Show HN: I built a DNS resolver from scratch in Rust – no DNS libraries
github.comI built a DNS resolver that lets me use https://frontend.numa instead of localhost:5173
— auto-generated TLS certs, WebSocket passthrough, path routing. No mkcert, no nginx, no /etc/hosts. Since I needed it to be my primary DNS, I also added: recursive resolution from root nameservers, DNSSEC chain-of-trust validation, ad blocking (385K+ domains), and LAN service discovery. I wrote about the DNSSEC implementation here: https://numa.rs/blog/posts/dnssec-from-scratch.html
It's now my daily system DNS.
Single binary (~8MB), macOS/Linux/Windows. `sudo numa install` Very interesting project! I have a couple of questions. With all the default blocked domains loaded, what is the average memory usage? Currently, I am using Pi-hole on a low memory single board computer. Is it possible to use this instead of Pi-hole? If so, I’d like to use it for all of my devices." With 390K blocked domains: ~31MB total process footprint.
Breakdown:
- Blocklist: 23.4MB (390K domains)
- Cache: 3.8MB (4.4K entries)
- Query log, SRTT, runtime: ~4MB It binds to 0.0.0.0:53 by default, so just point your devices' DNS to the board's IP Romanian project. Instant upvote. Great work Thanks! If you hit any issues during setup, feel free to open an issue — happy to help debug. The dashboard at localhost:5380 shows what's happening in real time. Why are you replying to your own coment? I think it's a bot? There's an identical version of this comment in another reply, except it cuts off half way through a sentence. I hit reply on the wrong post and you can't delete comments or at least I don't see how it can be done Above the comments I've written on HN I see: 5 minutes ago | parent | next | edit | delete That only lasts for a few minutes until it’s locked and you can no longer delete after that. It lasts 2 full hours, at least for edit. Delete stops working when someone replies afaik. because I clicked reply on the wrong one and you can't delete it... Of course I can’t prove it, but i am guessing some kind of “AI” is doing that. Humans rarely use emdashes. [flagged] Please don't break the site guidelines, regardless of how wrong someone is or you feel they are. You're right about em dashes of course (https://news.ycombinator.com/item?id=47154752) but being right on a point does not make it ok to attack another user or violate the rules of the site. [flagged] Please don't respond to a bad comment by breaking the site guidelines yourself. That only makes things worse. [flagged] It's neither here nor there but can I ask about the name? I only ask because when I see "numa" in relation to computing I immediately think "Non-Uniform Memory Access". Very cool project by the way. I wonder how this would run on an OpenWRT device. I see in your install.sh that you support Linux and Darwin/MacOS, do you think there would be any major hurdles in supporting FreeBSD? also in romanian nume = name(dns) and I also get the easter egg of that well known Romanian song numa numa :) https://www.youtube.com/watch?v=YnopHCL1Jk8 On OpenWRT — it's musl-based Linux so the binary should run the arm one would need a crosscompile
Free BSD can be done (pr's welcome?) When you've capitalized the N in "Numa", it's a lot more obvious to my brain you don't mean the acronym. But this is nitpicky bikeshedding and maybe I'm weird :) On the web site, it's named after the second King of Rome I have a couple of projects that once a month need to run a few million dns lookups as quickly as possible. I'm tempted to try this just to see how it performs and if it breaks. let me know if you do it! The interface looks vibecoded. I have no problem with people vibecoding things. In fact, I have zero frontend skills, so I rely on AI to be able to make easy-to-use interfaces. However, I feel like this should be clearly and prominently displayed in the project page. Furthermore it is a little off-putting to see a vibecoded UI because I have very little confidence that the rest of the backend code is not vibecoded. I know I am possibly being unfair, but this is how it looks to me. If the developer tells me they didn't use AI at all, I would believe it. It definitely is and you can see it in the git commits. The DNS wire protocol parser was the original learning project I wrote to understand the spec. Later features (recursive resolver, DNSSEC validation, the dashboard) were built with the help of AI That's fair, thanks for letting me know! I dont get this criticism at all, would you prefer someone write a shittier UI? And since when were people writing amazing bug free software before hand where not being vibe coded meant you could trust its good software? I guess to be fair, beforehand no body would be attempting this kind of thing and releasing it unless they knew what they were doing I literally said I'm fine with using LLMs for the frontend, but I think this should be disclosed clearly. I don't think having conditions to certain things qualify as "I'm fine with it" "I'm fine with people eating meat, as long as they declare so when we go out" like why? Why does it matter? Both GP's and your example in effect mean "I'm fine with other people doing this, but I don't want to have anything to do with it, or at least be able to decide case-by-case." Which is a valid stance IMO. In the OP, a vibecoded UI when the whole project emphasizes "I did this myself, from scratch" is a bit awkward. Does "I did this myself" mean they read all the relevant specs and then wrote the code - or did they just write the prompts themselves? Edit: OP already answered and confirmed that they in fact did write the code themselves. Given the state of webdev it is not a surprise. LLMs are my rubber gloves when working with web technologies. Nice idea. To test I ran a simple nextjs on port 3000. Added the service via the dashboard.
However, when I visit the url, (using chrome latest version), https://{mygivenname}.numa/ I hit a DNS resolution fail error.
If I do not use a trailing '/' then it is going to google search for {mygivenname}.numa and shows me some search results. Should I open an issue? Is it possible you didn't start it as root ( sudo numa install)?
Does dig {mygivenname}.numa @127.0.0.1 return 127.0.0.1 ?
What OS are you on?
Maybe you report it as an issue? Thanks for quick response. It started to work. I think it must be some caching issue. But it needs a trailing '/' . Maybe will raise the issue for this. Cool. I believe that is actually browser specific behavior. I sometimes use a fake TLD for stuff hosted at home, and both chrome and firefox resort to search if I don't include a trailing '/'. My assumption is the browser does a quick match against known TLDs and if it doesn't match then it resorts to search. exactly, I'll add a pr soon that tells the os (and browsers) that is'a a valid domain What's the reason you're not using hickory? Or was that the LLMs choice? Genuinely curious This was started as a learning project, went from the start to the lowest level then I've just added features I wanted one by one, it just made the most sense Same hack here ; I have no DSN running by default - much more handy than having to set up nginx as it has no opinion on the targeted infrastructure. And the bonus point is that you can see every sneaky request that happens when you browse ; so another side-project connected to this is to make an inventory and policy filter Yes sir!
The query log is at GET /querylog (or on the dashboard) shows every request with domain, type, path (forwarded/recursive/cached/blocked) and latency feature request: libnuma so i can use it programmatically with configuration. also, multiple user defined blocklists. Multiple blocklists already work -https://github.com/razvandimescu/numa/blob/main/numa.toml#L4...
The pieces are already there for libnuma, it could be done, would you share what use case you have in mind? The first thing I look at in new DNS code is whether it’s vulnerable to DNS name compression loops. This code passes the test! However it’s vulnerable to dots embedded in labels: it doesn’t escape bytes properly when converting from wire format to text. Thanks for pointing this out! I’ve created https://github.com/razvandimescu/numa/issues/36 How does one handle dots embedded in a label ? Isn’t that not valid? should be fixed by #54 in 0.10.3 thanks again! I have a project that requires DNS lookups and block ads. I am going to try this for it. let me know how it goes How is to compare to AdGuard? If it gets those features I would be switching over. Numa can do recursive resolution from root nameservers + DNSSEC, .numa local domains with auto HTTPS for dev, and LAN service discovery.
What features would you be interested in? What about split horizon dns so I can locally resolve home servers instead of going to tailscale Split DNS already works — Numa auto-detects Tailscale forwarding rules from the system config. Queries matching .<ts.net> go to Tailscale’s DNS, everything else goes through Numa If you want to skip Tailscale entirely for home servers, Numa’s LAN discovery auto-finds machines running Numa on the same network. Or add static records in numa.toml for machines that don’t run it. Just normal ad filtering. Cool idea, every developer running apps in dev on their machine knows this pain for sure. I'll give it a spin and let you know how it goes! Thanks! If you hit any issues during setup, feel free to open an issue — happy to help debug. The dashboard at localhost:5380 (or at https://numa.numa) I think I need to give this a go. Cool project. Thanks! Let me know how it goes. Nice work. What made you choose this license? very interesting. how does the blocklist work? can one manage the lists? like StevenBlack or others. Yes, it is configurable as a list
https://github.com/razvandimescu/numa/blob/main/numa.toml#L4... There's also a per-domain allowlist and you can pause/unpause blocking from the dashboard or API. Here's how the resolution pipeline looks like: https://numa.rs/blog/posts/dns-from-scratch.html#the-resolut... nice Great idea, pity about the slop. we need a [slop] flag in the headlines