Provisioning: update unhealthy status only when not recent (#112014)

pkg/registry/apis/provisioning/controller/health.go

@@ -12,6 +12,13 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 )
 
+const (
+	// recentHealthyDuration defines how recent a health check must be to be considered "recent" when healthy
+	recentHealthyDuration = 5 * time.Minute
+	// recentHealthyDuration defines how recent a health check must be to be considered "recent" when unhealthy
+	recentUnhealthyDuration = 1 * time.Minute
+)
+
 // StatusPatcher defines the interface for updating repository status
 //
 //go:generate mockery --name=StatusPatcher
@@ -60,9 +67,9 @@ func (hc *HealthChecker) hasRecentHealthCheck(healthStatus provisioning.HealthSt
 
 	age := time.Since(time.UnixMilli(healthStatus.Checked))
 	if healthStatus.Healthy {
-		return age <= time.Minute*5 // Recent if checked within 5 minutes when healthy
+		return age <= recentHealthyDuration
 	}
-	return age <= time.Minute // Recent if checked within 1 minute when unhealthy
+	return age <= recentUnhealthyDuration // Recent if checked within 1 minute when unhealthy
 }
 
 // HasRecentFailure checks if there's a recent failure of a specific type
@@ -72,7 +79,7 @@ func (hc *HealthChecker) HasRecentFailure(healthStatus provisioning.HealthStatus
 	}
 
 	age := time.Since(time.UnixMilli(healthStatus.Checked))
-	return age <= time.Minute // Recent if within 1 minute
+	return age <= recentUnhealthyDuration
 }
 
 // RecordFailureAndUpdate records a failure and updates the repository status
@@ -111,7 +118,11 @@ func (hc *HealthChecker) hasHealthStatusChanged(old, new provisioning.HealthStat
 		return true
 	}
 
-	if old.Checked != new.Checked {
+	recent := recentUnhealthyDuration
+	if new.Healthy {
+		recent = recentHealthyDuration
+	}
+	if time.UnixMilli(new.Checked).Sub(time.UnixMilli(old.Checked)) > recent {
 		return true
 	}
 

pkg/registry/apis/provisioning/controller/health_test.go

@@ -400,6 +400,26 @@ func TestRefreshHealth(t *testing.T) {
 			expectedHealth: true,
 			expectPatch:    false,
 		},
+		{
+			name: "no status change - no patch needed for unhealthy repo (recent check)",
+			testResult: &provisioning.TestResults{
+				Success: false,
+				Code:    500,
+				Errors: []provisioning.ErrorDetails{
+					{Detail: "connection failed"},
+					{Detail: "timeout"},
+				},
+			},
+			testError: nil,
+			existingStatus: provisioning.HealthStatus{
+				Healthy: false,
+				Checked: time.Now().Add(-15 * time.Second).UnixMilli(),
+				Message: []string{"connection failed", "timeout"},
+			},
+			expectError:    false,
+			expectedHealth: false,
+			expectPatch:    false,
+		},
 		{
 			name: "status unchanged but timestamp needs update (old check)",
 			testResult: &provisioning.TestResults{
@@ -415,6 +435,26 @@ func TestRefreshHealth(t *testing.T) {
 			expectedHealth: true,
 			expectPatch:    true,
 		},
+		{
+			name: "status unchanged but timestamp needs update (old unhealthy check)",
+			testResult: &provisioning.TestResults{
+				Success: false,
+				Code:    500,
+				Errors: []provisioning.ErrorDetails{
+					{Detail: "connection failed"},
+					{Detail: "timeout"},
+				},
+			},
+			testError: nil,
+			existingStatus: provisioning.HealthStatus{
+				Healthy: false,
+				Checked: time.Now().Add(-2 * time.Minute).UnixMilli(),
+				Message: []string{"connection failed", "timeout"},
+			},
+			expectError:    false,
+			expectedHealth: false,
+			expectPatch:    true,
+		},
 		{
 			name: "patch error",
 			testResult: &provisioning.TestResults{