Add workaround to prevent go build from hanging

[SalDaniele]

There is an issue that causes go multiarch to hang indefinitely sometimes: golang/go#70329

Per the comments on the issue, it seems setting GOMACPROCS to a small number can relieve this issue as a temporary WA

[thom311]

@SalDaniele are you sure this works?

you are setting the environment variable for running make local-buildx. However, that make step runs buildah build commands, which I think is the one that hangs (inside the containers).

I don't think external environment variables are passed down via buildah/podman, unless you pass buildah build --env.

I think what might works is (untested)

env["LOCAL_BUILDX_ARGS"] = "--env=GOMAXPROCS=2"

Or better patch this in dpu-operator instead of CDA. After all, it's dpu-operator that uses known-problematic base images like registry.ci.openshift.org/ocp/builder:rhel-9-golang-1.23-openshift-4.19. CDA shouldn't care what make local-buildx does or how to workaround some golang issue deep down in what dpu-operator does.

[thom311]

Or better patch this in dpu-operator instead of CDA. After all, it's dpu-operator that uses known-problematic base images like registry.ci.openshift.org/ocp/builder:rhel-9-golang-1.23-openshift-4.19. CDA shouldn't care what make local-buildx does or how to workaround some golang issue deep down in what dpu-operator does.

To be precise, the Dockerfiles that use the problematic base images and that run the hanging build command, should contain a ENV GOMAXPROCS=2. Or even better

RUN GOMAXPROCS=2 CGO_ENABLED=0 GOOS=${TARGETOS:-linux} GOARCH=${TARGETARCH} make build-daemon
#   ^^^^^^^^^^^^

[bn222]

Or better patch this in dpu-operator instead of CDA. After all, it's dpu-operator that uses known-problematic base images like registry.ci.openshift.org/ocp/builder:rhel-9-golang-1.23-openshift-4.19. CDA shouldn't care what make local-buildx does or how to workaround some golang issue deep down in what dpu-operator does.

This is a very good point. If you fix it in DPU operator, a developer can do a rebuild without depending on "logic" in CDA. @thom311 Thank you for your observation.

[SalDaniele]

Yes this makes sense, I will try adding this in the dpu operator directly