Dagger.jl VS buildkit

Firstly, I'll say that we already have work started to implement out-of-core directly in Dagger: https://github.com/JuliaParallel/Dagger.jl/pull/289.

With that PR in place, it should be possible to define a "storage device" which is backed by a database. I haven't had a chance to actually try this, since the PR still needs quite some work and testing, but it's definitely something on my radar!

I'm not sure that's necessarily the domain of a low-level package like CUDA.jl though (which I assume you're referring to). That kind of interface is more the domain of higher-level packages like https://github.com/JuliaParallel/Dagger.jl/ and to a lesser extent https://juliagpu.github.io/KernelAbstractions.jl/stable/. Moreover, the jury is still out on whether the built-in Distributed module is an ideal abstraction for every use-case (clusters, heterogeneous compute, etc.)

WRT Nx, my biggest question is how they'll crack the problem of still needing big balls of C++ and the shims everywhere to get acceleration. Creating a compiler that generates efficient GPU or other accelerator code is a massive research project with no clear winners, never mind the challenge of reconciling the very mutation-heavy needs of GPU compute with a mostly immutable language model.

> We are uding docker-in-docker at the moment

You can also run a "less privileged" container with all the features of Docker by using rootless buildkit in Kubernetes. Here are some examples:

https://github.com/moby/buildkit/tree/master/examples/kubern...

https://github.com/moby/buildkit/blob/master/examples/kubern...

It's also possible to run dedicated buildkitd workers and connect to them remotely.

- BuildOp evaluates its input as additional LLB operations to add to the graph to allow for dynamic build graphs (also unused in the Dockerfile frontend)

With the Dockerfile Explorer, we run the Dockerfile frontend[1] that BuildKit uses inside of WASM to parse and produce the LLB output locally in your browser. We then embed the Monaco Editor so that you can change your Dockerfile to see how it impacts the LLB output that BuildKit will use to build your Docker image.

You can see a quick video and read more details on how it all works here: https://depot.dev/blog/dockerfile-explorer.

We'd love any feedback or ideas folks would like around this type of tool!

[0] https://github.com/moby/buildkit#exploring-llb

Now since Kubernetes works off of containerd I'll be taking a different approach on handling container builds by using nerdctl and the buildkit that comes bundled with it. I'll do this on the amd64 control plane node since it's beefier than my Raspberry Pi workers for handling builds and build related services. Go ahead and download and unpack the latest nerdctl release as of writing (make sure to check the release page in case there's a new one):

There's a 2-year-old moby/buildkit GitHub issue about frequent build cache misses when using the BuildKit docker-container driver and multi-stage builds. Anyone else in this sub run into this problem and/or have reasonable workarounds? It seems like something that should come up pretty often.

A couple months ago I encountered a bug in buildkit - when enabling OpenTelemetry tracing, we got occasional panics. With a bit of investigation, we found the cause, fixed and tested in our fork and internal deployments, and pushed to upstream.

I do some search in the internet and there seems to be no good solution, so I just create a feature request: https://github.com/moby/buildkit/issues/3859

Yeah, only Linux containers at the moment, BuildKit is the way we are constructing pipelines and doing caching. Split on if we will support non-linux hosts, but definitely want to find a good solution to not doing Docker-in-Docker.

Relevant issues are https://github.com/moby/buildkit/issues/3011 and https://github.com/moby/buildkit/issues/1512.