feat: update health probe to make use of healthChecker

healthChecker was introduced in open-cluster-management-io/addon-framework#289
and it allow us to define a list of ProbeFields as before but it also
allows us to define HealthChecker. HealthChecker has the ability to go
through all the resultFields and decide which resources are mandatory
and should trigger
unheathly probes and which resources are optional

Author: JoaoBraveCoding

go.mod

@@ -17,11 +17,13 @@ require (
 	k8s.io/client-go v0.30.3
 	k8s.io/component-base v0.30.3
 	k8s.io/klog/v2 v2.130.1
-	open-cluster-management.io/addon-framework v0.11.0
-	open-cluster-management.io/api v0.15.0
+	open-cluster-management.io/addon-framework v0.0.0-00010101000000-000000000000
+	open-cluster-management.io/api v0.15.1-0.20241120090202-cb7ce98ab874
 	sigs.k8s.io/controller-runtime v0.18.5
 )
 
+replace open-cluster-management.io/addon-framework => github.com/zhiweiyin318/addon-framework v0.0.0-20241125074036-754aecccaf2b
+
 replace github.com/prometheus/common => github.com/prometheus/common v0.60.1
 
 replace github.com/prometheus/client_golang => github.com/prometheus/client_golang v1.20.5

go.sum

@@ -219,6 +219,8 @@ github.com/xiang90/probing v0.0.0-20190116061207-43a291ad63a2/go.mod h1:UETIi67q
 github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/yuin/goldmark v1.4.13/go.mod h1:6yULJ656Px+3vBD8DxQVa3kxgyrAnzto9xy5taEt/CY=
+github.com/zhiweiyin318/addon-framework v0.0.0-20241125074036-754aecccaf2b h1:xBnQWUqASEhZfyUZ8Bsl7fBe78ZF+Y2od524n6dZfCA=
+github.com/zhiweiyin318/addon-framework v0.0.0-20241125074036-754aecccaf2b/go.mod h1:tsBSNs9mGfVQQjXBnjgpiX6r0UM+G3iNfmzQgKhEfw4=
 go.etcd.io/bbolt v1.3.8 h1:xs88BrvEv273UsB79e0hcVrlUWmS0a8upikMFhSyAtA=
 go.etcd.io/bbolt v1.3.8/go.mod h1:N9Mkw9X8x5fupy0IKsmuqVtoGDyxsaDlbk4Rd05IAQw=
 go.etcd.io/etcd/api/v3 v3.5.12 h1:W4sw5ZoU2Juc9gBWuLk5U6fHfNVyY1WC5g9uiXZio/c=
@@ -378,10 +380,8 @@ k8s.io/kube-openapi v0.0.0-20240620174524-b456828f718b h1:Q9xmGWBvOGd8UJyccgpYlL
 k8s.io/kube-openapi v0.0.0-20240620174524-b456828f718b/go.mod h1:UxDHUPsUwTOOxSU+oXURfFBcAS6JwiRXTYqYwfuGowc=
 k8s.io/utils v0.0.0-20240502163921-fe8a2dddb1d0 h1:jgGTlFYnhF1PM1Ax/lAlxUPE+KfCIXHaathvJg1C3ak=
 k8s.io/utils v0.0.0-20240502163921-fe8a2dddb1d0/go.mod h1:OLgZIPagt7ERELqWJFomSt595RzquPNLL48iOWgYOg0=
-open-cluster-management.io/addon-framework v0.11.0 h1:ZJxphgHQ36VUJF0RIag+nzcEn5PNyep2rsEPdz6wT7o=
-open-cluster-management.io/addon-framework v0.11.0/go.mod h1:ruMU8i/dciz3qCv2CQ46Cu1b7rkK7TpvB+W4bRwHf+I=
-open-cluster-management.io/api v0.15.0 h1:lRee1KOlGHZb2scTA7ff9E9Fxt2hJc7jpkHnaCbvkOU=
-open-cluster-management.io/api v0.15.0/go.mod h1:9erZEWEn4bEqh0nIX2wA7f/s3KCuFycQdBrPrRzi0QM=
+open-cluster-management.io/api v0.15.1-0.20241120090202-cb7ce98ab874 h1:WgkuYXTbJV7EK+qtiMq3soa21faGUKeTG5w0C8Mn1Ok=
+open-cluster-management.io/api v0.15.1-0.20241120090202-cb7ce98ab874/go.mod h1:9erZEWEn4bEqh0nIX2wA7f/s3KCuFycQdBrPrRzi0QM=
 open-cluster-management.io/sdk-go v0.15.0 h1:2IAJnPfUoY6rPC5w7LhqAnvIlgekPoVW03LdZO1unIM=
 open-cluster-management.io/sdk-go v0.15.0/go.mod h1:fi5WBsbC5K3txKb8eRLuP0Sim/Oqz/PHX18skAEyjiA=
 sigs.k8s.io/apiserver-network-proxy/konnectivity-client v0.29.0 h1:/U5vjBbQn3RChhv7P11uhYvCSm5G2GaIi5AIGBS6r4c=

internal/addon/addon.go

@@ -14,9 +14,11 @@ import (
 )
 
 var (
+	errMissingFields              = errors.New("no fields found in health checker")
 	errProbeConditionNotSatisfied = errors.New("probe condition is not satisfied")
 	errProbeValueIsNil            = errors.New("probe value is nil")
-	errValueNotProbed             = errors.New("value not probed")
+	errUnknownProbeKey            = errors.New("probe has key that doesn't match the key defined")
+	errUnknownResource            = errors.New("undefined health check for resource")
 )
 
 func NewRegistrationOption(agentName string) *agent.RegistrationOption {
@@ -90,42 +92,53 @@ func AgentHealthProber() *agent.HealthProber {
 					},
 				},
 			},
-			HealthCheck: func(identifier workv1.ResourceIdentifier, result workv1.StatusFeedbackResult) error {
-				for _, value := range result.Values {
+			HealthChecker: func(fields []agent.ResultField) error {
+				if len(fields) == 0 {
+					return errMissingFields
+				}
+				for _, field := range fields {
+					if len(field.FeedbackResult.Values) == 0 {
+						// If a probe didn't get any values maybe the resources was not deployed
+						continue
+					}
+					identifier := field.ResourceIdentifier
 					switch {
 					case identifier.Resource == ClusterLogForwardersResource:
-						if value.Name != clfProbeKey {
-							continue
-						}
+						for _, value := range field.FeedbackResult.Values {
+							if value.Name != clfProbeKey {
+								return fmt.Errorf("%w: %s with key %s/%s unknown probe keys %s", errUnknownProbeKey, identifier.Resource, identifier.Namespace, identifier.Name, value.Name)
+							}
 
-						if value.Value.String == nil {
-							return fmt.Errorf("%w: clusterlogforwarder with key %s/%s", errProbeValueIsNil, identifier.Namespace, identifier.Name)
-						}
+							if value.Value.String == nil {
+								return fmt.Errorf("%w: %s with key %s/%s", errProbeValueIsNil, identifier.Resource, identifier.Namespace, identifier.Name)
+							}
 
-						if *value.Value.String != "True" {
-							return fmt.Errorf("%w: clusterlogforwarder status condition type is %s for %s/%s", errProbeConditionNotSatisfied, *value.Value.String, identifier.Namespace, identifier.Name)
+							if *value.Value.String != "True" {
+								return fmt.Errorf("%w: %s status condition type is %s for %s/%s", errProbeConditionNotSatisfied, identifier.Resource, *value.Value.String, identifier.Namespace, identifier.Name)
+							}
+							// everything checks we should skip to the next field
 						}
-
-						return nil
 					case identifier.Resource == OpenTelemetryCollectorsResource:
-						if value.Name != otelColProbeKey {
-							continue
-						}
+						for _, value := range field.FeedbackResult.Values {
+							if value.Name != otelColProbeKey {
+								return fmt.Errorf("%w: %s with key %s/%s unknown probe keys %s", errUnknownProbeKey, identifier.Resource, identifier.Namespace, identifier.Name, value.Name)
+							}
 
-						if value.Value.Integer == nil {
-							return fmt.Errorf("%w: opentelemetrycollector with key %s/%s", errProbeValueIsNil, identifier.Namespace, identifier.Name)
-						}
+							if value.Value.Integer == nil {
+								return fmt.Errorf("%w: %s with key %s/%s", errProbeValueIsNil, identifier.Resource, identifier.Namespace, identifier.Name)
+							}
 
-						if *value.Value.Integer < 1 {
-							return fmt.Errorf("%w: opentelemetrycollector replicas is %d for %s/%s", errProbeConditionNotSatisfied, *value.Value.Integer, identifier.Namespace, identifier.Name)
+							if *value.Value.Integer < 1 {
+								return fmt.Errorf("%w: %s replicas is %d for %s/%s", errProbeConditionNotSatisfied, identifier.Resource, *value.Value.Integer, identifier.Namespace, identifier.Name)
+							}
+							// everything checks we should skip to the next field
 						}
-
-						return nil
 					default:
-						continue
+						// If a resource is being probed it should have a health check defined
+						return fmt.Errorf("%w: %s with key %s/%s", errUnknownResource, identifier.Resource, identifier.Namespace, identifier.Name)
 					}
 				}
-				return fmt.Errorf("%w: for resource %s with key %s/%s", errValueNotProbed, identifier.Resource, identifier.Namespace, identifier.Name)
+				return nil
 			},
 		},
 	}