Add integration tests for arm64

[dharmit]

No description provided.

[mallardduck]

I think once the OS fields have both values the action will work as expected. 👍

[mallardduck]

Since i've just pushed our new main branch, I've repointed this PR to target main.
It should be a smooth transition since main is based on release/v5.0.

[mallardduck]

Ok @dharmit - I put some more dots together here in this puzzle for us.
The labels we saw Alex working with weren't wrong, just 4 months old before our full adoption of GHA.
And things must have changed since then by a lot since our new docs look a bit different: https://github.com/rancherlabs/eio/wiki/GitHub-Actions-Runners

[dharmit]

And things must have changed since then by a lot since our new docs look a bit different: https://github.com/rancherlabs/eio/wiki/GitHub-Actions-Runners

They don't mention arm64 in containerized runners section. All occurrences of arm64 are under the "Non-Containerized Org-Level Runners (VMs and Bare Metal Instances)" section. 🤔

Also that doc doesn't mention "ubuntu-latest" anywhere, and yet we are using it in our tests.

[mallardduck]

I think we just need to try: https://github.com/rancherlabs/eio/wiki/GitHub-Actions-Runners#containerized-org-level-runners

And I suspect that they have arm ones ready to go?

[mallardduck]

@dharmit - Disregard my suggestion to use a new CI file, I've worked with EIO to get this repo access to non-container runners. This provides us a much more direct drop-in replacement for the GHA runners. With those setup I was able to tweak this PR to use those runners for both arm and amd64 without major changes.

[mallardduck]

Still have some changes to be made before this works, I will leave these for you to resolve though now that we know the runners are sorted out and working. 😄

[mallardduck]

Might have to run workflows as debug to get additional logging. Looks like we're getting close but now seeing a new error related to no ARCH set and go binary missing.

[dharmit]

Looks like we're getting close but now seeing a new error related to no ARCH set and go binary missing.

I think the two are related because ARCH is set in ./scripts/version as:

ARCH=$TARGET_ARCH
if [ -z "$ARCH" ]; then
    ARCH=$(go env GOHOSTARCH)
fi

Since go binary isn't available, both problems are popping up. Let me try to install go.

[ericpromislow]

Please use uname -m instead. -i isn't supported on macOS and is documented as non-portable on ubuntu. As much as I can tell, they do the same thing.

[ericpromislow]

It's clearer to set PLATFORM=amd64|arm64 in a separate case block to avoid duplicating a long command.

But you should probably set this in ci.yaml and set the variables if they're unset so .github/scripts/X can still be run outside GHA, mostly for testing/dev purposes.

[ericpromislow]

Instead do

set -eu
set -x

where the set -x could get dropped at a later date. And our scripts should always fail on unset vars

[ericpromislow]

I would drop the entire install-go.sh file and replace it by running actions/setup-go in ci.yaml

Also see how we deal with the two different ways of naming the amd architecture in webhook/.../ci.yaml. And I would then try to pass a value for ARCH to install-mc.sh and for TARGET_ARCH for ./scripts/integration

[dharmit]

I would drop the entire install-go.sh file and replace it by running actions/setup-go in ci.yaml

Thanks. I wasn't aware of it. 👍🏽

Also see how we deal with the two different ways of naming the amd architecture in webhook/.../ci.yaml. And I would then try to pass a value for ARCH to install-mc.sh and for TARGET_ARCH for ./scripts/integration

I haven't made any changes that you suggested here. And, I think, it didn't cause any trouble because go installation happened before we run integration tests.

[mallardduck]

/backport NONE release/v5.0