Merge pull request #369 from aneeshkp/add-holdover-event

OCPBUGS-44208: Fix Holdover Event Triggering for T-GM and Minimize Redundant Clock Class Events:
redhat-cne · Nov 7, 2024 · bacea3b · bacea3b
2 parents 8c4d35f + 48c32f6
commit bacea3b
Show file tree

Hide file tree

Showing 3 changed files with 47 additions and 7 deletions.
diff --git a/plugins/ptp_operator/metrics/manager.go b/plugins/ptp_operator/metrics/manager.go
@@ -393,6 +393,8 @@ func (p *PTPEventManager) GenPTPEvent(ptpProfileName string, oStats *stats.Stats
 				oStats.SetLastOffset(ptpOffset)
 			}
 		case ptp.HOLDOVER:
+			//FOR OC/BC  do nothing, the timeout will switch holdover to FREE-RUN OR LOCKED  if it is not within the holdover timeout
+			// FOR T-GM its handled differently
 			// previous state was HOLDOVER, now it is in LOCKED state, cancel any HOLDOVER
 			if isOffsetInRange(ptpOffset, threshold.MaxOffsetThreshold, threshold.MinOffsetThreshold) {
 				log.Infof("interface %s is in LOCKED state, cancel any holdover states", eventResourceName)
@@ -403,7 +405,8 @@ func (p *PTPEventManager) GenPTPEvent(ptpProfileName string, oStats *stats.Stats
 				oStats.SetLastSyncState(clockState)
 				oStats.SetLastOffset(ptpOffset)
 				oStats.AddValue(ptpOffset) // update off set when its in locked state and hold over only/ update off set when its in locked state and hold over only
-			} // else continue to stay in HOLDOVER
+			} // else continue to stay in HOLDOVER UNTIL its really LOCKED state
+
 		default: // not yet used states
 			clockState = ptp.FREERUN
 			if isOffsetInRange(ptpOffset, threshold.MaxOffsetThreshold, threshold.MinOffsetThreshold) {
@@ -418,11 +421,19 @@ func (p *PTPEventManager) GenPTPEvent(ptpProfileName string, oStats *stats.Stats
 			oStats.SetLastOffset(ptpOffset)
 		}
 	case ptp.HOLDOVER:
-		return // do nothing for holdover
+		if lastClockState != ptp.HOLDOVER { //send event only once
+			log.Infof(" publishing event for (profile %s) %s with last state %s and current clock state %s and offset %d )",
+				ptpProfileName, eventResourceName, oStats.LastSyncState(), clockState, ptpOffset)
+			p.PublishEvent(clockState, ptpOffset, eventResourceName, eventType)
+		}
+		oStats.SetLastSyncState(clockState)
+		oStats.SetLastOffset(ptpOffset)
+		oStats.AddValue(ptpOffset)
+		return
 	case ptp.FREERUN:
 		if lastClockState != ptp.HOLDOVER {
 			if lastClockState != ptp.FREERUN { // don't send event if last event was freerun
-				log.Infof(" publishing event for (profile %s) %s with last state %s and current clock state %s and offset %d for ( Max/Min Threshold %d/%d )",
+				log.Infof("publishing event for (profile %s) %s with last state %s and current clock state %s and offset %d for ( Max/Min Threshold %d/%d )",
 					ptpProfileName, eventResourceName, oStats.LastSyncState(), clockState, ptpOffset, threshold.MaxOffsetThreshold, threshold.MinOffsetThreshold)
 				p.PublishEvent(clockState, ptpOffset, eventResourceName, eventType)
 			}

diff --git a/plugins/ptp_operator/metrics/manager_test.go b/plugins/ptp_operator/metrics/manager_test.go
@@ -1,3 +1,6 @@
+//go:build unittests
+// +build unittests
+
 package metrics_test
 
 import (
@@ -65,7 +68,7 @@ func TestPTPEventManager_GenPTPEvent(t *testing.T) {
 			eventResourceName: "resource3",
 			ptpOffset:         500,
 			lastClockState:    ptp.HOLDOVER,
-			clockState:        ptp.FREERUN,
+			clockState:        ptp.FREERUN, // stays in HOLDOVER until times out
 			eventType:         ptp.PtpStateChange,
 			mock:              true,
 			wantLastSyncState: ptp.HOLDOVER,
@@ -82,6 +85,30 @@ func TestPTPEventManager_GenPTPEvent(t *testing.T) {
 			mock:              true,
 			wantLastSyncState: ptp.LOCKED,
 		},
+		{
+			name:              "holdover to locked state",
+			ptpProfileName:    "T-GM",
+			oStats:            stats.NewStats("T-GM"),
+			eventResourceName: "resource3",
+			ptpOffset:         50,
+			lastClockState:    ptp.LOCKED,
+			clockState:        ptp.HOLDOVER,
+			eventType:         ptp.PtpStateChange,
+			mock:              true,
+			wantLastSyncState: ptp.HOLDOVER,
+		},
+		{
+			name:              "holdover to locked state",
+			ptpProfileName:    "T-GM",
+			oStats:            stats.NewStats("T-GM"),
+			eventResourceName: "resource3",
+			ptpOffset:         50,
+			lastClockState:    ptp.HOLDOVER,
+			clockState:        ptp.HOLDOVER,
+			eventType:         ptp.PtpStateChange,
+			mock:              true,
+			wantLastSyncState: ptp.HOLDOVER,
+		},
 	}
 
 	for _, tt := range tests {

diff --git a/plugins/ptp_operator/metrics/ptp4lParse.go b/plugins/ptp_operator/metrics/ptp4lParse.go
@@ -43,11 +43,13 @@ func (p *PTPEventManager) ParsePTP4l(processName, configName, profileName, outpu
 				alias, _ = ptp4lCfg.GetUnknownAlias()
 			}
 			masterResource := fmt.Sprintf("%s/%s", alias, MasterClockType)
-			ptpStats[master].SetClockClass(int64(clockClass))
+
 			ClockClassMetrics.With(prometheus.Labels{
 				"process": processName, "node": ptpNodeName}).Set(clockClass)
-
-			p.PublishClockClassEvent(clockClass, masterResource, ptp.PtpClockClassChange)
+			if ptpStats[master].ClockClass() != int64(clockClass) {
+				ptpStats[master].SetClockClass(int64(clockClass))
+				p.PublishClockClassEvent(clockClass, masterResource, ptp.PtpClockClassChange)
+			}
 		}
 	} else if strings.Contains(output, " port ") {
 		portID, role, syncState := extractPTP4lEventState(output)