Compiler gets killed in AArch64 CI

[straight-shoota]

aarch64-gnu-test-compiler is failing in CI relatively frequently. The compiler gets killed while building compiler_spec.
It's not consistently but seems to get more frequent.

The first failure on master was https://github.com/crystal-lang/crystal/actions/runs/13944742306/job/39029384603. But we've had previous unrelated failures like https://github.com/crystal-lang/crystal/actions/runs/13899834511/job/38888845048?pr=15305 and https://github.com/crystal-lang/crystal/actions/runs/13897500809/job/38881528755?pr=15558
The earliest failure I could find was last week https://github.com/crystal-lang/crystal/actions/runs/13788774130/job/38563172327?pr=15548

So it doesn't seem to be triggered by a specific change (at least it's not possible to identify such due to the flaky nature).

[Blacksmoke16]

Error 137 indicates it's running out of memory.

[straight-shoota]

Ah, good call.
aarch64-gnu-test-compiler runs on instances with 8GB of memory. We previously increased memory size of the test-stdlib jobs to 16GB in #15044. So we apparently need to increase the memory size for the test-compiler jobs as well.

It seems a bit odd that building the compiler specs would take more than 8GB though.
Although on my x86_64 machine it takes about 5.6 GB. 🤔 Maybe it's a bit more on aarch64, so it does not appear entirely implausible.

[Blacksmoke16]

https://github.blog/changelog/2025-01-16-linux-arm64-hosted-runners-now-available-for-free-in-public-repositories-public-preview/ is a thing now for public repos. So could just set it to run on ubuntu-24.04-arm and have it be free. Standard GH hosted linux runners all have 16GB.

[straight-shoota]

The cost factor isn't very significant. I think we can reconsider runs-on vs GitHub hosted-runners when the latter are out of beta/preview.