[ADXT-275] [k8s] add dump of k8s cluster state on k8s suites errors (#…

…26198)

test/new-e2e/pkg/e2e/provisioner.go

@@ -10,6 +10,11 @@ import (
 	"io"
 )
 
+// Diagnosable defines the interface for a diagnosable provider.
+type Diagnosable interface {
+	Diagnose(ctx context.Context, stackName string) (string, error)
+}
+
 // Provisioner defines the interface for a provisioner.
 type Provisioner interface {
 	ID() string

test/new-e2e/pkg/e2e/pulumi_provisioner.go

@@ -26,9 +26,10 @@ type PulumiEnvRunFunc[Env any] func(ctx *pulumi.Context, env *Env) error
 
 // PulumiProvisioner is a provisioner based on Pulumi with binding to an environment.
 type PulumiProvisioner[Env any] struct {
-	id        string
-	runFunc   PulumiEnvRunFunc[Env]
-	configMap runner.ConfigMap
+	id           string
+	runFunc      PulumiEnvRunFunc[Env]
+	configMap    runner.ConfigMap
+	diagnoseFunc func(ctx context.Context, stackName string) (string, error)
 }
 
 var (
@@ -101,6 +102,19 @@ func (pp *PulumiProvisioner[Env]) ProvisionEnv(ctx context.Context, stackName st
 	return resources, nil
 }
 
+// Diagnose runs the diagnose function if it is set diagnoseFunc
+func (pp *PulumiProvisioner[Env]) Diagnose(ctx context.Context, stackName string) (string, error) {
+	if pp.diagnoseFunc != nil {
+		return pp.diagnoseFunc(ctx, stackName)
+	}
+	return "", nil
+}
+
+// SetDiagnoseFunc sets the diagnose function.
+func (pp *PulumiProvisioner[Env]) SetDiagnoseFunc(diagnoseFunc func(ctx context.Context, stackName string) (string, error)) {
+	pp.diagnoseFunc = diagnoseFunc
+}
+
 // Destroy deletes the Pulumi stack.
 func (pp *PulumiProvisioner[Env]) Destroy(ctx context.Context, stackName string, logger io.Writer) error {
 	return infra.GetStackManager().DeleteStack(ctx, stackName, logger)

test/new-e2e/pkg/e2e/suite.go

@@ -150,6 +150,7 @@ import (
 
 	"github.com/DataDog/datadog-agent/test/new-e2e/pkg/runner"
 	"github.com/DataDog/datadog-agent/test/new-e2e/pkg/runner/parameters"
+	"github.com/DataDog/datadog-agent/test/new-e2e/pkg/utils/infra"
 	"github.com/DataDog/test-infra-definitions/common/utils"
 	"github.com/DataDog/test-infra-definitions/components"
 
@@ -284,6 +285,20 @@ func (bs *BaseSuite[Env]) reconcileEnv(targetProvisioners ProvisionerMap) error
 		}
 
 		if err != nil {
+			if diagnosableProvisioner, ok := provisioner.(Diagnosable); ok {
+				stackName, err := infra.GetStackManager().GetPulumiStackName(bs.params.stackName)
+				if err != nil {
+					bs.T().Logf("unable to get stack name for diagnose, err: %v", err)
+				} else {
+					diagnoseResult, diagnoseErr := diagnosableProvisioner.Diagnose(ctx, stackName)
+					if diagnoseErr != nil {
+						bs.T().Logf("WARNING: Diagnose failed: %v", diagnoseErr)
+					} else if diagnoseResult != "" {
+						bs.T().Logf("Diagnose result: %s", diagnoseResult)
+					}
+				}
+
+			}
 			return fmt.Errorf("your stack '%s' provisioning failed, check logs above. Provisioner was %s, failed with err: %v", bs.params.stackName, id, err)
 		}
 

test/new-e2e/pkg/environments/aws/kubernetes/eks.go

@@ -7,6 +7,9 @@
 package awskubernetes
 
 import (
+	"context"
+	"fmt"
+
 	"github.com/DataDog/test-infra-definitions/common/config"
 	"github.com/DataDog/test-infra-definitions/common/utils"
 	"github.com/DataDog/test-infra-definitions/components"
@@ -34,6 +37,14 @@ import (
 	"github.com/pulumi/pulumi/sdk/v3/go/pulumi"
 )
 
+func eksDiagnoseFunc(ctx context.Context, stackName string) (string, error) {
+	dumpResult, err := dumpEKSClusterState(ctx, stackName)
+	if err != nil {
+		return "", err
+	}
+	return fmt.Sprintf("Dumping EKS cluster state:\n%s", dumpResult), nil
+}
+
 // EKSProvisioner creates a new provisioner
 func EKSProvisioner(opts ...ProvisionerOption) e2e.TypedProvisioner[environments.AwsKubernetes] {
 	// We ALWAYS need to make a deep copy of `params`, as the provisioner can be called multiple times.
@@ -50,6 +61,8 @@ func EKSProvisioner(opts ...ProvisionerOption) e2e.TypedProvisioner[environments
 		return EKSRunFunc(ctx, env, params)
 	}, params.extraConfigParams)
 
+	provisioner.SetDiagnoseFunc(eksDiagnoseFunc)
+
 	return provisioner
 }
 

test/new-e2e/pkg/environments/aws/kubernetes/kind.go

@@ -7,6 +7,7 @@
 package awskubernetes
 
 import (
+	"context"
 	"fmt"
 
 	"github.com/DataDog/datadog-agent/test/new-e2e/pkg/e2e"
@@ -31,6 +32,14 @@ const (
 	defaultVMName     = "kind"
 )
 
+func kindDiagnoseFunc(ctx context.Context, stackName string) (string, error) {
+	dumpResult, err := dumpKindClusterState(ctx, stackName)
+	if err != nil {
+		return "", err
+	}
+	return fmt.Sprintf("Dumping Kind cluster state:\n%s", dumpResult), nil
+}
+
 // KindProvisioner creates a new provisioner
 func KindProvisioner(opts ...ProvisionerOption) e2e.TypedProvisioner[environments.Kubernetes] {
 	// We ALWAYS need to make a deep copy of `params`, as the provisioner can be called multiple times.
@@ -47,6 +56,8 @@ func KindProvisioner(opts ...ProvisionerOption) e2e.TypedProvisioner[environment
 		return KindRunFunc(ctx, env, params)
 	}, params.extraConfigParams)
 
+	provisioner.SetDiagnoseFunc(kindDiagnoseFunc)
+
 	return provisioner
 }