Reduce false positives in chart testing

Without this change, the "Test helm chart (install) action" will
rightfully succeed when our helm chart gets installed and has
no syntax issues. However, it doesn't test if kured is properly
installed. For example, the helm chart can try to install a
yet unpublished image, and our test will succeed, as the syntax
is still valid.

This is a problem, as everything looks green, but it's not
effectively working. Our other jobs are focusing on code changes,
so they rightfully override the image tag, which is not what
we want in this "Test helm chart" action.

This fixes it by adding an extra job in the workflow, depending
on the chart testing.

.github/workflows/on-pr-charts.yaml

@@ -39,3 +39,35 @@ jobs:
 
       - name: Run chart tests
         run: ct ${{ matrix.test-action }} --config .github/ct.yaml
+
+  # This doesn't re-use the ct actions, due to many limitations (auto tear down, no real testing)
+  deploy-chart:
+    name: "Functional test of helm chart with pre-published images"
+    runs-on: ubuntu-latest
+    needs: test-chart
+    steps:
+      - uses: actions/checkout@v2
+      
+      # Default name for helm/kind-action kind clusters is "chart-testing"
+      - name: Create 1 node kind cluster
+        uses: helm/kind-action@master
+
+      - name: Deploy kured on default namespace with its helm chart
+        run: |
+          # Documented in official helm doc to live on the edge
+          curl https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash
+          # Refresh bins
+          hash -r
+          helm install kured ./charts/kured/ --set configuration.period=1m --wait
+          kubectl config set-context kind-chart-testing
+          kubectl get ds --all-namespaces
+          kubectl describe ds kured
+
+      - name: Test if successful deploy
+        uses: nick-invision/retry@v2.4.0
+        with:
+          timeout_minutes: 10
+          max_attempts: 10
+          retry_wait_seconds: 10
+          # DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE should all be = to cluster_size
+          command: "kubectl get ds kured | grep -E 'kured.*1.*1.*1.*1.*1'"