For the past couple of years we’ve had access to a sponsored aarch64 server thanks to the Works on Arm project.
We’ve used this to host CI runners for GitHub Actions. This has been very valuable to ensure Crystal works well on Arm.
This sponsorship has come to an end and the server will shut off tomorrow. At this point we’ll have no means to run aarch64-linux CI runners for GitHub Actions.
How much of a server do you need? I have an 8-core ARM board here and am happy to register it as a runner. Heck, I have a Pi4 doing nothing, will happily turn it on and give you people a ssh key into it.
I’d be happy to donate capacity in one of my Kubernetes clusters. I can commit up to 8 vCPUs and 32GB of RAM on GCP T2A nodes.
Nearly all of my services run on T2A nodes on GCP (I was getting them for free for almost 2 years), so I have a vested interest in keeping aarch64-linux green.
I don’t think Raspberry Pis are really feasible as CI runners. The compiler build jobs need high compute power. An rpi would take quite long to complete the jobs.
For reference: Our current workload is handled by 8 runners, each with 16 GB and 4 cores of a Neoverse N1.
This was maybe a bit overallocated. Half of the memory and cores per runner would probably be fine. We’re limited on single-thread performance, because compilation isn’t easily parallelizable.
They were working 450 hours of CI jobs last month, so very low utilization. But we often have multiple CI workflows running in parallel, so it can be a bit congested at peak hours.
Overall, setup and continued maintenance of the CI configuration are a relevant cost factor. I’d rather have it require as little effort as possible.
Ideally we’ll get access to sponsored arm runners soon and we can wait a couple of weeks for that.
If this doesn’t work out we may need a different solution to bridge the gap until GitHub offers free arm runners, either with comunity-provided machines or a paid solution (maybe something like https://runs-on.com/ could work well?).
