Gentoo Linux

cross-posted from: https://lemmy.srcfiles.zip/post/2231272

I've been interested in switching away from $PROPRIETARY_VENDOR's HPC node / cluster management offering for a while, and the opportunity has finally arisen - $VENDOR has decided to massively hike up their prices, so it's time to look at alternatives.

The top option on my list is Warewulf. Warewulf is a stateless node management tool, where stateless means "we boot any image you want into memory" (compared with competing implementations which do 'magic' to image a node's disk every boot). There are advantages and disadvantages to each approach.

The thing that attracts me most to Warewulf is that they've come to the conclusion that most HPC "disk images" are basically container images. Rather than using a chroot directory as an image (as do so many competing implementations) Warewulf have leaned wholeheartedly into the concept, and have adopted the OCI image tooling to define HPC images!

This offers an astounding amount of flexibility that the current $VENDOR solution does not - the ability to define, build, and run any (reasonable) flavour of Linux as an image for HPC nodes; images need only the kernel, networking, systemd, and (optionally) a nfs client (this is for convenience, it's not required for node functionality).

Based on that I've taken it upon myself to have some fun and investigate the current state of Warewulf as a node management tool - the first step, of course, was creating a Gentoo ebuild for Warewulf that compiles and installs. I'm happy to say that, after fixing some bugs in the offline build process, I have a working ebuild.

If you're a Gentooer with an interest in Node Management or HPC, please give my Warewulf ebuild a try; it compiles and installs but I haven't yet had a chance to do any real cluster management and I'm interested in hearing about any bugs you encounter!

Next Week(-ish): Gentoo HPC Base Images - I'm going to have to dig into Gentoo Networking to update the wwinit image.