Merge branch 'master' into release-v0.1.18

Signed-off-by: Matt Richerson <matthew.richerson@hpe.com>

Author: matthew-richerson

go.mod

@@ -1,14 +1,12 @@
 module github.com/NearNodeFlash/nnf-integration-test
 
-go 1.22.0
-
-toolchain go1.22.5
+go 1.23.9
 
 require (
-	github.com/DataWorkflowServices/dws v0.0.1-0.20250610194327-bc91fef8e840
-	github.com/NearNodeFlash/lustre-fs-operator v0.0.1-0.20250708135942-4a4d148d7772
-	github.com/NearNodeFlash/nnf-ec v0.0.1-0.20250702200322-db4e4089d5e0 // indirect
-	github.com/NearNodeFlash/nnf-sos v0.0.1-0.20250708141732-79259d489832
+	github.com/DataWorkflowServices/dws v0.0.1-0.20250826235427-3ba3755ecbb4
+	github.com/NearNodeFlash/lustre-fs-operator v0.0.1-0.20250827135601-4d150db05d08
+	github.com/NearNodeFlash/nnf-ec v0.0.1-0.20250829141301-67f617d97f7b // indirect
+	github.com/NearNodeFlash/nnf-sos v0.0.1-0.20250829145954-af98382abcac
 	github.com/onsi/ginkgo/v2 v2.22.2
 	github.com/onsi/gomega v1.36.2
 	go.openly.dev/pointy v1.3.0

go.sum

@@ -1,11 +1,11 @@
-github.com/DataWorkflowServices/dws v0.0.1-0.20250610194327-bc91fef8e840 h1:2Jh7al4IH9R+d2kFX/pBE8A63TM4nGOwBSZcL1FBFzo=
-github.com/DataWorkflowServices/dws v0.0.1-0.20250610194327-bc91fef8e840/go.mod h1:i9v4K2d64a9uyd9ZazOJ85RWzVQjIgHWXkLo895KDAE=
-github.com/NearNodeFlash/lustre-fs-operator v0.0.1-0.20250708135942-4a4d148d7772 h1:7yokdQ0uKl1OJLSNL0x6dMmpFpbvcwJS3/aX7xVcTdg=
-github.com/NearNodeFlash/lustre-fs-operator v0.0.1-0.20250708135942-4a4d148d7772/go.mod h1:Hzd47Ti0pzNB44yeWgDDdNuRqfWFDkJNbMLEXwtjEyE=
-github.com/NearNodeFlash/nnf-ec v0.0.1-0.20250702200322-db4e4089d5e0 h1:meUKolZ546eXAfReA4ZxS+s1D9Rek4IuKfyT00mr1dw=
-github.com/NearNodeFlash/nnf-ec v0.0.1-0.20250702200322-db4e4089d5e0/go.mod h1:lx13ustzE/+39fLECky+CFKkAV8GYlX9eaI6IGmHQkY=
-github.com/NearNodeFlash/nnf-sos v0.0.1-0.20250708141732-79259d489832 h1:eAvmmB/FkzGwdaLxOBONC2WJ9VSYz2cHsfOK8QuAzh0=
-github.com/NearNodeFlash/nnf-sos v0.0.1-0.20250708141732-79259d489832/go.mod h1:f5duBSZaTrkf3fIksVxLXXTAW9rpziK6aMBn7RNFm/M=
+github.com/DataWorkflowServices/dws v0.0.1-0.20250826235427-3ba3755ecbb4 h1:c5bKT/hZUHzgx4HxiFMGoxEFDXPv/qlFmLwDTbJAA00=
+github.com/DataWorkflowServices/dws v0.0.1-0.20250826235427-3ba3755ecbb4/go.mod h1:NbNOIG2ybY2XHVSJn0KF7edPlX3I1ji7je6MGzuPeaA=
+github.com/NearNodeFlash/lustre-fs-operator v0.0.1-0.20250827135601-4d150db05d08 h1:1IYv2gDH/+hxBo6TZenIRs30PSpzO49USDkj/aBo/g4=
+github.com/NearNodeFlash/lustre-fs-operator v0.0.1-0.20250827135601-4d150db05d08/go.mod h1:1lpNL1ZYtVwDU6Fuju5u/xiza6Z2Ua8KWZeaGBKpn5o=
+github.com/NearNodeFlash/nnf-ec v0.0.1-0.20250829141301-67f617d97f7b h1:jkegsuJ6p7yaTcg+jK4sHy5xzR7z4qlQRR7euy8TSZ0=
+github.com/NearNodeFlash/nnf-ec v0.0.1-0.20250829141301-67f617d97f7b/go.mod h1:slDsqUJtN2uZ5Z7z4J6wWrtKksQAaH4JqRzUonDXNKo=
+github.com/NearNodeFlash/nnf-sos v0.0.1-0.20250829145954-af98382abcac h1:VrYaMmjYt89dzPsDQI2Hcu22HBPAV1SkcVfxHo8Hh9g=
+github.com/NearNodeFlash/nnf-sos v0.0.1-0.20250829145954-af98382abcac/go.mod h1:XMK6ByWuISN7DghUxYm4jdSEfA/+5LVrkZy4jpRHlqA=
 github.com/benbjohnson/clock v1.1.0/go.mod h1:J11/hYXuz8f4ySSvYwY0FKfm+ezbsZBKZxNJlLklBHA=
 github.com/benbjohnson/clock v1.3.0 h1:ip6w0uFQkncKQ979AypyG0ER7mqUSBdKLOgAle/AT8A=
 github.com/benbjohnson/clock v1.3.0/go.mod h1:J11/hYXuz8f4ySSvYwY0FKfm+ezbsZBKZxNJlLklBHA=

int_test.go

@@ -28,7 +28,7 @@ import (
 	. "github.com/onsi/gomega"
 	"go.openly.dev/pointy"
 
-	dwsv1alpha4 "github.com/DataWorkflowServices/dws/api/v1alpha4"
+	dwsv1alpha6 "github.com/DataWorkflowServices/dws/api/v1alpha6"
 )
 
 var (
@@ -41,9 +41,9 @@ var (
 	highTimeout = "5m"
 
 	// Which states use the high timeout
-	highTimeoutStates = []dwsv1alpha4.WorkflowState{
-		dwsv1alpha4.StateSetup,
-		dwsv1alpha4.StateTeardown,
+	highTimeoutStates = []dwsv1alpha6.WorkflowState{
+		dwsv1alpha6.StateSetup,
+		dwsv1alpha6.StateTeardown,
 	}
 )
 
@@ -143,17 +143,17 @@ var tests = []*T{
 		"#DW container name=prerun-timeout-mpi profile=example-mpi-prerun-timeout").
 		WithPermissions(1050, 1051).WithLabels("mpi", "timeout").
 		WithContainerProfile("example-mpi", &ContainerProfileOptions{PrerunTimeoutSeconds: pointy.Int(1), NoStorage: true}).
-		ExpectError(dwsv1alpha4.StatePreRun),
+		ExpectError(dwsv1alpha6.StatePreRun),
 	MakeTest("PostRun timeout on MPI containers",
 		"#DW container name=postrun-timeout-mpi profile=example-mpi-postrun-timeout").
 		WithPermissions(1050, 1051).WithLabels("mpi", "timeout").
 		WithContainerProfile("example-mpi-webserver", &ContainerProfileOptions{PostrunTimeoutSeconds: pointy.Int(1), NoStorage: true}).
-		ExpectError(dwsv1alpha4.StatePostRun),
+		ExpectError(dwsv1alpha6.StatePostRun),
 	MakeTest("Non-zero exit on MPI containers",
 		"#DW container name=mpi-container-fail profile=example-mpi-fail-noretry").
 		WithPermissions(1050, 1051).WithLabels("mpi", "fail").
 		WithContainerProfile("example-mpi-fail", &ContainerProfileOptions{RetryLimit: pointy.Int(0)}).
-		ExpectError(dwsv1alpha4.StatePostRun),
+		ExpectError(dwsv1alpha6.StatePostRun),
 
 	// Containers - Non-MPI
 	MakeTest("GFS2 with Containers",
@@ -175,27 +175,27 @@ var tests = []*T{
 		"#DW container name=prerun-timeout profile=example-prerun-timeout").
 		WithPermissions(1050, 1051).WithLabels("non-mpi", "timeout").
 		WithContainerProfile("example-forever", &ContainerProfileOptions{PrerunTimeoutSeconds: pointy.Int(1), NoStorage: true}).
-		ExpectError(dwsv1alpha4.StatePreRun),
+		ExpectError(dwsv1alpha6.StatePreRun),
 	MakeTest("PostRun timeout on non-MPI containers",
 		"#DW container name=postrun-timeout profile=example-postrun-timeout").
 		WithPermissions(1050, 1051).WithLabels("non-mpi", "timeout").
 		WithContainerProfile("example-forever", &ContainerProfileOptions{PostrunTimeoutSeconds: pointy.Int(1), NoStorage: true}).
-		ExpectError(dwsv1alpha4.StatePostRun),
+		ExpectError(dwsv1alpha6.StatePostRun),
 	MakeTest("Non-zero exit on non-MPI containers",
 		"#DW container name=container-fail profile=example-fail-noretry").
 		WithPermissions(1050, 1051).WithLabels("non-mpi", "fail").
 		WithContainerProfile("example-fail", &ContainerProfileOptions{RetryLimit: pointy.Int(0)}).
-		ExpectError(dwsv1alpha4.StatePostRun),
+		ExpectError(dwsv1alpha6.StatePostRun),
 
 	// Containers - Unsupported Filesystems. These should fail as xfs/raw filesystems are not supported for containers.
 	MakeTest("XFS with Containers",
 		"#DW jobdw type=xfs name=xfs-with-containers capacity=100GB",
 		"#DW container name=xfs-with-containers profile=example-success DW_JOB_foo_local_storage=xfs-with-containers").
-		ExpectError(dwsv1alpha4.StateProposal).WithLabels("unsupported-fs"),
+		ExpectError(dwsv1alpha6.StateProposal).WithLabels("unsupported-fs"),
 	MakeTest("Raw with Containers",
 		"#DW jobdw type=raw name=raw-with-containers capacity=100GB",
 		"#DW container name=raw-with-containers profile=example-success DW_JOB_foo_local_storage=raw-with-containers").
-		ExpectError(dwsv1alpha4.StateProposal).WithLabels("unsupported-fs"),
+		ExpectError(dwsv1alpha6.StateProposal).WithLabels("unsupported-fs"),
 
 	// Containers - Multiple Storages
 	MakeTest("GFS2 and Lustre with Containers",
@@ -250,7 +250,7 @@ var _ = Describe("NNF Integration Test", func() {
 						// TODO: Ginkgo's `--fail-fast` option still seems to execute DeferCleanup() calls
 						//       See if this is by design or if we might need to move this to an AfterEach()
 						if !context.SpecReport().Failed() {
-							t.AdvanceStateAndWaitForReady(ctx, k8sClient, workflow, dwsv1alpha4.StateTeardown)
+							t.AdvanceStateAndWaitForReady(ctx, k8sClient, workflow, dwsv1alpha6.StateTeardown)
 
 							Expect(k8sClient.Delete(ctx, workflow)).To(Succeed())
 						}

internal/internal.go

@@ -29,7 +29,7 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/intstr"
 
-	dwsv1alpha4 "github.com/DataWorkflowServices/dws/api/v1alpha4"
+	dwsv1alpha6 "github.com/DataWorkflowServices/dws/api/v1alpha6"
 	"github.com/DataWorkflowServices/dws/utils/dwdparse"
 )
 
@@ -68,7 +68,7 @@ type T struct {
 	decorators []interface{}
 
 	// Workflow defines the DWS Workflow resource that is the target of the test.
-	workflow *dwsv1alpha4.Workflow
+	workflow *dwsv1alpha6.Workflow
 
 	// User Id to use for the workflow
 	userId int
@@ -84,7 +84,7 @@ type T struct {
 	helperPods []*corev1.Pod
 
 	// Compute nodes that were assigned to the test. This is determined at test runtime.
-	computes *dwsv1alpha4.Computes
+	computes *dwsv1alpha6.Computes
 }
 
 func MakeTest(name string, directives ...string) *T {
@@ -118,13 +118,13 @@ func MakeTest(name string, directives ...string) *T {
 		t.directives = []string{}
 	}
 
-	t.workflow = &dwsv1alpha4.Workflow{
+	t.workflow = &dwsv1alpha6.Workflow{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      t.WorkflowName(),
 			Namespace: corev1.NamespaceDefault,
 		},
-		Spec: dwsv1alpha4.WorkflowSpec{
-			DesiredState: dwsv1alpha4.StateProposal,
+		Spec: dwsv1alpha6.WorkflowSpec{
+			DesiredState: dwsv1alpha6.StateProposal,
 			DWDirectives: t.WorkflowDirectives(),
 			JobID:        intstr.FromInt(GinkgoParallelProcess()),
 			WLMID:        strconv.Itoa(GinkgoParallelProcess()),
@@ -145,7 +145,7 @@ func (t *T) WorkflowDirectives() []string {
 	return t.directives
 }
 
-func (t *T) Workflow() *dwsv1alpha4.Workflow {
+func (t *T) Workflow() *dwsv1alpha6.Workflow {
 	return t.workflow
 }
 

internal/options.go

@@ -35,9 +35,9 @@ import (
 	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
-	dwsv1alpha4 "github.com/DataWorkflowServices/dws/api/v1alpha4"
+	dwsv1alpha6 "github.com/DataWorkflowServices/dws/api/v1alpha6"
 	lusv1alpha1 "github.com/NearNodeFlash/lustre-fs-operator/api/v1alpha1"
-	nnfv1alpha7 "github.com/NearNodeFlash/nnf-sos/api/v1alpha7"
+	nnfv1alpha8 "github.com/NearNodeFlash/nnf-sos/api/v1alpha8"
 
 	"github.com/DataWorkflowServices/dws/utils/dwdparse"
 )
@@ -57,7 +57,7 @@ type TOptions struct {
 	hardwareRequired    bool
 	lowTimeout          time.Duration
 	highTimeout         time.Duration
-	highTimeoutStates   []dwsv1alpha4.WorkflowState
+	highTimeoutStates   []dwsv1alpha6.WorkflowState
 	useExternalComputes bool
 }
 
@@ -67,21 +67,21 @@ func (o *TOptions) hasComplexOptions() bool {
 }
 
 type TStopAfter struct {
-	state dwsv1alpha4.WorkflowState
+	state dwsv1alpha6.WorkflowState
 }
 
 // Stop after lets you stop a test after a given state is reached
-func (t *T) StopAfter(state dwsv1alpha4.WorkflowState) *T {
+func (t *T) StopAfter(state dwsv1alpha6.WorkflowState) *T {
 	t.options.stopAfter = &TStopAfter{state: state}
 	return t
 }
 
 type TExpectError struct {
-	state dwsv1alpha4.WorkflowState
+	state dwsv1alpha6.WorkflowState
 }
 
 // Expect an error at the designed state; Proceed to teardown
-func (t *T) ExpectError(state dwsv1alpha4.WorkflowState) *T {
+func (t *T) ExpectError(state dwsv1alpha6.WorkflowState) *T {
 	t.options.expectError = &TExpectError{state: state}
 	t.options.stopAfter = &TStopAfter{state: state}
 	return t.WithLabels("error")
@@ -312,7 +312,7 @@ func (t *T) Prepare(ctx context.Context, k8sClient client.Client) error {
 		By(fmt.Sprintf("Creating storage profile '%s'", o.storageProfile.name))
 
 		// Clone the default profile.
-		defaultProf := &nnfv1alpha7.NnfStorageProfile{
+		defaultProf := &nnfv1alpha8.NnfStorageProfile{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      "default",
 				Namespace: "nnf-system",
@@ -321,7 +321,7 @@ func (t *T) Prepare(ctx context.Context, k8sClient client.Client) error {
 
 		Expect(k8sClient.Get(ctx, client.ObjectKeyFromObject(defaultProf), defaultProf)).To(Succeed())
 
-		profile := &nnfv1alpha7.NnfStorageProfile{
+		profile := &nnfv1alpha8.NnfStorageProfile{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      o.storageProfile.name,
 				Namespace: "nnf-system",
@@ -345,7 +345,7 @@ func (t *T) Prepare(ctx context.Context, k8sClient client.Client) error {
 
 	if o.containerProfile != nil {
 		// Clone the provided base profile
-		baseProfile := &nnfv1alpha7.NnfContainerProfile{
+		baseProfile := &nnfv1alpha8.NnfContainerProfile{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      o.containerProfile.base,
 				Namespace: "nnf-system",
@@ -354,7 +354,7 @@ func (t *T) Prepare(ctx context.Context, k8sClient client.Client) error {
 
 		Expect(k8sClient.Get(ctx, client.ObjectKeyFromObject(baseProfile), baseProfile)).To(Succeed())
 
-		profile := &nnfv1alpha7.NnfContainerProfile{
+		profile := &nnfv1alpha8.NnfContainerProfile{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      o.containerProfile.name,
 				Namespace: "nnf-system",
@@ -408,7 +408,7 @@ func (t *T) Prepare(ctx context.Context, k8sClient client.Client) error {
 
 		// Extract the File System Name and MGSNids from the persistent lustre instance. This
 		// assumes an NNF Storage resource is created in the same name as the persistent instance
-		storage := &nnfv1alpha7.NnfStorage{
+		storage := &nnfv1alpha8.NnfStorage{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      name,
 				Namespace: corev1.NamespaceDefault,
@@ -545,7 +545,7 @@ func (t *T) Cleanup(ctx context.Context, k8sClient client.Client) error {
 	if t.options.storageProfile != nil {
 		By(fmt.Sprintf("Deleting storage profile '%s'", o.storageProfile.name))
 
-		profile := &nnfv1alpha7.NnfStorageProfile{
+		profile := &nnfv1alpha8.NnfStorageProfile{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      o.storageProfile.name,
 				Namespace: "nnf-system",
@@ -559,7 +559,7 @@ func (t *T) Cleanup(ctx context.Context, k8sClient client.Client) error {
 	if t.options.containerProfile != nil {
 		By(fmt.Sprintf("Deleting container profile '%s'", o.containerProfile.name))
 
-		profile := &nnfv1alpha7.NnfContainerProfile{
+		profile := &nnfv1alpha8.NnfContainerProfile{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      o.containerProfile.name,
 				Namespace: "nnf-system",