From 7b12d38e650c913c3f26412cfa323bc36d646429 Mon Sep 17 00:00:00 2001
From: Eric Promislow <epromislow@suse.com>
Date: Wed, 12 Jun 2024 11:57:20 -0700
Subject: [PATCH] Skip the RKE Machine config on CI/arm64

- Retry the integration tests one more time on failure
  - Document where the rerun syntax comes from.
---
 .github/workflows/ci.yaml                     | 39 +++++++------------
 .github/workflows/publish.yaml                |  1 +
 .github/workflows/release.yaml                |  1 +
 .github/workflows/scripts/integration-test-ci |  3 +-
 tests/integration/rkeMachineConfig_test.go    |  6 +++
 5 files changed, 25 insertions(+), 25 deletions(-)

diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
index 5f55ac29c..125a4c100 100644
--- a/.github/workflows/ci.yaml
+++ b/.github/workflows/ci.yaml
@@ -3,7 +3,6 @@
 name: Webhook CI
 
 on:
-  workflow_call:
   push:
     branches:
       - release/v*
@@ -16,6 +15,7 @@ on:
     - '.gitignore'
     - 'CODEOWNERS'
     - 'LICENSE'
+  workflow_dispatch:
 
 permissions:
   contents: read
@@ -23,16 +23,12 @@ permissions:
 jobs:
   build:
     name: CI
-    # This matrix syntax courtesy of stackoverflow.com/questions/66025220
     strategy:
       matrix:
         archBox:
         - { arch: amd64, vmArch: x64 }
         - { arch: arm64, vmArch: arm64 }
     runs-on: runs-on,runner=1cpu-linux-${{ matrix.archBox.vmArch }},run-id=${{ github.run_id }}
-# org-${{ github.repository_owner_id }}-${{ matrix.archBox.arch }}-k8s
-#    container:
-#      image: registry.suse.com/bci/nodejs:20
     steps:
       - name : Checkout repository
         # https://github.com/actions/checkout/releases/tag/v4.1.1
@@ -55,30 +51,13 @@ jobs:
           tar cfz "${{ runner.temp }}/rancher.tgz" -C build/chart/rancher .
           popd
 
-#          # Install yq
-#          F=yq_linux_${{ matrix.archBox.arch }}
-#          curl --output yq.tgz -L --silent https://github.com/mikefarah/yq/releases/download/v4.44.1/${F}.tar.gz && tar xfz yq.tgz && chmod +x $F && mv $F /usr/bin/yq
-#      - name: zypper-install non rancher/rancher prerecs
-#        run: zypper install -y make docker
-
-#      - name: Start the Docker Daemon
-#        run: 'nohup dockerd > "${{ runner.temp }}/docker.out" &
-
       - name: install K3d
         run: ./.github/workflows/scripts/install-k3d.sh
         env:
           K3D_VERSION: latest
 
-      - name: Setup Helm
-        # https://github.com/Azure/setup-helm/commit/fe7b79cd5ee1e45176fcad797de68ecaf3ca4814
-        uses: azure/setup-helm@fe7b79cd5ee1e45176fcad797de68ecaf3ca4814
-
-      - name: Setup Kubectl
-        # https://github.com/Azure/setup-kubectl/commit/3e0aec4d80787158d308d7b364cb1b702e7feb7f
-        uses: azure/setup-kubectl@3e0aec4d80787158d308d7b364cb1b702e7feb7f
-
-      - run: make ci
-        name: make ci
+      - name: ci
+        run: make ci
 
       - name: setup cluster
         run: ./.github/workflows/scripts/setup-cluster.sh
@@ -101,10 +80,22 @@ jobs:
         run: cat dist/image_tag >> $GITHUB_ENV
 
       - name: Run integration tests
+        id: integrationTests
         run: ./.github/workflows/scripts/integration-test-ci
+        # See https://docs.github.com/en/actions/learn-github-actions/contexts#steps-context
+        # for explanation of steps.continue-on-error and steps.ID.outcome
+        continue-on-error: true
         env:
           ARCH: "${{ matrix.archBox.arch }}"
           CLUSTER_NAME: webhook
           IMAGE_REPO: rancher/webhook
           IMAGE_TAG: "${{ env.IMAGE_TAG }}"
 
+      - name: Rerun integration tests one more time if they failed the first time
+        if: steps.integrationTests.outcome == 'failure'
+        run: ./.github/workflows/scripts/integration-test-ci
+        env:
+          ARCH: "${{ matrix.archBox.arch }}"
+          CLUSTER_NAME: webhook
+          IMAGE_REPO: rancher/webhook
+          IMAGE_TAG: "${{ env.IMAGE_TAG }}"
diff --git a/.github/workflows/publish.yaml b/.github/workflows/publish.yaml
index bcd907fd3..8363b04cf 100644
--- a/.github/workflows/publish.yaml
+++ b/.github/workflows/publish.yaml
@@ -4,6 +4,7 @@ on:
   push:
     tags:
       - v*
+  workflow_dispatch:
 
 env:
   REGISTRY: docker.io
diff --git a/.github/workflows/release.yaml b/.github/workflows/release.yaml
index 02b533db9..7405b6028 100644
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@@ -4,6 +4,7 @@ on:
   push:
     tags:
       - v*
+  workflow_dispatch:
 
 permissions:
   contents: write
diff --git a/.github/workflows/scripts/integration-test-ci b/.github/workflows/scripts/integration-test-ci
index d5552c89b..a1078b6b4 100755
--- a/.github/workflows/scripts/integration-test-ci
+++ b/.github/workflows/scripts/integration-test-ci
@@ -51,7 +51,8 @@ upgrade_rancher_webhook() {
 
 set +e
 
-upgrade_rancher_webhook
+try --max 3 --delay 2 --waitmsg "Upgrading Webhook" --failmsg "Failed to upgrade webhook" upgrade_rancher_webhook
+
 x=$(kubectl get pods -n cattle-system -l app=rancher-webhook | grep -v -e Running -e NAME | awk '{print $1}')
 if [ -n "$x" ] ; then
   echo "Logs for failed rancher-webhook $x ":
diff --git a/tests/integration/rkeMachineConfig_test.go b/tests/integration/rkeMachineConfig_test.go
index 7c8ae72c1..edcfabaf4 100644
--- a/tests/integration/rkeMachineConfig_test.go
+++ b/tests/integration/rkeMachineConfig_test.go
@@ -1,12 +1,18 @@
 package integration_test
 
 import (
+	"os"
+	"runtime"
+
 	"github.com/rancher/webhook/pkg/auth"
 	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
 	"k8s.io/apimachinery/pkg/runtime/schema"
 )
 
 func (m *IntegrationSuite) TestRKEMachineConfig() {
+	if runtime.GOARCH == "arm64" && os.Getenv("CI") != "" {
+		m.T().Skip("Skipping the RKE Machine-Config test on arm64 in CI -- machine info not available")
+	}
 	objGVK := schema.GroupVersionKind{
 		Group:   "rke-machine-config.cattle.io",
 		Version: "v1",