chore: health fixes

pkg/health/health.go

@@ -110,15 +110,21 @@ func IsWorse(current, new HealthStatusCode) bool {
 	return newIndex > currentIndex
 }
 
-func GetHealthByConfigType(configType string, obj map[string]any, states ...string) HealthStatus {
-	switch configType {
-	case "AWS::ECS::Task":
-		return GetECSTaskHealth(obj)
-	}
+func get(obj map[string]any, keys ...string) string {
+	v, _, _ := unstructured.NestedString(obj, keys...)
+	return strings.TrimSpace(v)
+}
 
+func isArgoHealth(s HealthStatusCode) bool {
+	return s == "Suspended" || s == "Degraded" || s == "Progressing"
+}
+
+func GetHealthByConfigType(configType string, obj map[string]any, states ...string) HealthStatus {
 	configClass := strings.Split(configType, "::")[0]
 
 	switch strings.ToLower(configClass) {
+	case "aws":
+		return getAWSHealthByConfigType(configType, obj, states...)
 	case "mongo":
 		return GetMongoHealth(obj)
 	case "kubernetes", "crossplane", "missioncontrol", "flux", "argo":
@@ -172,37 +178,56 @@ func GetResourceHealth(
 				Status:  HealthStatusUnknown,
 				Message: err.Error(),
 			}
-		} else {
-			return health, nil
 		}
 	}
 
-	if healthOverride != nil {
-		health, err := healthOverride.GetResourceHealth(obj)
+	if health == nil && healthOverride != nil {
+		health, err = healthOverride.GetResourceHealth(obj)
 		if err != nil {
-			health = &HealthStatus{
+			return &HealthStatus{
 				Status:  HealthStatusUnknown,
 				Message: err.Error(),
-			}
-			return health, err
-		}
-		if health != nil {
-			return health, nil
+			}, err
 		}
 	}
 
-	if obj.GetDeletionTimestamp() != nil {
-		return &HealthStatus{
-			Status: HealthStatusTerminating,
-		}, nil
+	if health == nil ||
+		health.Status == "" ||
+		isArgoHealth(health.Status) {
+		// try and get a better status from conditions
+		defaultHealth, err := GetDefaultHealth(obj)
+		if err != nil {
+			return &HealthStatus{
+				Status:  "HealthParseError",
+				Message: lo.Elipse(err.Error(), 500),
+			}, nil
+		}
+		if health == nil {
+			health = defaultHealth
+		}
+		if health.Status == "" {
+			health.Status = defaultHealth.Status
+		}
+
+		if defaultHealth.Status != "" && isArgoHealth(health.Status) && !isArgoHealth(defaultHealth.Status) {
+			health.Status = defaultHealth.Status
+		}
+		if health.Message == "" {
+			health.Message = defaultHealth.Message
+		}
 	}
 
 	if health == nil {
-		return &HealthStatus{
+		health = &HealthStatus{
 			Status: HealthStatusUnknown,
 			Ready:  true,
-		}, nil
+		}
 	}
+	if obj.GetDeletionTimestamp() != nil {
+		health.Status = HealthStatusTerminating
+		health.Ready = false
+	}
+
 	return health, err
 }
 
@@ -274,5 +299,5 @@ func GetHealthCheckFunc(gvk schema.GroupVersionKind) func(obj *unstructured.Unst
 			return getHPAHealth
 		}
 	}
-	return GetDefaultHealth
+	return nil
 }

pkg/health/health_aws.go

@@ -1,5 +1,50 @@
 package health
 
+import (
+	"fmt"
+	"strings"
+)
+
 func GetAWSResourceHealth(_, status string) (health HealthStatus) {
 	return GetHealthFromStatusName(status)
 }
+
+func getAWSHealthByConfigType(configType string, obj map[string]any, states ...string) HealthStatus {
+	switch configType {
+	case "AWS::ECS::Task":
+		return GetECSTaskHealth(obj)
+	case "AWS::Cloudformation::Stack":
+		return GetHealthFromStatusName(get(obj, "StackStatus"), get(obj, "StackStatusReason"))
+	case "AWS::EC2::Instance":
+		return GetHealthFromStatusName(get(obj, "State"))
+	case "AWS::RDS::DBInstance":
+		return GetHealthFromStatusName(get(obj, "DBInstanceStatus"))
+	case "AWS::ElasticLoadBalancing::LoadBalancer":
+		return GetHealthFromStatusName(get(obj, "State", "Code"))
+	case "AWS::AutoScaling::AutoScalingGroup":
+		return GetHealthFromStatusName(get(obj, "Status"))
+	case "AWS::Lambda::Function":
+		return GetHealthFromStatusName(get(obj, "State"), get(obj, "StateReasonCode"))
+	case "AWS::DynamoDB::Table":
+		return GetHealthFromStatusName(get(obj, "TableStatus"))
+	case "AWS::ElastiCache::CacheCluster":
+		return GetHealthFromStatusName(get(obj, "CacheClusterStatus"))
+	}
+
+	if len(states) > 0 {
+		return GetHealthFromStatusName(states[0])
+	} else {
+		for k, v := range obj {
+			_k := strings.ToLower(k)
+			_v := fmt.Sprintf("%s", v)
+			if _k == "status" || _k == "state" ||
+				strings.HasSuffix(_k, "status") {
+				return GetHealthFromStatusName(_v)
+			}
+		}
+	}
+	return HealthStatus{
+		Health: HealthUnknown,
+	}
+
+}

pkg/health/health_aws_ecs.go

@@ -14,14 +14,14 @@ func GetECSTaskHealth(obj map[string]any) (health HealthStatus) {
 	}
 
 	if v, ok := obj["HealthStatus"].(string); ok {
-		hr.Health = Health(lo.CamelCase(v))
+		hr.Status = HealthStatusCode(HumanCase(v))
 	}
 
-	switch hr.Status {
+	switch strings.ToUpper(string(hr.Status)) {
 	case "RUNNING":
 		hr.Health = HealthHealthy
 		hr.Ready = true
-	case "STOPPED", "DELETED":
+	case "STOPPED", "DELETED", "UNKNOWN":
 		hr.Ready = true
 		hr.Health = HealthUnknown
 	}
@@ -31,15 +31,16 @@ func GetECSTaskHealth(obj map[string]any) (health HealthStatus) {
 	if stopCode != "" {
 		hr.Status = HealthStatusCode(stopCode)
 	}
-	switch stopCode {
-	case "TaskFailedToStart":
+
+	switch strings.ToUpper(stopCode) {
+	case "TASKFAILEDTOSTART":
 		hr.Health = HealthUnhealthy
-	case "EssentialContainerExited":
+	case "ESSENTIALCONTAINEREXITED":
 		hr.Status = HealthStatusCrashed
 		hr.Health = HealthUnhealthy
-	case "UserInitiated":
+	case "USERINITIATED":
 		hr.Status = HealthStatusStopped
-	case "ServiceSchedulerInitiated":
+	case "SERVICESCHEDULERINITIATED":
 		hr.Status = HealthStatusTerminating
 	}
 
@@ -52,15 +53,15 @@ func GetECSTaskHealth(obj map[string]any) (health HealthStatus) {
 				hr.Message = strings.TrimSpace(reason[idx+1:])
 			}
 
-			switch hr.Status {
-			case "ContainerRuntimeError", "ContainerRuntimeTimeoutError", "OutOfMemoryError":
+			switch strings.ToUpper(string(hr.Status)) {
+			case "CONTAINERRUNTIMEERROR", "CONTAINERRUNTIMETIMEOUTERROR", "OUTOFMEMORYERROR":
 				hr.Health = HealthUnhealthy
-			case "InternalError", "CannotCreateVolumeError", "ResourceNotFoundException", "CannotStartContainerError":
+			case "INTERNALERROR", "CANNOTCREATEVOLUMEERROR", "RESOURCENOTFOUNDERROR", "CANNOTSTARTCONTAINERERROR":
 				hr.Health = HealthUnhealthy
 				hr.Ready = true
-			case "SpotInterruptionError", "CannotStopContainerError", "CannotInspectContainerError":
+			case "SPOTINTERRUPTIONERROR", "CANNOTSTOPCONTAINERERROR", "CANNOTINSPECTCONTAINERERROR":
 				hr.Health = HealthWarning
-			case "TaskFailedToStart", "ResourceInitializationError", "CannotPullContainer":
+			case "TASKFAILEDTOSTART", "RESOURCEINITIALIZATIONERROR", "CANNOTPULLCONTAINER":
 				hr.Health = HealthUnhealthy
 			default:
 				hr.Health = HealthUnhealthy