From 39a7f7d8b03edab76adbd6a1bfb7fadd65e47c83 Mon Sep 17 00:00:00 2001
From: Amanuel Engeda <amanueng@gmail.com>
Date: Sat, 28 Oct 2023 00:04:11 -0700
Subject: [PATCH] Soak testing

---
 .github/workflows/e2e-sock-trigger.yaml | 55 +++++++++++++++++++++++++
 .github/workflows/e2e.yaml              | 23 ++++++++---
 2 files changed, 73 insertions(+), 5 deletions(-)
 create mode 100644 .github/workflows/e2e-sock-trigger.yaml

diff --git a/.github/workflows/e2e-sock-trigger.yaml b/.github/workflows/e2e-sock-trigger.yaml
new file mode 100644
index 000000000000..d587adc897f4
--- /dev/null
+++ b/.github/workflows/e2e-sock-trigger.yaml
@@ -0,0 +1,55 @@
+name: E2ESoakTrigger
+on:
+  schedule:
+    - cron: '0 */3 * * *'
+  workflow_run:
+    workflows: [ApprovalComment]
+    types: [completed]
+  workflow_dispatch:
+    inputs:
+      region:
+        required: true
+        default: 'us-west-2'
+        type: choice
+        options:
+          - "us-east-1"
+          - "us-west-2"
+      cleanup:
+        required: true
+        default: true
+        type: boolean
+jobs:
+  reslove_cluster:
+    runs-on: ubuntu-latest
+    outputs:
+      CREATE_CLUSTER: ${{ steps.create_cluster.outputs.SHOULD_RUN }}
+      PREEXISTING: ${{ steps.create_cluster.outputs.GIT_REF }}
+    steps:
+        - id: create_cluster
+          run: |
+            export PREEXISTING=$(eksctl get cluster -o json | jq '.[].Name' | grep soak)
+            echo "Found existing cluster name \"$PREEXISTING\""
+            if [[ $PREEXISTING != "" ]]; then 
+                echo PREEXISTING=$PREEXISTING >> $GITHUB_OUTPUT
+                echo CREATE_CLUSTER=true >> $GITHUB_OUTPUT
+            else
+                echo CREATE_CLUSTER=false >> $GITHUB_OUTPUT
+            fi
+  sock:
+    needs: [reslove_cluster]
+    strategy:
+      fail-fast: false
+      matrix:
+        suite:
+        - Beta/Integration
+    uses: ./.github/workflows/e2e.yaml
+    with:
+      suite: ${{ matrix.suite }}
+      region: ${{ inputs.region || 'us-west-1' }}
+      workflow_trigger: "soak"
+      create_cluster: ${{needs.reslove_cluster.outputs.CREATE_CLUSTER }}
+      preexisiting_cluster: ${{needs.reslove_cluster.outputs.PREEXISTING || '' }}
+      # Default to true unless using a workflow_dispatch
+      cleanup: false
+    secrets:
+      SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}
diff --git a/.github/workflows/e2e.yaml b/.github/workflows/e2e.yaml
index d2dee5362735..3aa7d702f3cf 100644
--- a/.github/workflows/e2e.yaml
+++ b/.github/workflows/e2e.yaml
@@ -73,6 +73,10 @@ on:
         required: true
       workflow_trigger:
         type: string
+      preexisiting_cluster:
+        type: string
+      create_cluster:
+        type: string
     secrets:
       SLACK_WEBHOOK_URL:
         required: true
@@ -100,15 +104,22 @@ jobs:
           aws-region: ${{ inputs.region }}
           role-duration-seconds: 21600
       - name: add jitter on cluster creation
+        if: inputs.create_cluster
         run: |
           # Creating jitter so that we can stagger cluster creation to avoid throttling
           sleep $(( $RANDOM % 300 + 1 ))
       - id: generate-cluster-name
+        if: inputs.create_cluster
         run: |
-          CLUSTER_NAME=$(echo ${{ inputs.suite }}-$RANDOM$RANDOM | awk '{print tolower($0)}' | tr / -)
+          if [[ ${{ inputs.workflow_trigger }} == 'soak' ]]; then 
+            CLUSTER_NAME=$(echo ${{ inputs.workflow_trigger }}-$RANDOM$RANDOM | awk '{print tolower($0)}' | tr / -)
+          else 
+            CLUSTER_NAME=$(echo ${{ inputs.suite }}-$RANDOM$RANDOM | awk '{print tolower($0)}' | tr / -)
+          fi 
           echo "Using cluster name \"$CLUSTER_NAME\""
           echo CLUSTER_NAME=$CLUSTER_NAME >> $GITHUB_OUTPUT
-      - name: create eks cluster '${{ steps.generate-cluster-name.outputs.CLUSTER_NAME }}'
+      - name: create eks cluster '${{ steps.generate-cluster-name.outputs.CLUSTER_NAME || inputs.preexisiting_cluster }}'
+        if: inputs.create_cluster
         uses: ./.github/actions/e2e/create-cluster
         with:
           account_id: ${{ vars.ACCOUNT_ID }}
@@ -120,6 +131,7 @@ jobs:
           ip_family: ${{ contains(inputs.suite, 'IPv6') && 'IPv6' || 'IPv4' }} # Set the value to IPv6 if IPv6 suite, else IPv4
           git_ref: ${{ inputs.git_ref }}
       - name: install prometheus
+        if: inputs.create_cluster
         uses: ./.github/actions/e2e/install-prometheus
         with:
           account_id: ${{ vars.ACCOUNT_ID }}
@@ -129,6 +141,7 @@ jobs:
           workspace_id: ${{ vars.WORKSPACE_ID }}
           git_ref: ${{ inputs.git_ref }}
       - name: install karpenter
+        if: inputs.create_cluster
         uses: ./.github/actions/e2e/install-karpenter
         with:
           account_id: ${{ vars.ACCOUNT_ID }}
@@ -140,10 +153,10 @@ jobs:
           git_ref: ${{ inputs.git_ref }}
       - name: run the ${{ inputs.suite }} test suite
         run: |
-          aws eks update-kubeconfig --name ${{ steps.generate-cluster-name.outputs.CLUSTER_NAME }}
+          aws eks update-kubeconfig --name ${{ steps.generate-cluster-name.outputs.CLUSTER_NAME || inputs.preexisiting_cluster }}
           TEST_SUITE="${{ inputs.suite }}" ENABLE_METRICS=${{ inputs.enable_metrics }} METRICS_REGION=${{ vars.TIMESTREAM_REGION }} GIT_REF="$(git rev-parse HEAD)" \
-            CLUSTER_NAME="${{ steps.generate-cluster-name.outputs.CLUSTER_NAME }}" CLUSTER_ENDPOINT="$(aws eks describe-cluster --name ${{ steps.generate-cluster-name.outputs.CLUSTER_NAME }} --query "cluster.endpoint" --output text)" \
-            INTERRUPTION_QUEUE="${{ steps.generate-cluster-name.outputs.CLUSTER_NAME }}" make e2etests
+            CLUSTER_NAME="${{ steps.generate-cluster-name.outputs.CLUSTER_NAME || inputs.preexisiting_cluster }}" CLUSTER_ENDPOINT="$(aws eks describe-cluster --name ${{ steps.generate-cluster-name.outputs.CLUSTER_NAME || inputs.preexisiting_cluster }} --query "cluster.endpoint" --output text)" \
+            INTERRUPTION_QUEUE="${{ steps.generate-cluster-name.outputs.CLUSTER_NAME || inputs.preexisiting_cluster }}" make e2etests
       - name: notify slack of success or failure
         uses: ./.github/actions/e2e/slack/notify
         if: (success() || failure()) && github.event_name != 'workflow_run' && inputs.workflow_trigger != 'versionCompatibility'