Bug 86133: validate duplicate Nutanix failureDomain names and topology

pkg/types/nutanix/validation/platform.go

@@ -3,6 +3,8 @@ package validation
 import (
 	"fmt"
 	"regexp"
+	"sort"
+	"strings"
 
 	"k8s.io/apimachinery/pkg/util/validation/field"
 
@@ -103,11 +105,20 @@ func ValidatePlatform(p *nutanix.Platform, fldPath *field.Path, c *types.Install
 		if err != nil {
 			allErrs = append(allErrs, field.InternalError(fldPath.Child("failureDomain", "name"), fmt.Errorf("fail to compile the pattern %q: %w", pattern, err)))
 		} else {
-			for _, fd := range p.FailureDomains {
+			fdNames := make(map[string]int)
+			fdTopologies := make(map[string]string)
+
+			for idx, fd := range p.FailureDomains {
 				if !rexp.MatchString(fd.Name) {
 					allErrs = append(allErrs, field.Invalid(fldPath.Child("failureDomain", "name"), fd.Name, fmt.Sprintf("failureDomain name should match the pattern %q.", pattern)))
 				}
 
+				if prevIdx, exists := fdNames[fd.Name]; exists {
+					allErrs = append(allErrs, field.Duplicate(fldPath.Child("failureDomains").Index(idx).Child("name"), fmt.Sprintf("failure domain name %q is already used by failureDomains[%d]", fd.Name, prevIdx)))
+				} else {
+					fdNames[fd.Name] = idx
+				}
+
 				if fd.PrismElement.UUID == "" {
 					allErrs = append(allErrs, field.Required(fldPath.Child("failureDomain", "prismElement", "uuid"), "failureDomain prismElement uuid cannot be empty"))
 				}
@@ -117,6 +128,19 @@ func ValidatePlatform(p *nutanix.Platform, fldPath *field.Path, c *types.Install
 					allErrs = append(allErrs, errs...)
 				}
 
+				// Only check for duplicate topology after basic required-field validation,
+				// so users see actionable "required field" errors instead of misleading
+				// "identical topology" when fields are simply empty.
+				if fd.PrismElement.UUID != "" && len(fd.SubnetUUIDs) > 0 {
+					topoKey := nutanixFailureDomainTopologyKey(fd)
+					if prevName, exists := fdTopologies[topoKey]; exists {
+						allErrs = append(allErrs, field.Invalid(fldPath.Child("failureDomains").Index(idx), fd.Name,
+							fmt.Sprintf("failure domain %q has identical topology (same prismElement and subnets) as %q; this provides no additional fault tolerance", fd.Name, prevName)))
+					} else {
+						fdTopologies[topoKey] = fd.Name
+					}
+				}
+
 				for _, sc := range fd.StorageContainers {
 					if sc.ReferenceName == "" {
 						allErrs = append(allErrs, field.Required(fldPath.Child("failureDomain", "storageContainers", "referenceName"), fmt.Sprintf("failureDomain %q: missing storageContainer referenceName", fd.Name)))
@@ -153,6 +177,15 @@ func validateLoadBalancer(lbType configv1.PlatformLoadBalancerType) bool {
 	}
 }
 
+// nutanixFailureDomainTopologyKey builds a comparable key from the infrastructure-defining
+// fields of a failure domain: Prism Element UUID and sorted subnet UUIDs.
+func nutanixFailureDomainTopologyKey(fd nutanix.FailureDomain) string {
+	subnets := make([]string, len(fd.SubnetUUIDs))
+	copy(subnets, fd.SubnetUUIDs)
+	sort.Strings(subnets)
+	return fmt.Sprintf("pe=%s;subnets=%s", fd.PrismElement.UUID, strings.Join(subnets, "\x00"))
+}
+
 // validateSubnets validates the input subnetUUIDs meet the configuration requirements.
 func validateSubnets(fldPath *field.Path, subnetUUIDs []string) field.ErrorList {
 	var errs field.ErrorList

pkg/types/nutanix/validation/platform_test.go

@@ -409,6 +409,85 @@ func TestValidatePlatform(t *testing.T) {
 				return p
 			}(),
 		},
+		{
+			name: "failureDomain with duplicate name",
+			platform: func() *nutanix.Platform {
+				p := validPlatform()
+				p.FailureDomains = []nutanix.FailureDomain{
+					{
+						Name:         "fd-1",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid-1", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe-1", Port: 9440}},
+						SubnetUUIDs:  []string{"fd-subnet-uuid-1"},
+					},
+					{
+						Name:         "fd-1",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid-2", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe-2", Port: 9440}},
+						SubnetUUIDs:  []string{"fd-subnet-uuid-2"},
+					},
+				}
+				return p
+			}(),
+			expectedError: `test-path\.failureDomains\[1\]\.name: Duplicate value: "failure domain name \\"fd-1\\" is already used by failureDomains\[0\]"`,
+		},
+		{
+			name: "failureDomain with duplicate topology same prismElement and subnet",
+			platform: func() *nutanix.Platform {
+				p := validPlatform()
+				p.FailureDomains = []nutanix.FailureDomain{
+					{
+						Name:         "fd-1",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe", Port: 9440}},
+						SubnetUUIDs:  []string{"fd-subnet-uuid"},
+					},
+					{
+						Name:         "fd-2",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe", Port: 9440}},
+						SubnetUUIDs:  []string{"fd-subnet-uuid"},
+					},
+				}
+				return p
+			}(),
+			expectedError: `test-path\.failureDomains\[1\]: Invalid value: "fd-2": failure domain "fd-2" has identical topology \(same prismElement and subnets\) as "fd-1"; this provides no additional fault tolerance`,
+		},
+		{
+			name: "valid failureDomain with different prismElements",
+			platform: func() *nutanix.Platform {
+				p := validPlatform()
+				p.FailureDomains = []nutanix.FailureDomain{
+					{
+						Name:         "fd-1",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid-1", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe-1", Port: 9440}},
+						SubnetUUIDs:  []string{"fd-subnet-uuid"},
+					},
+					{
+						Name:         "fd-2",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid-2", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe-2", Port: 9440}},
+						SubnetUUIDs:  []string{"fd-subnet-uuid"},
+					},
+				}
+				return p
+			}(),
+		},
+		{
+			name: "failureDomain duplicate topology with same subnets in different order",
+			platform: func() *nutanix.Platform {
+				p := validPlatform()
+				p.FailureDomains = []nutanix.FailureDomain{
+					{
+						Name:         "fd-1",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe", Port: 9440}},
+						SubnetUUIDs:  []string{"subnet-a", "subnet-b"},
+					},
+					{
+						Name:         "fd-2",
+						PrismElement: nutanix.PrismElement{UUID: "fd-pe-uuid", Endpoint: nutanix.PrismEndpoint{Address: "fd-pe", Port: 9440}},
+						SubnetUUIDs:  []string{"subnet-b", "subnet-a"},
+					},
+				}
+				return p
+			}(),
+			expectedError: `test-path\.failureDomains\[1\]: Invalid value: "fd-2": failure domain "fd-2" has identical topology \(same prismElement and subnets\) as "fd-1"; this provides no additional fault tolerance`,
+		},
 		{
 			name: "valid failureDomain with multiple subnets for multi-NIC",
 			platform: func() *nutanix.Platform {