fix recreation of metrics on edit or delete operation on ptpconfig

Signed-off-by: Aneesh Puttur <aputtur@redhat.com>

plugins/ptp_operator/config/config.go

@@ -130,16 +130,26 @@ func (pt *PtpClockThreshold) SafeClose() (justClosed bool) {
 
 // LinuxPTPConfigMapUpdate for ptp-conf update
 type LinuxPTPConfigMapUpdate struct {
-	lock                   sync.RWMutex
-	UpdateCh               chan bool
-	NodeProfiles           []PtpProfile
-	appliedNodeProfileJSON []byte
-	profilePath            string
-	intervalUpdate         int
-	EventThreshold         map[string]*PtpClockThreshold
-	PtpProcessOpts         map[string]*PtpProcessOpts
-	PtpSettings            map[string]map[string]string
-	DeleteConfigInProcess  bool
+	lock                        sync.RWMutex
+	UpdateCh                    chan bool
+	NodeProfiles                []PtpProfile
+	appliedNodeProfileJSON      []byte
+	profilePath                 string
+	intervalUpdate              int
+	EventThreshold              map[string]*PtpClockThreshold
+	PtpProcessOpts              map[string]*PtpProcessOpts
+	PtpSettings                 map[string]map[string]string
+	fileWatcherUpdateInProgress bool
+}
+
+// AppliedNodeProfileJSON ....
+func (l *LinuxPTPConfigMapUpdate) AppliedNodeProfileJSON() []byte {
+	return l.appliedNodeProfileJSON
+}
+
+// SetAppliedNodeProfileJSON ...
+func (l *LinuxPTPConfigMapUpdate) SetAppliedNodeProfileJSON(appliedNodeProfileJSON []byte) {
+	l.appliedNodeProfileJSON = appliedNodeProfileJSON
 }
 
 // NewLinuxPTPConfUpdate -- profile updater
@@ -206,13 +216,18 @@ func (l *LinuxPTPConfigMapUpdate) DeleteAllPTPThreshold() {
 	}
 }
 
-// UpdateDeleteConfigInProcess ... if config file is deleted delay config update reads
-func (l *LinuxPTPConfigMapUpdate) UpdateDeleteConfigInProcess(delete bool) {
+// FileWatcherUpdateInProgress ... if config file watcher delay config update reads
+func (l *LinuxPTPConfigMapUpdate) FileWatcherUpdateInProgress(inProgress bool) {
 	l.lock.Lock()
-	l.DeleteConfigInProcess = delete
+	l.fileWatcherUpdateInProgress = inProgress
 	l.lock.Unlock()
 }
 
+// IsFileWatcherUpdateInProgress ... if config file watcher delay config update reads
+func (l *LinuxPTPConfigMapUpdate) IsFileWatcherUpdateInProgress() bool {
+	return l.fileWatcherUpdateInProgress
+}
+
 func closeHoldover(t *PtpClockThreshold) {
 	defer func() {
 		if err := recover(); err != nil {
@@ -355,16 +370,22 @@ func (l *LinuxPTPConfigMapUpdate) WatchConfigMapUpdate(nodeName string, closeCh
 		select {
 		case <-tickerPull.C:
 			l.updatePtpConfig(nodeName)
+			return // only do it once until ptpconfig is found
 		case <-closeCh:
 			log.Info("signal received, shutting down")
 			return
 		}
 	}
 }
 
+// PushPtpConfigMapChanges  ... push configMap updates
+func (l *LinuxPTPConfigMapUpdate) PushPtpConfigMapChanges(nodeName string) {
+	l.updatePtpConfig(nodeName)
+}
+
 func (l *LinuxPTPConfigMapUpdate) updatePtpConfig(nodeName string) {
 	nodeProfile := filepath.Join(l.profilePath, nodeName)
-	if l.DeleteConfigInProcess {
+	if l.IsFileWatcherUpdateInProgress() {
 		log.Infof("delete action on config was detetcted; delaying updating ptpconfig")
 		return // wait until delete is completed
 	}

plugins/ptp_operator/metrics/metrics.go

@@ -111,6 +111,7 @@ func (p *PTPEventManager) ExtractMetrics(msg string) {
 	profileName := ptp4lCfg.Profile
 
 	if !p.validLogToProcess(profileName, processName, len(ptp4lCfg.Interfaces)) {
+		log.Infof("skipped parsing %s", output)
 		return
 	}
 
@@ -335,7 +336,7 @@ func (p *PTPEventManager) processDownEvent(profileName, processName string, ptpS
 
 func (p *PTPEventManager) validLogToProcess(profileName, processName string, iFaceSize int) bool {
 	if profileName == "" {
-		log.Errorf("ptp4l config does not have profile name, aborting. ")
+		log.Info("ptp4l config does not have profile name, skipping. ")
 		return false
 	}
 	if !p.hasHAProfile(profileName) && profileName != "" && iFaceSize == 0 { //TODO: Use PMC to update port and roles

plugins/ptp_operator/metrics/registry.go

@@ -181,7 +181,7 @@ func DeletedPTPMetrics(clockType, processName, eventResourceName string) {
 		"process": processName, "node": ptpNodeName, "iface": eventResourceName})
 }
 
-// DeleteThresholdMetrics .. delete threshold metrics
+// DeleteThresholdMetrics ... delete threshold metrics
 func DeleteThresholdMetrics(profile string) {
 	Threshold.Delete(prometheus.Labels{
 		"threshold": "MinOffsetThreshold", "node": ptpNodeName, "profile": profile})

plugins/ptp_operator/ptp_operator_plugin.go

@@ -22,6 +22,7 @@ import (
 	"os"
 	"strings"
 	"sync"
+	"time"
 
 	"k8s.io/utils/pointer"
 
@@ -105,7 +106,6 @@ func Start(wg *sync.WaitGroup, configuration *common.SCConfiguration, fn func(e
 	go listenToSocket(wg)
 	// watch for ptp any config updates
 	go eventManager.PtpConfigMapUpdates.WatchConfigMapUpdate(nodeName, configuration.CloseCh)
-
 	// watch and monitor ptp4l config file creation and deletion
 	// create a ptpConfigDir directory watcher for config changes
 	notifyConfigDirUpdates = make(chan *ptp4lconf.PtpConfigUpdate, 20) // max 20 profiles since all files will be deleted in reapply
@@ -114,6 +114,7 @@ func Start(wg *sync.WaitGroup, configuration *common.SCConfiguration, fn func(e
 		log.Errorf("cannot monitor ptp4l configuation folder at %s : %s", ptpConfigDir, err)
 		return err
 	}
+
 	// process ptp4l conf file updates under /var/run/ptp4l.X.conf
 	go processPtp4lConfigFileUpdates()
 
@@ -271,9 +272,19 @@ func getCurrentStatOverrideFn() func(e v2.Event, d *channel.DataChan) error {
 	}
 }
 
-// update interface and threshold details when ptpConfig change found
-// these are updated when config file is created or deleted under /var/run folder
+// update interface details  and threshold details when ptpConfig change found.
+// These are updated when config file is created or deleted under /var/run folder
+// by  linuxptp-daemon.
+// issue to resolve: linuxptp-daemon-container will read ptpconfig and update ptpConfigMap
+// the updates are incremental and file creation happens when  ptpConfigMap is updated
+// NOTE: there is delay between configmap updates and file creation
+// FIRST ptpConfigMap created and then the daemon ticker when starting the process
+// creates the files or deletes the  file when process is stopped
+// This routine tries to rely on file creation to detect ptp config updates
+// and on file creation call ptpConfig updates
+// on file deletion reset to read fresh ptpconfig updates
 func processPtp4lConfigFileUpdates() {
+	fileModified := false
 	for {
 		select {
 		case ptpConfigEvent := <-notifyConfigDirUpdates:
@@ -362,13 +373,24 @@ func processPtp4lConfigFileUpdates() {
 							strings.Replace(cName, ptp4lProcessName, ts2PhcProcessName, 1), ts2PhcProcessName)
 					}
 				}
+				for key, np := range eventManager.PtpConfigMapUpdates.EventThreshold {
+					ptpMetrics.Threshold.With(prometheus.Labels{
+						"threshold": "MinOffsetThreshold", "node": eventManager.NodeName(), "profile": key}).Set(float64(np.MinOffsetThreshold))
+					ptpMetrics.Threshold.With(prometheus.Labels{
+						"threshold": "MaxOffsetThreshold", "node": eventManager.NodeName(), "profile": key}).Set(float64(np.MaxOffsetThreshold))
+					ptpMetrics.Threshold.With(prometheus.Labels{
+						"threshold": "HoldOverTimeout", "node": eventManager.NodeName(), "profile": key}).Set(float64(np.HoldOverTimeout))
+				}
 			case true: // ptp4l.X.conf is deleted ; how to prevent updates happening at same time
 				// delete metrics, ptp4l config is removed
 				// get config fileName
+				fileModified = true
+				eventManager.PtpConfigMapUpdates.FileWatcherUpdateInProgress(true)
+				log.Infof("delete ptp config %s at %s", *ptpConfigEvent.Name, time.Now())
 				ptpConfigFileName := ptpTypes.ConfigName(*ptpConfigEvent.Name)
 				ptp4lConfig := eventManager.GetPTPConfig(ptpConfigFileName)
 				log.Infof("deleting config %s with profile %s\n", *ptpConfigEvent.Name, ptp4lConfig.Profile)
-				eventManager.PtpConfigMapUpdates.UpdateDeleteConfigInProcess(true)
+
 				ptpStats := eventManager.GetStats(ptpConfigFileName)
 				// to avoid new one getting created , make a copy of this stats
 				ptpStats.SetConfigAsDeleted(true)
@@ -430,11 +452,19 @@ func processPtp4lConfigFileUpdates() {
 				// clean up process metrics
 				ptpMetrics.DeleteProcessStatusMetricsForConfig(eventManager.NodeName(), string(ptpConfigFileName), ptp4lProcessName, phc2sysProcessName)
 				ptpMetrics.DeleteProcessStatusMetricsForConfig(eventManager.NodeName(), strings.Replace(string(ptpConfigFileName), ptp4lProcessName, ts2PhcProcessName, 1), ts2PhcProcessName)
-				eventManager.PtpConfigMapUpdates.UpdateDeleteConfigInProcess(false)
 			}
 		case <-config.CloseCh:
 			fileWatcher.Close()
 			return
+		default:
+			// decision maker before ptpConfig map updates can be processes
+			if fileModified {
+				eventManager.PtpConfigMapUpdates.FileWatcherUpdateInProgress(false)
+				eventManager.PtpConfigMapUpdates.SetAppliedNodeProfileJSON([]byte{}) // reset ptpconfig dataset and update again
+				eventManager.PtpConfigMapUpdates.PushPtpConfigMapChanges(eventManager.NodeName())
+				fileModified = false
+			}
+
 		}
 	}
 }