fix: only execute cloud-init-status-check.sh when it exists on the node (#7456)

Author: Devinwong

.vscode/settings.json

@@ -75,5 +75,6 @@
   ],
   "files.trimTrailingWhitespace": true,
   "files.trimTrailingWhitespaceInRegexAndStrings": true,
-  "files.insertFinalNewline": true
+  "files.insertFinalNewline": true,
+  "go.testTimeout": "30m"
 }

aks-node-controller/parser/templates/cse_cmd.sh.gtpl

@@ -1,11 +1,22 @@
 echo $(date),$(hostname) > ${PROVISION_OUTPUT};
 {{if not .GetDisableCustomData}}
-cloud-init status --wait > /dev/null 2>&1;
-[ "$?" -ne 0 ] && echo 'cloud-init failed' >> ${PROVISION_OUTPUT} && exit 1;
-echo "cloud-init succeeded" >> ${PROVISION_OUTPUT};
+CLOUD_INIT_STATUS_SCRIPT="/opt/azure/containers/cloud-init-status-check.sh";
+cloudInitExitCode=0;
+if [ -f "${CLOUD_INIT_STATUS_SCRIPT}" ]; then
+	/bin/bash -c "source ${CLOUD_INIT_STATUS_SCRIPT}; handleCloudInitStatus \"${PROVISION_OUTPUT}\"; returnStatus=\$?; echo \"Cloud init status check exit code: \$returnStatus\" >> ${PROVISION_OUTPUT}; exit \$returnStatus" >> ${PROVISION_OUTPUT} 2>&1;
+else
+    cloud-init status --wait > /dev/null 2>&1;
+fi;
+cloudInitExitCode=$?;
+if [ "$cloudInitExitCode" -eq 0 ]; then
+	echo "cloud-init succeeded" >> ${PROVISION_OUTPUT};
+else
+	echo "cloud-init failed with exit code ${cloudInitExitCode}" >> ${PROVISION_OUTPUT};
+	exit ${cloudInitExitCode};
+fi;
 {{end}}
 {{if getIsAksCustomCloud .CustomCloudConfig}}
 REPO_DEPOT_ENDPOINT="{{.CustomCloudConfig.RepoDepotEndpoint}}"
 {{getInitAKSCustomCloudFilepath}} >> /var/log/azure/cluster-provision.log 2>&1;
 {{end}}
-/usr/bin/nohup /bin/bash -c "/bin/bash /opt/azure/containers/provision_start.sh"
+/usr/bin/nohup /bin/bash -c "/bin/bash /opt/azure/containers/provision_start.sh"

parts/linux/cloud-init/artifacts/cse_cmd.sh

@@ -1,9 +1,20 @@
 PROVISION_OUTPUT="/var/log/azure/cluster-provision-cse-output.log";
 echo $(date),$(hostname) > ${PROVISION_OUTPUT};
 {{if ShouldEnableCustomData}}
-/bin/bash -c "source /opt/azure/containers/cloud-init-status-check.sh; handleCloudInitStatus \"${PROVISION_OUTPUT}\"; returnStatus=\$?; echo \"Cloud init status check exit code: \$returnStatus\" >> ${PROVISION_OUTPUT}; exit \$returnStatus" >> ${PROVISION_OUTPUT} 2>&1;
+CLOUD_INIT_STATUS_SCRIPT="/opt/azure/containers/cloud-init-status-check.sh";
+cloudInitExitCode=0;
+if [ -f "${CLOUD_INIT_STATUS_SCRIPT}" ]; then
+	/bin/bash -c "source ${CLOUD_INIT_STATUS_SCRIPT}; handleCloudInitStatus \"${PROVISION_OUTPUT}\"; returnStatus=\$?; echo \"Cloud init status check exit code: \$returnStatus\" >> ${PROVISION_OUTPUT}; exit \$returnStatus" >> ${PROVISION_OUTPUT} 2>&1;
+else
+    cloud-init status --wait > /dev/null 2>&1;
+fi;
 cloudInitExitCode=$?;
-[ "$cloudInitExitCode" -ne 0 ] && echo "cloud-init failed with exit code ${cloudInitExitCode}" >> ${PROVISION_OUTPUT} && exit ${cloudInitExitCode};
+if [ "$cloudInitExitCode" -eq 0 ]; then
+	echo "cloud-init succeeded" >> ${PROVISION_OUTPUT};
+else
+	echo "cloud-init failed with exit code ${cloudInitExitCode}" >> ${PROVISION_OUTPUT};
+	exit ${cloudInitExitCode};
+fi;
 {{end}}
 {{if IsAKSCustomCloud}}
 REPO_DEPOT_ENDPOINT="{{AKSCustomCloudRepoDepotEndpoint}}"

pkg/agent/baker.go

@@ -774,6 +774,9 @@ func getContainerServiceFuncMap(config *datamodel.NodeBootstrappingConfiguration
 			return datamodel.AzureADIdentitySystem
 		},
 		"GetPodInfraContainerSpec": func() string {
+			if config.K8sComponents == nil {
+				return ""
+			}
 			return config.K8sComponents.PodInfraContainerImageURL
 		},
 		"IsKubenet": func() bool {
@@ -838,9 +841,15 @@ func getContainerServiceFuncMap(config *datamodel.NodeBootstrappingConfiguration
 			return cs.Properties.HasDCSeriesSKU()
 		},
 		"GetHyperkubeImageReference": func() string {
+			if config.K8sComponents == nil {
+				return ""
+			}
 			return config.K8sComponents.HyperkubeImageURL
 		},
 		"GetLinuxPrivatePackageURL": func() string {
+			if config.K8sComponents == nil {
+				return ""
+			}
 			return config.K8sComponents.LinuxPrivatePackageURL
 		},
 		"GetTargetEnvironment": func() string {