Add Github Actions

[cdesiniotis]

No description provided.

[elezar]

As a general comment, does it make sense to split the actions based on scope. For example in the device plugin we have:

.github/workflows
├── e2e.yaml
├── golang.yaml
├── image.yaml
└── stale.yaml

1 directory, 4 files

Here golang.yaml and image.yaml are generally generic enough to reuse across all our components. We can add a separate file for the other validation (linting etc).

[ArangoGutierrez]

Upcoming holodeck release won't require this line :)

[ArangoGutierrez]

NVIDIA/holodeck#63

[cdesiniotis]

As a general comment, does it make sense to split the actions based on scope. For example in the device plugin we have:

.github/workflows
├── e2e.yaml
├── golang.yaml
├── image.yaml
└── stale.yaml

1 directory, 4 files

Here golang.yaml and image.yaml are generally generic enough to reuse across all our components. We can add a separate file for the other validation (linting etc).

In general I agree we should split the actions and align common workflows, like golang.yaml and image.yaml, across all of our projects. And @ArangoGutierrez I know we discussed your plans for removing some of the current limitations, like not being able to run e2e tests on PRs, in the future. However, are you okay if we keep this PR as is for the time being so that we mimic the existing pipelines on Gitlab as much as possible? I am happy to re-factor this into multiple files and align with other projects once the limitations are addressed.

To prevent e2e tests from running on forks (which will likely fail anyways if AWS creds are not configured on the fork), we can consider adding the below conditional to the e2e jobs:

if: github.event.pull_request.head.repo.full_name == "github.com/NVIDIA/gpu-operator"

[ArangoGutierrez]

Let's add this line to ensure we only run e2e on merge for now

[ArangoGutierrez]

@cdesiniotis holodeck is fixed, now it fails for real reasons

[cdesiniotis]

@cdesiniotis holodeck is fixed, now it fails for real reasons

Thanks. The e2e jobs are failing because the container images for this repo have not been made public yet. I have pinged George to make them public.

[tariq1890]

To prevent e2e tests from running on forks (which will likely fail anyways if AWS creds are not configured on the fork), we can consider adding the below conditional to the e2e jobs:
if: github.event.pull_request.head.repo.full_name == "github.com/NVIDIA/gpu-operator"

I think we can get around this by requiring manual approves by the code owners for e2e github pipelines in PRs at all times.

[cdesiniotis]

All checks have passed. @ArangoGutierrez @elezar could you take another look?

For now, the Github Actions resemble our Gitlab pipelines. In a follow-up, we can look to restrict when the e2e test jobs run.

[tariq1890]

Can we choose a different name here? Perhaps, something along the lines of gpu-operator-e2e-env ?

[cdesiniotis]

I believe the holodeck action automatically generates a name, but I could be wrong. @ArangoGutierrez what is the recommendation for naming this?

[elezar]

Since this is the same as for the device plugin, let's leave it as is for now -- pending guidance on what this should be.

To clarify, the name is set here for the github action: https://github.com/NVIDIA/holodeck/blob/58fe703797a66510af3598bfca139fbaa46f47fa/cmd/action/ci/ci.go#L58-L69

Comments addressed.

[elezar]

Thanks @cdesiniotis. I'm happy to start with this and iterate once we have the initial kinks ironed out.