ci: move CI to GitHub Actions

[aktech]

Fixes #346

First pass at converting from .travis.yml to GitHub Actions.

Notes:

• This doesn't includes skipping tests based on commit message at the moment, not sure if we need them (thoughts?)
• Currently runs on all branches (will change before merge)
• The tests are failing (probably because they are failing in Travis as well)
• Fixing flaky tests is beyond the scope of this, we can fix them, after this is merged (?).
• Also, it seems most of the tests haven't ran for a long long time except the main tests.

Here are the sample runs:
Docs: https://github.com/aktech/dask-gateway/actions/runs/1024666853
Tests: https://github.com/aktech/dask-gateway/actions/runs/1024666852

@jacobtomlinson @martindurant

[martindurant]

cc @douglasdavis

[TomAugspurger]

Do we have any hope for fixing the failing tests with CI running on Travis and then merging this once all the tests are passing?

[martindurant]

Are we missing YARN tests, or are those all the Errored ones I am seeing?

[aktech]

Are we missing YARN tests, or are those all the Errored ones I am seeing?

They are marked as hadoop-tests

Do we have any hope for fixing the failing tests with CI running on Travis and then merging this once all the tests are passing?

We can, although GitHub Actions doesn't need to "replace" Travis yet. Also, I am not sure how much effort is that, I'll check it though.

[martindurant]

They are marked as hadoop-tests

Oh! And they are green :)

[aktech]

pbs-tests and slurm-tests tests are sometimes flaky though.

[consideRatio]

Wieee nice work on this @aktech!

In case you think it's useful, I want to highlight github actions we developed in the JupyterHub organization as part of creating CI infrastructure for the JupyterHub and BinderHub helm charts.

An action to setup a k8s cluster - includes install of kubectl and helm, and includes a reliable NetworkPolicy controller to validate such functions

Full context of example here, and a link to the github action's documentation.

      # Starts a k8s cluster with NetworkPolicy enforcement and installs both
      # kubectl and helm
      #
      # ref: https://github.com/jupyterhub/action-k3s-helm/
      - uses: jupyterhub/action-k3s-helm@v1
        with:
          k3s-channel: ${{ matrix.k3s-channel }}
          metrics-enabled: false
          traefik-enabled: false
          docker-enabled: true

An action to provide insights if something went wrong or not

Full context of example here, and a link to the github action's documentation.

      # ref: https://github.com/jupyterhub/action-k8s-namespace-report
      - name: Kubernetes namespace report
        uses: jupyterhub/action-k8s-namespace-report@v1
        if: always()
        with:
          important-workloads: deploy/hub deploy/proxy

An action to await installed Helm charts efficiently - more informative than helm install --wait

Full context of example here, and a link to the github action's documentation.

      # It is only needed at this point forward as this is when we install
      # jupyterhub and the autohttps pod is about to start, so for CI
      # performance we delayed this until now and did other things in between.
      - name: Await local ACME server
        uses: jupyterhub/action-k8s-await-workloads@v1
        with:
          timeout: 150
          max-restarts: 1

[martindurant]

Those do look like useful action recipes! I should not be surprised that k8s-on-gh is a well-developed space.

[aktech]

@consideRatio That's excellent stuff! thanks for sharing.

[jacobtomlinson]

Another tool that could be useful that we use over on dask-kubernetes is pytest-kind which handles k8s cluster creation as a pytest fixture. Makes local testing consistent with CI testing. Docker is your only dependency.

[aktech]

Travis tests are fixed in this PR: #411

[martindurant]

Can you merge your fixes in here and rerun your actions, please?

[aktech]

Just ran: https://github.com/aktech/dask-gateway/actions/runs/1060226799

All green except the kubernetes tests, I haven't had the chance to looked into it yet. Can we fix that in another PR maybe?

There maybe some flaky tests, which needs to be fixed but would be good to get the CI in first, I guess.

[aktech]

Note that I had to use the pytest marker (kubernetes) to mark the kubernetes test to avoid running them in the main tests job as helm is probably present by default in GitHub Actions, hence the k8s tests are not skipped by default.

[martindurant]

That all seems fine to me. Anyone else here have an opinion? If no -1, I intend to merge, and we can then iterate with GA tests.

[martindurant]

Thank you @aktech !

[martindurant]

https://github.com/dask/dask-gateway/actions/runs/1067491048

[aktech]

Thanks for getting this in @martindurant 🎉

[martindurant]

We should start a new issue discussing what it will take to get kubernetes passing

[aktech]

We should start a new issue discussing what it will take to get kubernetes passing

Yep, makes sense. I think I might have some idea.

[aktech]

I have created an issue here: #412