Bugfix/5373 tm log inconsistency

CHANGELOG.md

@@ -21,6 +21,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - [#6259](https://github.com/apache/trafficcontrol/issues/6259) - Traffic Portal No Longer Allows Spaces in Server Object "Router Port Name"
 - [#6175](https://github.com/apache/trafficcontrol/issues/6175) - POST request to /api/4.0/phys_locations accepts mismatch values for regionName.
 - [#6283](https://github.com/apache/trafficcontrol/issues/6283) - The Traffic Ops Postinstall script will work in CentOS 7, even if Python 3 is installed
+- [#5373](https://github.com/apache/trafficcontrol/issues/5373) - Traffic Monitor logs not consistent
 
 ### Changed
 - Updated `t3c` to request less unnecessary deliveryservice-server assignment and invalidation jobs data via new query params supported by Traffic Ops

traffic_monitor/ds/stat.go

@@ -105,30 +105,6 @@ func addAvailableData(dsStats *dsdata.Stats, crStates tc.CRStates, serverCachegr
 		}
 	}
 
-	for dsName, stat := range dsStats.DeliveryService {
-		lastStat, lastStatExists := lastStats.DeliveryServices[dsName]
-		if !lastStatExists {
-			continue
-		}
-
-		getEvent := func(desc string) health.Event {
-			// TODO sync.Pool?
-			return health.Event{
-				Time:        health.Time(time.Now()),
-				Description: desc,
-				Name:        dsName.String(),
-				Hostname:    dsName.String(),
-				Type:        "Delivery Service",
-				Available:   stat.CommonStats.IsAvailable.Value,
-			}
-		}
-		if stat.CommonStats.IsAvailable.Value == false && lastStat.Available == true {
-			events.Add(getEvent("no available caches"))
-		} else if stat.CommonStats.IsAvailable.Value == true && lastStat.Available == false {
-			events.Add(getEvent("available caches"))
-		}
-	}
-
 	// TODO move to its own func?
 	for dsName := range crStates.DeliveryService {
 		stat, ok := dsStats.DeliveryService[dsName]
@@ -325,14 +301,18 @@ func addDSPerSecStats(lastStats *dsdata.LastStats, dsStats *dsdata.Stats, dsName
 			Description: desc,
 			Name:        dsName.String(),
 			Hostname:    dsName.String(),
-			Type:        "DELIVERYSERVICE",
+			Type:        health.DeliveryServiceEventType,
 			Available:   stat.CommonStats.IsAvailable.Value,
 		}
 	}
-	if stat.CommonStats.IsAvailable.Value == false && lastStat.Available == true && dsErr != nil {
-		events.Add(getEvent(dsErr.Error())) // TODO change events.Add to not allocate new memory, after the limit is reached.
+	if stat.CommonStats.IsAvailable.Value == false && lastStat.Available == true {
+		eventDesc := "Unavailable"
+		if dsErr != nil {
+			eventDesc = eventDesc + " err: " + dsErr.Error()
+		}
+		events.Add(getEvent(eventDesc)) // TODO change events.Add to not allocate new memory, after the limit is reached.
 	} else if stat.CommonStats.IsAvailable.Value == true && lastStat.Available == false {
-		events.Add(getEvent("REPORTED - available"))
+		events.Add(getEvent("Available caches"))
 	}
 
 	lastStat.Available = stat.CommonStats.IsAvailable.Value

traffic_monitor/health/event.go

@@ -29,6 +29,10 @@ import (
 	"github.com/apache/trafficcontrol/lib/go-log"
 )
 
+const (
+	DeliveryServiceEventType = "DELIVERYSERVICE"
+)
+
 type Time time.Time
 
 func (t Time) MarshalJSON() ([]byte, error) {

traffic_monitor/manager/peer.go

@@ -52,6 +52,13 @@ func comparePeerState(events health.ThreadsafeEvents, result peer.Result, peerSt
 			description = "Peer is unreachable"
 		}
 
-		events.Add(health.Event{Time: health.Time(result.Time), Description: description, Name: result.ID.String(), Hostname: result.ID.String(), Type: "PEER", Available: result.Available})
+		events.Add(
+			health.Event{
+				Time:        health.Time(result.Time),
+				Description: description,
+				Name:        result.ID.String(),
+				Hostname:    result.ID.String(),
+				Type:        "PEER",
+				Available:   result.Available})
 	}
 }

traffic_monitor/manager/statecombiner.go

@@ -131,7 +131,16 @@ func combineCacheState(
 	}
 
 	if overrideCondition != "" {
-		events.Add(health.Event{Time: health.Time(time.Now()), Description: fmt.Sprintf("Health protocol override condition %s", overrideCondition), Name: cacheName.String(), Hostname: cacheName.String(), Type: toData.ServerTypes[cacheName].String(), Available: available, IPv4Available: ipv4Available, IPv6Available: ipv6Available})
+		events.Add(
+			health.Event{
+				Time:          health.Time(time.Now()),
+				Description:   fmt.Sprintf("Health protocol override condition %s", overrideCondition),
+				Name:          cacheName.String(),
+				Hostname:      cacheName.String(),
+				Type:          toData.ServerTypes[cacheName].String(),
+				Available:     available,
+				IPv4Available: ipv4Available,
+				IPv6Available: ipv6Available})
 	}
 
 	combinedStates.AddCache(cacheName, tc.IsAvailable{IsAvailable: available, Ipv4Available: ipv4Available, Ipv6Available: ipv6Available})