imo the APIs are far better than the rest of the product - I suspect enterprise customers are strong-arming GitHub to keep them reliable. and there's always third-party CI like tekton if Actions' yaml is too annoying
> windows store requires a MS account
they avoid mentioning it, but the Microsoft managed package format (MSIX) works just fine without the Microsoft Store. create an App Installer manifest, stick it on a website, and get semver-d differential updates across multiple architectures for free: https://learn.microsoft.com/en-us/windows/msix/app-installer...
msft have woefully underinvested in the ecosystem and docs though. I wish they'd fund me or others to contribute on the OSS side - electron could be far simpler and more secure with batteries-included MSIX
> Until now the only way to do that was on device
as usual, Apple's implementation is exceptional, but far from the first. see https://confidentialcomputing.io/ and its long history
thanks for this, can't believe I've never come across it. I've been building git scraping tools for years, but didn't know it was a popular pattern
this, because msft spent years and many $$$ to build an open-source ecosystem. apple hasn't done that yet, so I'm not sure why anyone would trust them
this is really helpful, thanks. how much are third-party models changing these workflows (LLMs etc)? would you still spend as much time on feature engineering and evaluation? I'm wondering whether any saved time would be refocused on hosting, especially optimizing GPU utilization
I host a usable web app like this, including a complete CLI via WebAssembly. it's surprisingly useful for teaching, kind of like CyberChef
https://openssl.tplant.com.au/
it's not as expensive as it looks, if you don't count time. CNC alu and the display would be the biggest costs. other carrier board projects use 3d printing over CNC, but the display/bandwidth is kinda what makes this novel so a lower res wouldn't make sense
really hoping Byran's excellent writeup helps encourage others. SoMs have lowered the barrier to entry and birthed a ton of SBC/carrier communities, but most of their tribal knowledge is buried in discord servers
thanks, really appreciate that your post was concise and included examples/followups. only needed 10 minutes to style my own rss feed
mainline LTS is good enough to boot and perform well, but it's still missing media and NPU features. as someone who briefly worked on mainlining, I feel the biggest problem is Turing's deceptive marketing - rk3588 and SBCs in general just aren't powerful enough for popular inference workloads like LLMs, let alone training. and ARM (rather than x86) is still very limiting for self-hosting apps
every trufflehog post I've seen on hn has been alarmist clickbait. could've been an opportunity to discuss security tradeoffs of `sub` vs `email` and how to handle `sub` changes, but nope their take is "sub doesn't fix the problem we found"
Microsoft Store and Xbox games/updates are distributed with a proprietary P2P protocol, which also includes ISP appliances. afaik it's the largest P2P network in the world.
https://learn.microsoft.com/en-us/windows/deployment/do/mcc-...Steam recently introduced LAN-based P2P to complement their significant appliance/CDN infrastructure, but idk if anyone has pulled it apart yet. and I don't think it does tunnelling like the msft network
its layout engine is the real magic for me, it allows complex diagrams that are just unreadable in graphviz or mermaid. a concise DSL and animation support are bonuses too
it's become much easier in the last few years. Talos + Flux + Renovate can be built in a day [0], and simplifies storage/backups/patching even for single-node clusters. There's also a great community, with services like kubesearch [1] providing templates for tons of apps
[0] https://github.com/onedr0p/cluster-template
[1] https://kubesearch.dev
and the niche customers that do need Proton's level of privacy are often more comfortable with a selfhosted product
it's a pain, you'll need to build the images on separate jobs then merge them into a multi-arch manifest. I moved to namespace.so for remote BuildKit builds a while ago and haven't looked back. depot.dev is also good, but pretty expensive
the docs might be misleading - New-MailboxImportRequest is "only supported on-prem", but it's used by the PST import UI and works just fine with EXO
> apt is unpinnable, as old versions get removed
not necessarily, eg snapshot.debian.org
> pinning on hashes is architecture dependent
can't you pin the multi-arch manifest instead?
I still like StableBuild for protection against package deletion, and mirroring non-pinnable deps
k8s is a lot easier for homelabs than it used to be, and imo it's quicker than nix for building a declarative homelab. templates like this one can deploy a cluster in a few hours:
https://github.com/onedr0p/cluster-templatehere's my home assistant deployment as a single file: https://github.com/pl4nty/homelab/blob/main/kubernetes/clust...
I deliberately nuked my onprem cluster a few weeks ago, and was fully restored within 2 hours (including host OS reinstalls). and most of that was waiting for backup restores over my slow internet connection. I think the break even for me was around 15-20 containers - managing backups, config, etc scales pretty well with k8s.
Nvidia's SODIMM compute module interface can prove this concept already. I have two 7W ARM Turing RK1s arriving soon, each with PCIe 3x4 at 4GB/s, and the Turing Pi 2 cluster board can fit four in an ITX form factor. I'm expecting over 3Gbps per watt at a total cost of 820USD
PCIe lanes are the bottleneck so far - even my $90 2TB SSDs are rated at 7GB/s on PCIe 4x4. So I don't think SBCs are the optimal solution yet. Looks like Ampere's Altra line can do PCIe 4x128 at 40W so a 1U blade with 100G networking could be interesting. I've seen lots of bugs and missing optimisations with ARM though, even in a homelab, so this kind of solution might not be ready for datacenters yet
they do offer an alternative, Azure Code Signing (ie cloud HSM). ImageMagick are looking at moving to it. it's free but someone difficult to access, due to the identity verification requirements
VxRail's the market leader? Last I touched it, it was a mountain of integration faults and vendors pointing fingers at each other. Maybe it's easier in larger or more traditional environments, but I'd be apprehensive of hyperconvered if VxRail's the best available
WSL2 doesn't really use VMs in the traditional sense. msft have invested heavily in lightweight virtualisation (eg better memory management) for certain security features, which allowed WSL to perform well enough without the maintenance/support overhead of WSL1
for one, it's bundled with consumer versions of Windows as winsqlite3.dll. not sure when this started though