Merge pull request #4254 from LiilyZhang/zhangl/Issue4253

Issue #4253 - Bug: Cert version info lost after config au…

clusterupgrade/cluster_upgrade_worker.go

@@ -116,7 +116,7 @@ func (w *ClusterUpgradeWorker) syncOnInit(db *bolt.DB, kubeClient *KubeClient, b
 				glog.Infof(cuwlog(fmt.Sprintf("Handling status %v during initialization", name)))
 				workDir := path.Join(baseWorkingDir, name)
 				if statusInStatusFile, err := checkDeploymentStatus(kubeClient, baseWorkingDir, name); err != nil {
-					errMessage := fmt.Sprintf("Failed to check deployment status duriing syncOnInit for nmp: %v, error: %v", name, err)
+					errMessage := fmt.Sprintf("Failed to check deployment status during syncOnInit for nmp: %v, error: %v", name, err)
 					if err = setErrorMessageInStatusFile(workDir, exchangecommon.STATUS_FAILED_JOB, errMessage); err != nil {
 						glog.Errorf(fmt.Sprintf("Failed to set error message (%v) for nmp: %v in the status file, error: %v", errMessage, name, err))
 					}

common/node_management.go

@@ -59,6 +59,11 @@ func SetNodeManagementPolicyStatus(db *bolt.DB, pDevice *persistence.ExchangeDev
 			if dbStatus.AgentUpgradeInternal != nil {
 				dbStatus.AgentUpgradeInternal.DownloadAttempts = 0
 			}
+
+			if dbStatus.AgentUpgrade != nil {
+				dbStatus.AgentUpgrade.ActualStartTime = ""
+				dbStatus.AgentUpgrade.CompletionTime = ""
+			}
 		}
 
 		// Update the NMP status in the local db
@@ -95,12 +100,19 @@ func SetNodeManagementPolicyStatus(db *bolt.DB, pDevice *persistence.ExchangeDev
 					if newCertVer != cert_version {
 						updateExch = true
 					}
+				} else {
+					// newCertVer somehow is empty, use the cert version in the exchange
+					newCertVer = cert_version
 				}
+
 				if newConfigVer != "" {
 					pDevice.SetConfigVersion(db, pDevice.Id, newConfigVer)
 					if newConfigVer != config_version {
 						updateExch = true
 					}
+				} else {
+					// newConfigVer somehow is empty, use the config version in the exchange
+					newConfigVer = config_version
 				}
 
 				// Update the agent software version

download/download_worker.go

@@ -220,6 +220,9 @@ func (w *DownloadWorker) DownloadAgentUpgradePackages(org string, filePath strin
 	swType, configType, certType := getUpgradeCSSType(upgradeVersions)
 	glog.V(3).Infof(dwlog(fmt.Sprintf("Upgrade versions: swType: %v, configType: %v, certType: %v", swType, configType, certType)))
 
+	configVersionToSave := upgradeVersions.ConfigVersion
+	certVersionToSave := upgradeVersions.CertVersion
+
 	missingPkgs := []string{}
 	if swType != "" {
 		if objIds != nil {
@@ -264,6 +267,9 @@ func (w *DownloadWorker) DownloadAgentUpgradePackages(org string, filePath strin
 			glog.Errorf(dwlog(fmt.Sprintf("No config upgrade object found of expected type %v found in manifest list.", HZN_CONFIG_FILE)))
 			missingPkgs = append(missingPkgs, HZN_CONFIG_FILE)
 		}
+	} else {
+		// no need to update config, then will keep current config version
+		configVersionToSave = dev.SoftwareVersions[persistence.CONFIG_VERSION]
 	}
 
 	if certType != "" {
@@ -283,13 +289,16 @@ func (w *DownloadWorker) DownloadAgentUpgradePackages(org string, filePath strin
 			glog.Errorf(dwlog(fmt.Sprintf("No cert upgrade object found of expected type %v found in manifest list.", HZN_CERT_FILE)))
 			missingPkgs = append(missingPkgs, HZN_CERT_FILE)
 		}
+	} else {
+		// no need to update cert, then will keep current cert version
+		certVersionToSave = dev.SoftwareVersions[persistence.CERT_VERSION]
 	}
 
 	latestVersions := checkForLatestKeywords(manifest)
 
 	// Return the software version regardless of whether or not it was upgraded as this version is set in the software
 	// The config and cert versions should be the actual version downloaded so after the upgrade is executed, these versions can be used to set the device versions
-	versionsToSave := exchangecommon.AgentUpgradeVersions{SoftwareVersion: manifestUpgradeVersions.SoftwareVersion, ConfigVersion: upgradeVersions.ConfigVersion, CertVersion: upgradeVersions.CertVersion}
+	versionsToSave := exchangecommon.AgentUpgradeVersions{SoftwareVersion: manifestUpgradeVersions.SoftwareVersion, ConfigVersion: configVersionToSave, CertVersion: certVersionToSave}
 
 	if swType == "" && configType == "" && certType == "" {
 		w.Messages() <- events.NewNMPDownloadCompleteMessage(events.NMP_DOWNLOAD_COMPLETE, exchangecommon.STATUS_NO_ACTION, "", nmpName, &versionsToSave, latestVersions)
@@ -662,3 +671,4 @@ func insertVersionToCert(filePath string, nmpName string, version string) error
 func dwlog(input string) string {
 	return fmt.Sprintf("Download worker: %v", input)
 }
+

exchange/cache.go

@@ -354,7 +354,13 @@ func UpdateCacheNodePutWriteThru(nodeOrg string, nodeId string, cachedDevice *De
 	pdr.UserInput = nil
 	pdr.SoftwareVersions = SoftwareVersion{}
 
-	if !reflect.DeepEqual(*pdr, PutDeviceRequest{}) {
+	// Setting this line to prevent unexpected unequal from !reflect.DeepEqual(*pdr, emptyPdr).
+	// type of pdr.SoftwareVersions is SoftwareVersion{}, type of PatchDeviceRequest{} is SoftwareVersion(nil) without setting this line
+	// For debugging, use %#v instead of %v in the log
+	emptyPdr := PutDeviceRequest{
+		SoftwareVersions: SoftwareVersion{},
+	}
+	if !reflect.DeepEqual(*pdr, emptyPdr) {
 		// If you see this error, most likely a new field has been added to the PutDeviceRequest struct and this function needs to be updated to accomadate it
 		glog.Errorf("Warning: Failed to completely update the cached device %s/%s. Changed fields present in the put request were not applied to the cached device. Dropping cache and continuing.", nodeOrg, nodeId)
 		DeleteCacheResource(NODE_DEF_TYPE_CACHE, NodeCacheMapKey(nodeOrg, nodeId))
@@ -381,6 +387,14 @@ func UpdateCacheNodePatchWriteThru(nodeOrg string, nodeId string, cachedDevice *
 			cachedDevice.Arch = *pdr.Arch
 			pdr.Arch = nil
 		}
+		if pdr.ClusterNamespace != nil && *pdr.ClusterNamespace != "" {
+			cachedDevice.ClusterNamespace = *pdr.ClusterNamespace
+			pdr.ClusterNamespace = nil
+		}
+		if pdr.IsNamespaceScoped != nil {
+			cachedDevice.IsNamespaceScoped = *pdr.IsNamespaceScoped
+			pdr.IsNamespaceScoped = nil
+		}
 		if pdr.RegisteredServices != nil {
 			cachedDevice.RegisteredServices = *pdr.RegisteredServices
 			pdr.RegisteredServices = nil
@@ -390,7 +404,14 @@ func UpdateCacheNodePatchWriteThru(nodeOrg string, nodeId string, cachedDevice *
 			pdr.SoftwareVersions = SoftwareVersion{}
 		}
 	}
-	if !reflect.DeepEqual(*pdr, PatchDeviceRequest{}) {
+
+	// Setting this line to prevent unexpected unequal from !reflect.DeepEqual(*pdr, emptyPdr).
+	// type of pdr.SoftwareVersions is SoftwareVersion{}, type of PatchDeviceRequest{} is SoftwareVersion(nil) without setting this line
+	// For debugging, use %#v instead of %v in the log
+	emptyPdr := PatchDeviceRequest{
+		SoftwareVersions: SoftwareVersion{},
+	}
+	if !reflect.DeepEqual(*pdr, emptyPdr) {
 		// If you see this error, most likely a new field has been added to the PatchDeviceRequest struct and this function needs to be updated to accomadate it
 		glog.Errorf("Warning: Failed to completely update the cached device %s/%s. Changed fields present in the patch request were not applied to the cached device. Dropping cache and continuing.", nodeOrg, nodeId)
 		DeleteCacheResource(NODE_DEF_TYPE_CACHE, NodeCacheMapKey(nodeOrg, nodeId))

exchange/node_management_policy.go

@@ -57,7 +57,7 @@ func GetAllExchangeNodeManagementPolicy(ec ExchangeContext, policyOrg string) (*
 
 // Get lists of the agent file versions currently availible in the css
 func GetNodeUpgradeVersions(ec ExchangeContext) (*exchangecommon.AgentFileVersions, error) {
-	glog.V(3).Infof("Getting the availible versions for agent upgrade packages from the exchange.")
+	glog.V(3).Infof("Getting the available versions for agent upgrade packages from the exchange.")
 
 	var resp interface{}
 	resp = new(exchangecommon.AgentFileVersions)

nodemanagement/agent_upgrade.go

@@ -328,7 +328,7 @@ func (n *NodeManagementWorker) HandleAgentFilesVersionChange(cmd *AgentFileVersi
 				}
 			}
 
-			// set the status to waiting for this nmp
+			// set the status to waiting, clear out the SetActualStartTime and CompletionTime for this nmp
 			if setStatusToWaiting {
 				glog.V(3).Infof(nmwlog(fmt.Sprintf("Change status to \"waiting\" for the nmp %v", statusName)))
 
@@ -344,6 +344,8 @@ func (n *NodeManagementWorker) HandleAgentFilesVersionChange(cmd *AgentFileVersi
 				}
 
 				status.AgentUpgrade.Status = exchangecommon.STATUS_NEW
+				status.SetActualStartTime("")
+				status.SetCompletionTime("")
 				err = n.UpdateStatus(statusName, status, exchange.GetPutNodeManagementPolicyStatusHandler(n), persistence.NewMessageMeta(EL_NMP_STATUS_CHANGED, statusName, exchangecommon.STATUS_NEW), persistence.EC_NMP_STATUS_UPDATE_NEW)
 				if err != nil {
 					glog.Errorf(nmwlog(fmt.Sprintf("Error changing nmp status for %v to \"waiting\". Error was %v.", statusName, err)))
@@ -385,6 +387,8 @@ func (w *NodeManagementWorker) HandleNmpStatusReset() {
 					glog.V(3).Infof(nmwlog(fmt.Sprintf("Change status from \"reset\" to \"waiting\" for the nmp %v", nmp_name)))
 
 					local_status.AgentUpgrade.Status = exchangecommon.STATUS_NEW
+					local_status.SetActualStartTime("")
+					local_status.SetCompletionTime("")
 					if local_status.AgentUpgradeInternal != nil {
 						local_status.AgentUpgradeInternal.DownloadAttempts = 0
 					}