drmorr

Every project has a Continuous Integration (CI) pipeline and every one of them complains its CI is too slow. [...] This can be the root cause of many problems, including lackluster productivity, low morale, high barrier of entry for newcomers, and overall suboptimal quality.

Here's another good post with a bunch of practical tips for speeding up your CI pipeline by @pgaultier.bsky.social

gaultier.github.io/blog/speed_u...

I have a friend who describes this process as "just show me the prompt, and if I need the lengthy verbocious text I can feed it into ChatGPT myself"

Also really enjoyed this post about LLM inflation by @ltratt.bsky.social

"It is very easy to use LLMs to turn short, simple content into something long and seemingly profound — and to use LLMs to turn long and seemingly profound content into something short and simple."

tratt.net/laurie/blog/...

The problem is that we need a way to _generate_ yaml programmatically, and helm is a templating language trying to be a programming language.

Just. Just stop. Use a real programming language to generate your configs.

Unlike most folks, I actually don't hate yaml, even though it's fun to dunk on. But I think it's approximately "fine" as a configuration language for computers to consume; yaml isn't the problem, and kyaml isn't fixing it.

I shouldn’t have to read installer code every day I don’t want helm charts to be my interface to off-the-shelf components on a daily basis. Kubernetes resources are simpler.

I Shouldn't Have to Read Installer Code Every Day -- "Many helm charts for off-the-shelf applications have hundreds of parameters. They are pushed in the direction of extreme flexibility by requests from their many users—the more users and use cases, the more parameters."

itnext.io/i-shouldnt-h...

Catching up on my blogroll: two different posts slamming Helm show up.

Helm Charts are Cloud Native Fast Food -- "Only 13,880 lines of yaml: If those were calories, it would be about 24 Big Macs."

kyle.cascade.family/posts/helm-c...

Using SimKube 1.0: Comparing Kubernetes Cluster Autoscaler and Karpenter This post is part 2 of a 3-part series on some "real" results that I've gotten from my Kubernetes simulator project called SimKube. If you recall from last week, I also announced that SimKube has matu...

The graphs at the end remind me a lot of the SimKube work I did last year, only at a much larger scale. I wonder if they'd be willing to release a trace of their work? I'd go bananas with SimKube on that.

blog.appliedcomputing.io/p/using-simk...

"However, we achieved consistently a high throughput of 500 pods/second even at the 100K node scale by carefully tailoring scheduler plugins based on the workload and optimizing node filtering/scoring parameters."

How does this work??? What did they do??? Again, frustratingly light on details.

I'm also interested in what they did to kube-scheduler. They say that the scheduler is a big source of latency because of its "one-pod-at-a-time" behaviour, and that this doesn't work for "ultra-scale" or big ML workloads.

"Offloading consensus to journal enabled us to freely scale etcd replicas without being bound by a quorum requirement and eliminated the need for peer-to-peer communication."

How on earth does that work? Would love to have more details here.

Aside from a bunch of "targeted" improvements, it seems like the biggest change is to move away from raft into some internal thing called "journal" in etcd. This seems cool, I guess, but also kinda like magic unicorn sauce?

Under the hood: Amazon EKS ultra scale clusters | Amazon Web Services This post was co-authored by Shyam Jeedigunta, Principal Engineer, Amazon EKS; Apoorva Kulkarni, Sr. Specialist Solutions Architect, Containers and Raghav Tripathi, Sr. Software Dev Manager, Amazon EK...

Finally finished reading the AWS post from a few weeks ago about their "ultra-scale" EKS perf improvements! There's some cool stuff in there, and lots of pretty graphs.

aws.amazon.com/blogs/contai...

"Chad Gippity" is even better imo.

Ok just booked my travel for KubeCon 2025! Looking forward to seeing folks there!

Not tired, not stressed, not dead: pick two

It's the year of Linux^H AGI on the desktop

bro just one more GPT bro. just one more you dont have to read or write emails anymore bro. one more GPT will fix everything bro

One of my biggest pet peeves is quickly becoming seeing times, dates, locations, or anything else that does not work as an absolute and context free reference.

I schedule things across continents, involving multiple parties in multiple time zones. Stop fucking saying “so, does 7 work?”

7 OF WHAT?!

I just purchased an ad on LinkedIn, I guess the devil has my soul now. Anyways, yay capitalism.

Astronomer: Saving Megabux with SQL and SimKube Ok, folks, get ready, this is a fun one!

Ok my blog post about my work with Astronomer is public now, go read it!

blog.appliedcomputing.io/p/astronomer...

GitHub - icedland/iced: Blazing fast and correct x86/x64 disassembler, assembler, decoder, encoder for Rust, .NET, Java, Python, Lua Blazing fast and correct x86/x64 disassembler, assembler, decoder, encoder for Rust, .NET, Java, Python, Lua - icedland/iced

Next up, Chapter 8, is Memory and Disassembly! I did a quick sneak peak ahead and discovered that the book relies on an external C++ lib for disassembly; fortunately, the `iced` crate for Rust seems to provide the functionality that we need here:

github.com/icedland/iced

Also, instead of shelling out to `readelf` in the tests, I just used the `elf` crate which already exists for Rust. I expect to rip this out in Chapter 11 when we implement our own ELF support in the debugger, but for now it feels less gross than readelf regex parsing.

I ended up making the breakpoint sites Clone-able, and then just wrapped the "important"/"shared" bits in Rc<Cell<...>>, which works pretty well aside from a couple dumb mistakes which took me too long to debug.

I tried to do it this way at first but ran into a point where I had to hold an immutable borrow across a mutable borrow. In the C++ code this is fine because they're clearly touching different parts of the code, but there's no way to tell the Rust borrow checker that, or restructure the code.

This chapter was reasonably straightforward. The hardest part was figuring out the right type for my breakpoint sites and convincing the borrow checker it was OK. In the C++ version of this code, Sy just returns a pointer to the breakpoint site and everything is fine.

GitHub - drmorr0/drbug: x86 debugger written in Rust x86 debugger written in Rust. Contribute to drmorr0/drbug development by creating an account on GitHub.

Chapter 7, "Software Breakpoints" of "Building a Debugger" by @tartanllama.xyz is done (in Rust)!

github.com/drmorr0/drbug

I don't know who needs to hear this, but QuickBooks is not your only option for accounting software. Xero is out there too, and it's cheaper and easier to use.

This is not an advertisement, I just finally got so mad at Intuit that I ragequit.

Astronomer: Saving Megabux with SQL and SimKube Ok, folks, get ready, this is a fun one!

Hey! Hey you! Do you want a sneak peek at the work I did over the last six months for Astronomer? Yes, that Astronomer! Sign up for a paid subscription and help me go to KubeCon this year. Or just read the article on Monday with all the normies, your choice.

blog.appliedcomputing.io/p/astronomer...