Parameterize workflow timeouts

.github/assets/php/createImageBuilderDataYaml.php

@@ -9,6 +9,8 @@ function tabEverything($content, $tab = 7)
 $scriptBuild = tabEverything(file_get_contents($argv[1]));
 $scriptValidate = tabEverything(file_get_contents($argv[2]));
 
+$timeout = getenv('IMAGE_BUILDER_STEP_TIMEOUT_SECONDS') ?: '1200';
+
 $yamlContent = <<<EOD
 name: Install Server Software
 description: Installs server software for EC2 instances
@@ -18,7 +20,7 @@ function tabEverything($content, $tab = 7)
     steps:
       - name: CustomBuildEc2Image
         action: ExecuteBash
-        timeoutSeconds: 1200
+        timeoutSeconds: $timeout
         onFailure: Abort
         maxAttempts: 1
         inputs:

.github/workflows/aws.yml

@@ -167,6 +167,12 @@ on:
         required: false
         type: number
         default: 30
+
+      imageBuilderStepTimeoutSeconds:
+        description: 'Timeout for each image builder step'
+        required: false
+        type: number
+        default: 1200
       deployALB:
         description: 'Deploy an Application Load Balancer (true or false)'
         required: false
@@ -306,6 +312,60 @@ on:
         required: false
         type: string
 
+      vpcStackPropagationSeconds:
+        description: 'Time to wait for VPC stack exports to propagate'
+        required: false
+        type: number
+        default: 240
+
+      dnsPropagationTimeoutSeconds:
+        description: 'Maximum wait for DNS propagation when validating certificates'
+        required: false
+        type: number
+        default: 300
+
+      dnsPropagationCheckIntervalSeconds:
+        description: 'Interval between DNS propagation checks'
+        required: false
+        type: number
+        default: 10
+
+      albStackPauseSeconds:
+        description: 'Pause before creating the ALB stack'
+        required: false
+        type: number
+        default: 3
+
+      eksClusterCheckIntervalSeconds:
+        description: 'Interval between EKS cluster status checks'
+        required: false
+        type: number
+        default: 30
+
+      imageBuilderDistributionSleepSeconds:
+        description: 'Pause after creating or updating image builder distribution configuration'
+        required: false
+        type: number
+        default: 240
+
+      instanceRefreshInitialSleepSeconds:
+        description: 'Initial wait before polling instance refresh status'
+        required: false
+        type: number
+        default: 240
+
+      instanceRefreshCheckIntervalSeconds:
+        description: 'Interval between instance refresh status checks'
+        required: false
+        type: number
+        default: 60
+
+      instanceWarmupSeconds:
+        description: 'InstanceWarmup value for auto-scaling instance refresh'
+        required: false
+        type: number
+        default: 1200
+
     secrets:
       ENCRYPTION_KEY:
         required: false
@@ -478,8 +538,8 @@ jobs:
                 ParameterKey=HighlyAvailableNat,ParameterValue="${{ needs.CONSTANTS.outputs.highlyAvailableNat }}"  \
                 ParameterKey=EnableVpcFlowLogs,ParameterValue="${{ needs.CONSTANTS.outputs.enableVpcFlowLogs }}"
 
-            echo "Sleeping for 240 seconds to allow VPC stack exports to propagate. We've seen timing issues with the VPC stack exports not being available immediately for networkshares."
-            sleep 240
+            echo "Sleeping for ${{ inputs.vpcStackPropagationSeconds }} seconds to allow VPC stack exports to propagate. We've seen timing issues with the VPC stack exports not being available immediately for networkshares."
+            sleep ${{ inputs.vpcStackPropagationSeconds }}
 
           else
             echo "The VPC stack already exists on the AWS network account."
@@ -678,8 +738,8 @@ jobs:
 
             echo "Waiting for certificate to be validated for $domain ($cert)"
 
-            timeout=300  # seconds
-            interval=10  # seconds
+            timeout=${{ inputs.dnsPropagationTimeoutSeconds }}  # seconds
+            interval=${{ inputs.dnsPropagationCheckIntervalSeconds }}  # seconds
             elapsed=0
 
             while ! dig +short "$CNAME" >/dev/null; do
@@ -765,7 +825,7 @@ jobs:
           set -eEBx
           php "./.github/assets/php/createAlbYaml.php" "${{ env.certificates }}" > ./CloudFormation/alb.yaml
           cat ./CloudFormation/alb.yaml
-          sleep 3
+          sleep ${{ inputs.albStackPauseSeconds }}
           ./.github/assets/shell/createUpdateCFStack.sh ${{ matrix.aws-region }} alb \
             --template-body file://./CloudFormation/alb.yaml \
             --parameters 'ParameterKey=PublicSubnets,ParameterValue="${{ env.publicSubnet }}"'
@@ -1057,7 +1117,7 @@ jobs:
               echo "❌ Cluster failed to become active"
               exit 1
             fi
-            sleep 30
+            sleep ${{ inputs.eksClusterCheckIntervalSeconds }}
           done
 
       - name: Add new node group
@@ -1255,7 +1315,7 @@ jobs:
             aws imagebuilder create-distribution-configuration \
                         --name ${{ inputs.accountName }}-distribution-configuration \
                         "${DISTRIBUTION_ARGUMENTS[@]}"
-            sleep 240
+            sleep ${{ inputs.imageBuilderDistributionSleepSeconds }}
           else
             aws imagebuilder update-distribution-configuration \
                     --distribution-configuration-arn \
@@ -1287,6 +1347,7 @@ jobs:
           IMAGE-BUILDER-VALIDATE-EOF
 
           chmod +x ./.github/assets/shell/createUpdateImageBuilder.sh
+          export IMAGE_BUILDER_STEP_TIMEOUT_SECONDS="${{ inputs.imageBuilderStepTimeoutSeconds }}"
           ./.github/assets/shell/createUpdateImageBuilder.sh \
             "${{ env.distribution }}" \
             "${{ matrix.aws-region }}" \
@@ -1471,7 +1532,7 @@ jobs:
         if: ${{ steps.web.outputs.refresh == '1' && needs.IMAGE-BUILDER.outputs.image_rebuilt == '1' }}
         run: |
           REFRESH_ID=$(aws autoscaling start-instance-refresh \
-                --preferences '{"InstanceWarmup": 1200, "MinHealthyPercentage": 100}' \
+                --preferences '{"InstanceWarmup": ${{ inputs.instanceWarmupSeconds }}, "MinHealthyPercentage": 100}' \
                 --strategy Rolling \
                 --auto-scaling-group-name "${{ inputs.environment }}-${{ needs.CONSTANTS.outputs.repositoryNicename }}-${{ env.version }}.${{ github.run_number }}-asg" --output text)
 
@@ -1497,12 +1558,12 @@ jobs:
 
           getStatus
 
-          sleep 240
+          sleep ${{ inputs.instanceRefreshInitialSleepSeconds }}
 
           while [[ "$STATUS" == "Pending" || "$STATUS" == "InProgress" ]]; do
               getLog ${{ env.version }}
-              echo "Waiting 60 seconds... <$STATUS> (attempt:$TRY)"
-              sleep 60
+              echo "Waiting ${{ inputs.instanceRefreshCheckIntervalSeconds }} seconds... <$STATUS> (attempt:$TRY)"
+              sleep ${{ inputs.instanceRefreshCheckIntervalSeconds }}
               getStatus
           done
           if [[ "$STATUS" == "Successful" ]]; then