refactor: require stored references for CR status

Signed-off-by: Alex Jun <aljun@nvidia.com>

Author: alexscjundev

fault-remediation/pkg/crstatus/checker.go

@@ -23,13 +23,11 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	"github.com/nvidia/nvsentinel/fault-remediation/pkg/annotation"
-	"github.com/nvidia/nvsentinel/fault-remediation/pkg/config"
 )
 
 type CRStatusChecker struct {
-	client             client.Client
-	remediationActions map[string]config.MaintenanceResource
-	dryRun             bool
+	client client.Client
+	dryRun bool
 }
 
 type CRState string
@@ -43,39 +41,14 @@ const (
 
 func NewCRStatusChecker(
 	client client.Client,
-	remediationActions map[string]config.MaintenanceResource,
 	dryRun bool,
 ) *CRStatusChecker {
 	return &CRStatusChecker{
-		client:             client,
-		remediationActions: remediationActions,
-		dryRun:             dryRun,
+		client: client,
+		dryRun: dryRun,
 	}
 }
 
-// ShouldSkipCRCreation returns true if an existing CR should suppress creation of a new CR.
-func (c *CRStatusChecker) ShouldSkipCRCreation(ctx context.Context, actionName string, crName string) bool {
-	state := c.GetCRState(ctx, actionName, crName)
-	return state == CRStateInProgress || state == CRStateSucceeded
-}
-
-func (c *CRStatusChecker) GetCRState(ctx context.Context, actionName string, crName string) CRState {
-	resource, exists := c.remediationActions[actionName]
-	if !exists {
-		slog.ErrorContext(ctx, "No remediation configuration found for action", "action", actionName)
-		return CRStateNotFound
-	}
-
-	resourceRef := annotation.MaintenanceResourceReference{
-		Namespace: resource.Namespace,
-		Version:   resource.Version,
-		ApiGroup:  resource.ApiGroup,
-		Kind:      resource.Kind,
-	}
-
-	return c.GetCRStateForReference(ctx, crName, resourceRef, resource.CompleteConditionType)
-}
-
 func (c *CRStatusChecker) GetCRStateForReference(
 	ctx context.Context,
 	crName string,
@@ -110,10 +83,6 @@ func (c *CRStatusChecker) GetCRStateForReference(
 	return c.checkConditionType(obj, completeConditionType)
 }
 
-func (c *CRStatusChecker) checkCondition(obj *unstructured.Unstructured, resource config.MaintenanceResource) CRState {
-	return c.checkConditionType(obj, resource.CompleteConditionType)
-}
-
 func (c *CRStatusChecker) checkConditionType(obj *unstructured.Unstructured, completeConditionType string) CRState {
 	status, found, err := unstructured.NestedMap(obj.Object, "status")
 	if err != nil || !found {

fault-remediation/pkg/crstatus/crstatus_interface.go

@@ -23,7 +23,5 @@ import (
 )
 
 type CRStatusCheckerInterface interface {
-	ShouldSkipCRCreation(context.Context, string, string) bool
-	GetCRState(context.Context, string, string) CRState
 	GetCRStateForReference(context.Context, string, annotation.MaintenanceResourceReference, string) CRState
 }

fault-remediation/pkg/crstatus/crstatus_test.go

@@ -23,18 +23,12 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client/fake"
 
 	"github.com/nvidia/nvsentinel/fault-remediation/pkg/annotation"
-	"github.com/nvidia/nvsentinel/fault-remediation/pkg/config"
 )
 
 func TestCheckCondition(t *testing.T) {
-	testResource := config.MaintenanceResource{
-		CompleteConditionType: "Completed",
-	}
-	cfg := map[string]config.MaintenanceResource{
-		"test": testResource,
-	}
+	completeConditionType := "Completed"
 
-	checker := NewCRStatusChecker(nil, cfg, false)
+	checker := NewCRStatusChecker(nil, false)
 
 	tests := []struct {
 		name     string
@@ -118,7 +112,7 @@ func TestCheckCondition(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := checker.checkCondition(tt.cr, testResource)
+			result := checker.checkConditionType(tt.cr, completeConditionType)
 			assert.Equal(t, tt.expected, result)
 		})
 	}
@@ -145,15 +139,7 @@ func TestGetCRStateForReferenceUsesStoredReference(t *testing.T) {
 	}
 
 	fakeClient := fake.NewClientBuilder().WithObjects(storedCR).Build()
-	checker := NewCRStatusChecker(fakeClient, map[string]config.MaintenanceResource{
-		"RESTART_BM": {
-			ApiGroup:              "config.example.com",
-			Version:               "v1",
-			Kind:                  "ConfigMaintenance",
-			Namespace:             "config-namespace",
-			CompleteConditionType: "NodeReady",
-		},
-	}, false)
+	checker := NewCRStatusChecker(fakeClient, false)
 
 	state := checker.GetCRStateForReference(context.Background(), "stored-cr", annotation.MaintenanceResourceReference{
 		ApiGroup:  "stored.example.com",

fault-remediation/pkg/reconciler/reconciler_test.go

@@ -65,26 +65,13 @@ func (m *MockK8sClient) GetStatusChecker() crstatus.CRStatusCheckerInterface {
 }
 
 type mockStatusChecker struct {
-	getCRStateFn             func(context.Context, string, string) crstatus.CRState
 	getCRStateForReferenceFn func(context.Context, string, annotation.MaintenanceResourceReference, string) crstatus.CRState
 	shouldSkip               []bool
 	states                   []crstatus.CRState
 	stateByCR                map[string]crstatus.CRState
 	callCount                int
 }
 
-func (statusChecker *mockStatusChecker) ShouldSkipCRCreation(context.Context, string, string) bool {
-	return statusChecker.GetCRState(context.Background(), "", "") != crstatus.CRStateFailed
-}
-
-func (statusChecker *mockStatusChecker) GetCRState(ctx context.Context, actionName string, crName string) crstatus.CRState {
-	if statusChecker.getCRStateFn != nil {
-		return statusChecker.getCRStateFn(ctx, actionName, crName)
-	}
-
-	return statusChecker.nextState(crName)
-}
-
 func (statusChecker *mockStatusChecker) GetCRStateForReference(
 	ctx context.Context,
 	crName string,

fault-remediation/pkg/remediation/remediation.go

@@ -127,7 +127,6 @@ func NewRemediationClient(
 
 	ctrlRuntimeRemediationClient.statusChecker = crstatus.NewCRStatusChecker(
 		client,
-		remediationConfig.RemediationActions,
 		dryRun,
 	)