PWX-38589: Fix the CPU spike and excessive logging in telemetry pods

libopenstorage · Aug 19, 2024 · 534f548 · 534f548
1 parent 303cc37
commit 534f548
Show file tree

Hide file tree

Showing 7 changed files with 41 additions and 10 deletions.
diff --git a/deploy/ccm/ccm.properties b/deploy/ccm/ccm.properties
@@ -28,7 +28,7 @@
       "/var/cores/px_info.log",
       "/var/cores/px_patch_fs.log"
     ],
-    "phonehome_hour_range": 8760,
+    "phonehome_hour_range": 744,
     "phonehome_sent": "/var/logs/phonehome.sent",
     "always_scan_range_days": 7,
     "max_retry_per_hour": 5,

diff --git a/drivers/storage/portworx/component/telemetry.go b/drivers/storage/portworx/component/telemetry.go
@@ -88,6 +88,7 @@ const (
 	daemonsetFileNameTelemetryPhonehome              = "phonehome-cluster.yaml"
 
 	configParameterApplianceID                           = "APPLIANCE_ID"
+	configParameterApplianceName                         = "APPLIANCE_NAME"
 	configParameterComponentSN                           = "COMPONENT_SN"
 	configParameterProductVersion                        = "PRODUCT_VERSION"
 	configParameterRegisterProxyURL                      = "REGISTER_PROXY_URL"
@@ -588,6 +589,7 @@ func (t *telemetry) createCCMGoConfigMapRegisterProxy(
 	cloudSupportPort, tcpProxyPort, envoyRedirectPort := getCCMCloudSupportPorts(cluster, defaultRegisterPort)
 	replaceMap := map[string]string{
 		configParameterApplianceID:              cluster.Status.ClusterUID,
+		configParameterApplianceName:            cluster.Status.ClusterName,
 		configParameterComponentSN:              cluster.Name,
 		configParameterProductVersion:           pxutil.GetPortworxVersion(cluster).String(),
 		configParameterRegisterProxyURL:         getArcusRegisterProxyURL(cluster),
@@ -644,6 +646,7 @@ func (t *telemetry) createCCMGoConfigMapTelemetryPhonehomeProxy(
 	cloudSupportPort, tcpProxyPort, envoyRedirectPort := getCCMCloudSupportPorts(cluster, defaultPhonehomePort)
 	replaceMap := map[string]string{
 		configParameterApplianceID:          cluster.Status.ClusterUID,
+		configParameterApplianceName:        cluster.Status.ClusterName,
 		configParameterProductVersion:       pxutil.GetPortworxVersion(cluster).String(),
 		configParameterRestProxyURL:         getArcusRestProxyURL(cluster),
 		configParameterRestCloudSupportPort: fmt.Sprint(cloudSupportPort),
@@ -701,6 +704,7 @@ func (t *telemetry) createCCMGoConfigMapCollectorProxyV2(
 	cloudSupportPort, tcpProxyPort, envoyRedirectPort := getCCMCloudSupportPorts(cluster, defaultCollectorPort)
 	replaceMap := map[string]string{
 		configParameterApplianceID:          cluster.Status.ClusterUID,
+		configParameterApplianceName:        cluster.Status.ClusterName,
 		configParameterProductVersion:       pxutil.GetPortworxVersion(cluster).String(),
 		configParameterRestProxyURL:         getArcusRestProxyURL(cluster),
 		configParameterRestCloudSupportPort: fmt.Sprint(cloudSupportPort),
@@ -830,6 +834,10 @@ func (t *telemetry) createDeploymentTelemetryRegistration(
 				Name:  configParameterApplianceID,
 				Value: cluster.Status.ClusterUID,
 			})
+			container.Env = append(container.Env, v1.EnvVar{
+				Name:  configParameterApplianceName,
+				Value: cluster.Status.ClusterName,
+			})
 		} else if container.Name == containerNameTelemetryProxy {
 			container.Image = proxyImage
 		}
@@ -893,6 +901,15 @@ func (t *telemetry) createDaemonSetTelemetryPhonehome(
 		container := &daemonset.Spec.Template.Spec.Containers[i]
 		if container.Name == containerNameLogUploader {
 			container.Image = logUploaderImage
+			// add APPLIANCE_ID env var
+			container.Env = append(container.Env, v1.EnvVar{
+				Name:  configParameterApplianceID,
+				Value: cluster.Status.ClusterUID,
+			})
+			container.Env = append(container.Env, v1.EnvVar{
+				Name:  configParameterApplianceName,
+				Value: cluster.Status.ClusterName,
+			})
 			for j := 0; j < len(container.Ports); j++ {
 				port := &container.Ports[j]
 				if port.Name == portNameLogUploaderContainer {

diff --git a/drivers/storage/portworx/components_test.go b/drivers/storage/portworx/components_test.go
@@ -15189,7 +15189,8 @@ func TestTelemetryCCMGoEnableAndDisable(t *testing.T) {
 			},
 		},
 		Status: corev1.StorageClusterStatus{
-			ClusterUID: "test-clusteruid",
+			ClusterUID:  "test-clusteruid",
+			ClusterName: "test-clustername",
 		},
 	}
 
@@ -15447,7 +15448,8 @@ func TestTelemetryContainerOrchestratorEnable(t *testing.T) {
 			},
 		},
 		Status: corev1.StorageClusterStatus{
-			ClusterUID: "test-clusteruid",
+			ClusterName: "test-clustername",
+			ClusterUID:  "test-clusteruid",
 		},
 	}
 
@@ -15466,7 +15468,7 @@ func TestTelemetryContainerOrchestratorEnable(t *testing.T) {
 	deployment := &appsv1.Deployment{}
 	err = testutil.Get(k8sClient, deployment, component.DeploymentNameTelemetryRegistration, cluster.Namespace)
 	require.NoError(t, err)
-	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 2)
+	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 3)
 
 	// Compatible PX & Incompatible Telemetry Images
 	cluster.Spec.Image = "portworx/image:3.2.0"
@@ -15483,7 +15485,7 @@ func TestTelemetryContainerOrchestratorEnable(t *testing.T) {
 	// Validate deployments
 	err = testutil.Get(k8sClient, deployment, component.DeploymentNameTelemetryRegistration, cluster.Namespace)
 	require.NoError(t, err)
-	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 2)
+	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 3)
 
 	// Incompatible PX & compatible Telemetry Images
 	cluster.Spec.Image = "portworx/image:3.0.0"
@@ -15500,7 +15502,7 @@ func TestTelemetryContainerOrchestratorEnable(t *testing.T) {
 	// Validate deployments
 	err = testutil.Get(k8sClient, deployment, component.DeploymentNameTelemetryRegistration, cluster.Namespace)
 	require.NoError(t, err)
-	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 2)
+	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 3)
 
 	// Compatible PX & Telemetry Images
 	cluster.Spec.Image = "portworx/image:3.2.0"
@@ -15518,13 +15520,15 @@ func TestTelemetryContainerOrchestratorEnable(t *testing.T) {
 	// Validate deployments
 	err = testutil.Get(k8sClient, deployment, component.DeploymentNameTelemetryRegistration, cluster.Namespace)
 	require.NoError(t, err)
-	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 3)
+	require.Len(t, deployment.Spec.Template.Spec.Containers[0].Env, 4)
 	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[0].Name, "CONFIG")
 	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[0].Value, "config/config_properties_px.yaml")
 	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[1].Name, "APPLIANCE_ID")
 	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[1].Value, "test-clusteruid")
-	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[2].Name, "REFRESH_TOKEN")
-	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[2].Value, "")
+	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[2].Name, "APPLIANCE_NAME")
+	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[2].Value, "test-clustername")
+	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[3].Name, "REFRESH_TOKEN")
+	require.Equal(t, deployment.Spec.Template.Spec.Containers[0].Env[3].Value, "")
 
 	// Port shift on OCP
 	cluster.Annotations[pxutil.AnnotationIsOpenshift] = "true"

diff --git a/drivers/storage/portworx/testspec/ccmGoPhonehomeConfigMap.yaml b/drivers/storage/portworx/testspec/ccmGoPhonehomeConfigMap.yaml
@@ -40,7 +40,7 @@ data:
           "/var/cores/px_info.log",
           "/var/cores/px_patch_fs.log"
         ],
-        "phonehome_hour_range": 8760,
+        "phonehome_hour_range": 744,
         "phonehome_sent": "/var/logs/phonehome.sent",
         "always_scan_range_days": 7,
         "max_retry_per_hour": 5,

diff --git a/drivers/storage/portworx/testspec/ccmGoPhonehomeDaemonSet.yaml b/drivers/storage/portworx/testspec/ccmGoPhonehomeDaemonSet.yaml
@@ -49,6 +49,10 @@ spec:
           valueFrom:
             fieldRef:
               fieldPath: spec.nodeName
+        - name: APPLIANCE_ID
+          value: test-clusteruid
+        - name: APPLIANCE_NAME
+          value: test-clustername
         image: docker.io/purestorage/log-upload:1.2.3
         imagePullPolicy: Always
         name: log-upload-service

diff --git a/drivers/storage/portworx/testspec/ccmGoPhonehomeDaemonSetDefaultPort.yaml b/drivers/storage/portworx/testspec/ccmGoPhonehomeDaemonSetDefaultPort.yaml
@@ -49,6 +49,10 @@ spec:
           valueFrom:
             fieldRef:
               fieldPath: spec.nodeName
+        - name: APPLIANCE_ID
+          value: test-clusteruid
+        - name: APPLIANCE_NAME
+          value: test-clustername
         image: docker.io/purestorage/log-upload:1.2.3
         imagePullPolicy: Always
         name: log-upload-service

diff --git a/drivers/storage/portworx/testspec/ccmGoRegisterDeployment.yaml b/drivers/storage/portworx/testspec/ccmGoRegisterDeployment.yaml
@@ -50,6 +50,8 @@ spec:
           value: config/config_properties_px.yaml
         - name: APPLIANCE_ID
           value: test-clusteruid
+        - name: APPLIANCE_NAME
+          value: test-clustername
         image: docker.io/portworx/px-telemetry:4.3.2
         imagePullPolicy: Always
         name: registration