Julia 1.11.2 hangs the aarch64 builds

[mathbunnyru]

More info: #2178

[mathbunnyru]

@giordano

Is there any way to reproduce the hang you observed on a local machine? Or at least, a clear indication of what's the workload you're running.

Hi,

1. The machine running the build is t2a-standard-1 on google cloud: https://gcloud-compute.com/t2a-standard-1.html
2. The only thing running on the machine is the GitHub actions runner, which actually performs the build
3. It hangs here (Downgrade to Julia 1.11.1 #2178 (comment)): https://github.com/jupyter/docker-stacks/blob/main/images/minimal-notebook/setup-scripts/setup-julia-packages.bash#L39

I think to reproduce you need to run:

1. https://github.com/jupyter/docker-stacks/blob/main/images/minimal-notebook/setup-scripts/setup_julia.py
2. https://github.com/jupyter/docker-stacks/blob/main/images/minimal-notebook/setup-scripts/setup-julia-packages.bash

Don't forget to revert the patch I made to not use 1.11.2

[giordano]

Sorry, I never came back to this because I don't have access to Google Cloud nor any aarch64-linux machine with Docker set up, but I just tried to run the following

JULIA_CPU_TARGET="generic;cortex-a57;thunderx2t99;carmel,clone_all;apple-m1,base(3);neoverse-512tvb,base(3)" julia --project=. -e '
import Pkg;
Pkg.update();
Pkg.add([
    "HDF5",
    "IJulia",
    "Pluto"
]);
Pkg.precompile();
'

with Julia v1.11.2 and v1.11.3 on Nvidia Grace, Apple M1 (with Asahi Linux), Ampere eMAG, and Ampere Altra Max M128-30 (the latter should be somewhat similar to the Google Cloud one) and all processes finished within 2-3 minutes for me.

[mathbunnyru]

Sorry, I never came back to this because I don't have access to Google Cloud nor any aarch64-linux machine with Docker set up, but I just tried to run the following

JULIA_CPU_TARGET="generic;cortex-a57;thunderx2t99;carmel,clone_all;apple-m1,base(3);neoverse-512tvb,base(3)" julia --project=. -e '
import Pkg;
Pkg.update();
Pkg.add([
"HDF5",
"IJulia",
"Pluto"
]);
Pkg.precompile();
'
with Julia v1.11.2 and v1.11.3 on Nvidia Grace, Apple M1 (with Asahi Linux), Ampere eMAG, and Ampere Altra Max M128-30 (the latter should be somewhat similar to the Google Cloud one) and all processes finished within 2-3 minutes for me.

Thanks for checking. I don’t think you should spend any more time on this, to be honest. As soon as GitHub-hosted runners start working properly, we’ll be able to close this issue, because these runners don’t have this problem.
And I created a PR already, just waiting GitHub to fix some issues with the runners (they affect almost everyone, so I think we won’t have to wait too long).