@austinlparker is this still needed? I thought we do multi-arch builds already.

We do not do multi arch builds because it takes like 4+ hours.

I have added it in particular because the featureflagservice keeps segfaulting on my M1 device with and without emulation.

I think it makes a lot of sense to have this available for multiple architectures. If the build time is a problem, it's probably a question of optimization

I also thought we fixed this but I just tried installing the helm chart locally again and featureflag service is crashing again (open-telemetry/opentelemetry-helm-charts#580).

Is there a way that we can do the existing build and publish process and separately build a specific arm64 image either after the existing build finishes or in parallel?

Is there a way that we can do the existing build and publish process and separately build a specific arm64 image either after the existing build finishes or in parallel?

We sure could. Not sure if this benefits us. Currently all images are built in parallel and in each build sequentially builds and pushed amd64 and arm64. I would argue there is a fair amount of parallelism already in place.

Current duration for the Build & Push job is [19m 47s] - so overall pretty far away from the projected "4hrs" :)

No, it takes about 4 hours: https://github.com/open-telemetry/opentelemetry-demo/actions/runs/3313405848

It might be better now with some of the optimizations we've made to build time but Rust and C++ images take forever to build under emulation.

I would suggest modifying the GHA runner file here to allow builds for PR's (you need to modify the release.yaml file to allow the build_and_push_images action to run on PR and add logic to disable pushing images on non-release builds) in order to see how long it takes.

This PR was marked stale due to lack of activity. It will be closed in 7 days.

I went ahead and tested this out on our runner config -- everything builds OK. Shipping service takes an hour which isn't great, but we can work on getting that down.