HNHacker News
TopNewBestAskShowJobs

kjuulh

512 karma · joined March 22, 2021

submissionscomments
kjuulh··on Zero-copy protobuf and ConnectRPC for Rust
On the other hands, having half the packages depend on packages such as serde, syn, procmacro2 might not be such a good idea. First of all it is annoying when creating new projects to have to move over table stakes. Second, it is a security nightmare. most of rust could be vulnerable if dtolnay decided to go rogue.

It is not that everything should go into the stdlib, but having syn, procmacro and serde would be a good start imo. And like golang having a native http stack would be really awesome, every time you have to do any HTTP, you end up pulling in some c-based crypto lib, which can really mess up your day when you want to cross-compile. With golang it mostly just works.

It isn't really in the flavor of rust to do, so I don't think it is going to happen, but it is nice when building services, that you can avoid most dependencies.

kjuulh··on Tell HN: Anthropic no longer allowing Claude Code subscriptions to use OpenClaw
Anthropic should calm down, I get that they're trying to either build a moat, or simply curb what is essentially subsidized tokens. It is technically true that when you've got a claude code subscription you pay for the product with its terms, and those terms doesn't include you grabbing the token and using it for another application. They're also trying to build a competitor to openclaw so it makes sense they're trying to crush it. But it feels like such a feeble moat, that it looks silly. Claude Code is nice, but it is not that nice.
kjuulh··on Moving from GitHub to Codeberg, for lazy people
For me it was specifically crawlers from the large companies, they we're at least announcing themselves as such. They did have different patterns, bytedance was relatively behaved, but some of the less known ones, did have weird patterns of looking at comparisons.

I do think they care about repos, and not just the code, but also how it evolves over time. I can see some use, if marginal in those traits. But if they really wanted that, I'd rather they clone my repos, I'd be totally fine with that. But i guess they'd have to deal with state, and they likely don't want to deal with that. Rather just increase my energy bill ;)

kjuulh··on We rewrote JSONata with AI in a day, saved $500k/year
Not saying it is a good thing, but an organization, especially if there has been a lot of turnover, can enter a state of status quo.

> it must have that architecture for a reason, we don't enough knowledge about it to touch it, etc.

That or they simply haven't had the time, cost can creep up over time. 300k is a lot though. Especially for just 200 replicas.

Seems wildly in-efficient. I also don't understand why you wouldn't just bundle these with the application in question. Have the go service and nodejs service in the same pod / container. It can even use sockets, it should be pretty much instant (sub ms) for rpc between them.

kjuulh··on Moving from GitHub to Codeberg, for lazy people
My own git server has been hit severely by scrapers. They're scraping everything. Commits, comparisons between commits, api calls for files, everything.

And pretty much all of them, ByteDance, OpenAI, AWS, Claude, various I couldn't recognize. I basically just had to block all of them to get reasonable performance for a server running on a mini-pc.

I was going to move to codeberg at some point, but they had downtime when I was considering it, I'd rather deal with that myself then.

kjuulh··on Why I'm Not Worried About Running Out of Work in the Age of AI
Will swe's be squeezed; yes. But I don't think everything will just be magically done by these models. Right now the wheels are completely off the wagon as we see more and more vibe coded apps going live with fatal security vulnerabilities, privacy issues. The act of putting pen to paper, will change, the positions will change, but I don't think these models are a silver bullet.

Nothing has been so simple until now, and it seems strange that we just get to a certain point and then all of our problems are now just solved, completely. From my experience until now, at my current start-up, it has reduced our need to hire a tad, but not too much. However, I've also seen early stage start ups needing to hire because they started out building a product, and it became too much to handle, it is anectodal and current, I'd just find it strange that we just end up automating ourselves away, my own role has sort of turned into an AI enablement for the rest of start up, mostly C-level, business, pretty much everyone else than swe's. There is potential but mixed success for now. Agent's a good enough to build something that works, but not good enough to build the right solution.

I had a guy that ended up building a local dashboard in perl (the only thing claude could find on his mac) and wanted to distribute it to his colleagues. Engineers sometimes forget that normal people don't usually work in the unknown, they will solve problems in any way they known, in this case a airdropped folder of perl code sent to each other.

kjuulh··on Our commitment to Windows quality
What is apparent from windows in my opinion is a lack of direction. Ask for feedback when something falls outside of common sense, you ask for feedback when you need another point of view. But windows is currently failing at the most essentials, these should be apparent inside Microsoft as well, and should've been for the last few years.

It is not that everything should stay the same, that is one choice, but there needs to be a steward that says, hey our right click menu on the desktop has an SLA of 100ms to open, it doesn't matter which features you put in there, if something causes it to be slow, kill it.

Can I access basic apps that are table stakes for an OS, an editor, screenshots etc without popups for unrelated nonsense. If you fail at that, then as a user I get confused. I am used to just being able to note down some text, why am I asked to transcribe with Copilot or login to microsoft.

It is clear that the adoption of Copilot was measured in activations, and as such was pushed in as many places as possible, simply because they needed all that exposure to meet their targets. Windows was not just a product but a funnel to other offerings and that cannibalized windows even more than it was previously.

I've got a slight bias, as I haven't had windows installed in about 10 years, but when I've helped my family with their issues, it is clear how much of a shitshow it actually is.

kjuulh··on Don't post generated/AI-edited comments. HN is for conversation between humans
Just had a look, it is pretty interesting, just from the few times I've checked the frontpage there was some interesting articles to me. with a variety of topic. Great suggestion!
kjuulh··on Don't post generated/AI-edited comments. HN is for conversation between humans
I am 100% behind this. I've been browsing hackernews since I started in tech, it is the only forum i regularly browse, and partake in. Simply because the quality of submissions and conversations are so high. There has been more AI related articles this part year, and it only seems ramping. I personally haven't found the AI part of the comments as big of a deal but dang and tom might be doing more than I realize on that front.

Though I do wish we'd see less AI related posts on the front page, they simply aren't sparking curiosity, it is the same wrapped in a different format, a different person commenting on our struggles and wins with AI, the 10th software "rewritten" by an AI.

At this point there nearly should be a "tax" on category, as of this moment I count 8-10 related posts on the front page related to AI / LLMs. It is a hot field, but I come to hackernews, to partake in discussions about things that are interesting, and many of those just doesn't cut it, in my opinion.

kjuulh··on MinIO Is Dead, Long Live MinIO
I considered it a while ago, but I wasn't totally clear on Read-After-Write. Which was the primary reason why I choose to just implement my own for testing.

I'll probably give GarageHQ a more serious look again.

kjuulh··on MinIO Is Dead, Long Live MinIO
It is an interesting time we're in right now where buying physical hardware and support is cheaper than a license.

Same goes for AWS markup on rented hardware. ;)

Man I sometimes miss having physical servers.

kjuulh··on Block the “Upgrade to Tahoe” alerts
Upgrading to Sequoia was a mistake, and so was upgrading to Tahoe.

I like new and shiny software, but these two releases aren't great. Outside of a good amount of bugs. It is wild to me that Apple can't even get their own UI consistent.

Apples own apps are pretty much the only things you can't close. Finder: can't quit. System settings, somehow doesn't expand horizontally (are we still in the 2000s apple?) I haven't felt the liquid glass or whatever too much on the laptop, but I just used one of my family members Iphone today, and man it was distracting, it seems crazy that contrast has gone out the window.

But especially the bugs. Apple should really take a release that is just bug fixing. I had to switch out Spotlight because it kept trying to want to index my entire system, which is hard when you work in both Rust and typescript projects (lots of small files).

kjuulh··on MinIO Is Dead, Long Live MinIO
Seems like a very balanced take on forking Minio. I don't have high hopes for the future Minio, but as mentioned it is more or less feature complete, good enough for most use-cases.

I was searching for a fairly simple replacement for s3 for testing. I'd been using Minio for a while now, and simply ended up implementing my own on top of Postgres. Fun intersection given the post. (Note, I know it isn't optimal, but as I always have Postgres available it fits well, and I don't have high storage needs, just the api compatibility)

kjuulh··on gRPC: From service definition to wire format
I think connect-rpc[0] strikes a good balance between normal http apis, and gRPC. It allows protobuf as json. So you could think of it as an opinionated http api spec. A health check would be just a call to an url /something.v1.MyService/MyMethod -d { "input": "something }.

it works really well, and the tooling is pretty good, though it isn't that widely supported yet. Rust for one doesn't have an implementation. But I've been using it at work, and we basically haven't had any issues with it (go and typescript).

But the good thing is that it can interoperate with normal grpc servers, etc. But that of course locks it into using the protobuf wireformat, which is part of the trouble ;)

0: https://connectrpc.com/

kjuulh··on Oxide raises $200M Series C
Looking forward to the podcast! Congrats. Had to do a double take, wasn't it around summer last year they closed 100 mil. Wild
kjuulh··on Termux
Use it all the time.

I've got a wireguard setup from phone or tablet to my workstation. Using mosh with zellij and I can do all the development I want. Whether it is restarting a machine, or actually writing code, using claude code etc. It works really well

kjuulh··on Rust’s Standard Library on the GPU
I think it fits quite well. Kind of like the rust standard lib runs on the cpu this does partially run on the gpu. The post does say they fall back on syscalls but for others there a native calls on the gpu itself such as Instant. The same way the standard lib uses syscalls on the cou instead of doing everything in process
kjuulh··on Statement by Denmark, Finland, France, Germany, the Netherlands,Norway,Sweden,UK
That is my gut feeling as well. I wonder if we're gonna see more expansion in the south americas as well in the coming year..
kjuulh··on Ask HN: Share your personal website
https://rawpotion.io

Blog and project site

kjuulh··on Ask HN: Vxlan over WireGuard or WireGuard over Vxlan?
You can see my reply below: https://news.ycombinator.com/item?id=46609044 I believe our setups are pretty equivalent.

I achieve load balancing by running native wireguard on a vps at hetzner, I've got a native wireguard mesh, I believe Talos can do the same, where the peers are manually set up, or via. tailscale etc. I then tell k3s that it should use the wireguard interface for vxlan, and boom my kubernetes mesh is now connected.

flannel-iface: "wg0" # Talos might have something similar.

I do use some node-labels and affinities to make sure the right pods end up in the right spot. For example the metallb annoucer always has to come from the hetzner node. As mentioned in my reply below, it takes about 20ms roundtrip back to my homelab, so my sites can take a bit of time to load, but it works pretty well otherwise, sort of similar to how cloudflare tunnels would work, except not as polished.

My setup is here if it is of help

https://git.kjuulh.io/kjuulh/clank-homelab-flux/src/branch/m...

kjuulh··on Ask HN: Vxlan over WireGuard or WireGuard over Vxlan?
I use vxlan on top of wireguard in my hobby set up. Probably wouldn't recommend it for an actual production use-case. But that is more or less because of how my homelab is setup (Hetzner -> Home about 20ms latency roundtrip).

I considered dropping my root wireguard and setting up just vxlan and flannel, but as I need NAT hole punching I kind of need the wireguard root so that is why i ended up with it.

Going Wireguard inside the vxlan (flannel) in my case, would likely be overkill, unless I wanted my traffic between nodes between regions to be separated from other peers on the network, not sure where that would be useful. It is an easy way of blocking out a peer however, but that could just as well be solved on the "root" wireguard node.

There might be some MTU things that would be messed up going nested wireguard networks.

kjuulh··on The World Happiness Report is beset with methodological problems
From the provided question by WHR I can definitely see how Scandinavian countries rank so high. Being Danish myself my answer would immediately go into long term thinking and whether I would have a better life elsewhere and to me the answer is a clear no. Not financially, socially or politically. So yes, Denmark scores really high, but is it really measuring happiness. I dont know. That said I dont think measuring how often we laugh as a better metric amongst other thing, I can be perfectly "happy" without being outwardly joyous, maybe contentment is a better word. Or well being. But it isn't as catchy i guess
kjuulh··on Pricing Changes for GitHub Actions
After self hosting our builds ended up so fast, that we were actually waiting for was GitHub scheduling our agents, rather than it being the job running. It sucked a bit, because we'd optimized it so much, but we on 90th percentile saw that it took 20-30 seconds for github to schedule the jobs as they should. Measured from when the commit hit the branch, to the webhook begin sent.
kjuulh··on Pricing Changes for GitHub Actions
We choose github actions because it was tied directly to github providing the best pull-request experience etc. We actually didn't really use github actions templating as we'd got our own stuff for that, so the only thing github actions actually had to do was start, run a few light jobs as the CI was technically run elsewhere and then report the final status.

When you've got many 100s of services managing these in actions yaml itself is no bueno. As you mentioned having the option to actually be able to run the CI/CD yourself is a must. Having to wait 5 minutes plus many commits just to test an action drains you very fast.

Granted we did end up making the CI so fast (~ 1 minute with dependency cache, ~4 minutes without), that we saw devs running their setup less and less on their personal workstations for development. Except when github actions went down... ;) We used Jenkins self-hosted before and it was far more stable, but a pain to maintain and understand.

kjuulh··on Pricing Changes for GitHub Actions
It is xD On the outside it feels like a product held together with duct tape, wood glue and prayers.
kjuulh··on Pricing Changes for GitHub Actions
We've self-hosted github actions in the past, and self-hosting it doesn't help all that much with the fragile part. For github it is just as much triggering the actions as it is running them. ;) I hope the product gets some investment, because it has been unstable for such a long time, that on the inside it must be just the usual right now. GitHub has by far the worst uptime of any SaaS tools we use at the moment, and it isn't even close.

> Actions is down again, call Brent so he can fix it again...

kjuulh··on Dagger: Define software delivery workflows and dev environments
Sounds good Solomon I look forward to seeing how it goes in about a year when I am going to tackle our CI again ;)

Best of luck and thx for taking my harsh feedback in strides!

kjuulh··on Dagger: Define software delivery workflows and dev environments
As someone that has used Dagger a lot (a previous daggernaut / ambassador dropped off after LLMs was announced, and was changing jobs at the time. implemented it at a previous company across 95% of services, built the rust sdk) the approach was and is amazing for building complex build chains.

It serves a place where a dockerfile is not enough, and CI workflows are too difficult to debug or reason about.

I do have some current problems with it though:

1. I don't care at all about the LLM agent workflows, I get that it is possible, but the same people that chose dagger for what it was, is not the same audience that runs agents like that. I can't choose dagger currently, because I don't know if they align with my interests as an engineer solving a specific problems for where I work (delivering software, not running agents).

2. I advocated for modules before it was a thing, but I never implemented it. It is too much magic, I want to write code, not a DSL that looks like code, dagger is already special in that regard, to modules takes it a step too far. You can't find the code in their docs anymore, but dagger can be written with just a .go, .py or .rs file. Simply take in dagger as a dependency and build your workflow.

3. Too complex to operate, dagger doesn't have runners currently, and it is difficult to run a setup in production for CI yourself, without running it in the actions themselves, which can be disastrous for build times, as dagger often leads you into using quite a few images, so having a cache is a must.

Dagger needs to choose and execute; not having runners, even when we we're willing to throw money at them was a mistake IMO. Love the tool, the team, the vision but it is too distracted, magical and impatient to pick up at the moment.

kjuulh··on Cloudflare outage should not have happened
It did happen, and cloudflare should learn from it, but not just the technical reasons.

Instead of focusing on the technical reasons why, they should answer how such a change bubbled out to cause such a massive impact instead.

Why: Proxy fails requests

Why: Handlers crashed because of OOM

Why: Clickhouse returns too much data

Why: A change was introduced causing double the amount of data

Why: A central change was rolled out immediately to all cluster (single point of failure)

Why: There are exemptions or standard operating procedure (gate) for releasing changes to the hot path for cloudflares network infra.

While the Clickhouse change is important, I personally think it is crucial that Cloudflare tackles the processes, and possibly gates / controls rollout for hot path system, no matter what kind of change they are when they're at that scale it should be possible. But that is probably enough co-driving. It to me seems like a process issue more than a technical one.

kjuulh··on Cloudflare Global Network experiencing issues
It depends on the infrastructure you're running on. There was a post yesterday going fairly into depth how you do such calculations https://authress.io/knowledge-base/articles/2025/11/01/how-w...

You probably cannot achieve this with a single node, so you'll at least need to replicate it a few times to combat the normal 2-3 9s you get from a single node. But then you've got load balancers and dns, which can also serve as single point of failure, as seen with cloudflare.

Depending on the database type and choice, it varies. If you've got a single node of postgres, you can likely never achieve more than 2-3 9s (aws guarantees 3 9s for a multi-az RDS). But if you do multi-master cockroach etc, you can maybe achieve 5 9s just on the database layer, or using spanner. But you'll basically need to have 5 9s which means quite a bit of redundancy in all the layers going to and from your app and data. The database and DNS being the most difficult.

Reliable DNS provider with 5 9s of uptime guarantees -> multi-master load balancer each with 3 9s, -> each load balancer serving 3 or more apps each with 3 9s of availability, going to a database(s) with 5 9s.

This page from google shows their uptime guarantees for big tables, 3 9s for a single region with a cluster. 4 9s for multi cluster and 5 9s for multi region

https://docs.cloud.google.com/architecture/infra-reliability...

In general it doesn't matter really what you're running, it is all about redundancy. Whether that is instances, cloud vendor, region, zone etc.

← PreviousPage 2 of 6Next →