Use ansibleEE library instead of AnsibleEE v1 CR

This change swaps out the use of the AnsibleEE custom resource for
a generic Go library implementation. This enables us to stop using the unnecessary
OpenStackAnsibleEE abstraction that we currently have around Kubernetes Jobs for
AnsibleEE executions.

Jira: https://issues.redhat.com/browse/OSPRH-8926
Signed-off-by: Brendan Shephard <[email protected]>

controllers/dataplane/openstackdataplanedeployment_controller.go

@@ -22,6 +22,7 @@ import (
 	"time"
 
 	"github.com/go-playground/validator/v10"
+	batchv1 "k8s.io/api/batch/v1"
 	corev1 "k8s.io/api/core/v1"
 	k8s_errors "k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -35,7 +36,6 @@ import (
 	condition "github.com/openstack-k8s-operators/lib-common/modules/common/condition"
 	"github.com/openstack-k8s-operators/lib-common/modules/common/helper"
 	"github.com/openstack-k8s-operators/lib-common/modules/common/util"
-	ansibleeev1 "github.com/openstack-k8s-operators/openstack-ansibleee-operator/api/v1beta1"
 	dataplanev1 "github.com/openstack-k8s-operators/openstack-operator/apis/dataplane/v1beta1"
 	deployment "github.com/openstack-k8s-operators/openstack-operator/pkg/dataplane"
 	dataplaneutil "github.com/openstack-k8s-operators/openstack-operator/pkg/dataplane/util"
@@ -427,6 +427,6 @@ func (r *OpenStackDataPlaneDeploymentReconciler) setHashes(
 func (r *OpenStackDataPlaneDeploymentReconciler) SetupWithManager(mgr ctrl.Manager) error {
 	return ctrl.NewControllerManagedBy(mgr).
 		For(&dataplanev1.OpenStackDataPlaneDeployment{}).
-		Owns(&ansibleeev1.OpenStackAnsibleEE{}).
+		Owns(&batchv1.Job{}).
 		Complete(r)
 }

controllers/dataplane/openstackdataplanenodeset_controller.go

@@ -24,6 +24,7 @@ import (
 
 	"github.com/go-playground/validator/v10"
 	"golang.org/x/exp/slices"
+	batchv1 "k8s.io/api/batch/v1"
 	corev1 "k8s.io/api/core/v1"
 	rbacv1 "k8s.io/api/rbac/v1"
 	k8s_errors "k8s.io/apimachinery/pkg/api/errors"
@@ -48,7 +49,6 @@ import (
 	"github.com/openstack-k8s-operators/lib-common/modules/common/secret"
 	"github.com/openstack-k8s-operators/lib-common/modules/common/serviceaccount"
 	"github.com/openstack-k8s-operators/lib-common/modules/common/util"
-	ansibleeev1 "github.com/openstack-k8s-operators/openstack-ansibleee-operator/api/v1beta1"
 	baremetalv1 "github.com/openstack-k8s-operators/openstack-baremetal-operator/api/v1beta1"
 	openstackv1 "github.com/openstack-k8s-operators/openstack-operator/apis/core/v1beta1"
 	dataplanev1 "github.com/openstack-k8s-operators/openstack-operator/apis/dataplane/v1beta1"
@@ -576,7 +576,8 @@ func checkDeployment(ctx context.Context, helper *helper.Helper,
 
 // SetupWithManager sets up the controller with the Manager.
 func (r *OpenStackDataPlaneNodeSetReconciler) SetupWithManager(
-	ctx context.Context, mgr ctrl.Manager) error {
+	ctx context.Context, mgr ctrl.Manager,
+) error {
 	// index for ConfigMaps listed on ansibleVarsFrom
 	if err := mgr.GetFieldIndexer().IndexField(ctx,
 		&dataplanev1.OpenStackDataPlaneNodeSet{}, "spec.ansibleVarsFrom.ansible.configMaps",
@@ -629,7 +630,7 @@ func (r *OpenStackDataPlaneNodeSetReconciler) SetupWithManager(
 	}
 	return ctrl.NewControllerManagedBy(mgr).
 		For(&dataplanev1.OpenStackDataPlaneNodeSet{}).
-		Owns(&ansibleeev1.OpenStackAnsibleEE{}).
+		Owns(&batchv1.Job{}).
 		Owns(&baremetalv1.OpenStackBaremetalSet{}).
 		Owns(&infranetworkv1.IPSet{}).
 		Owns(&infranetworkv1.DNSData{}).

go.mod

@@ -130,3 +130,5 @@ replace github.com/openshift/api => github.com/openshift/api v0.0.0-202304141430
 
 // custom RabbitmqClusterSpecCore for OpenStackControlplane (v2.6.0_patches_tag)
 replace github.com/rabbitmq/cluster-operator/v2 => github.com/openstack-k8s-operators/rabbitmq-cluster-operator/v2 v2.6.1-0.20240719064129-c0201810c8f5 //allow-merging
+
+replace github.com/openstack-k8s-operators/lib-common/modules/ansible => github.com/bshephar/lib-common/modules/ansible v0.0.0-20240724052033-e8fa85ebec00

pkg/dataplane/deployment.go

@@ -24,6 +24,9 @@ import (
 	"sort"
 	"strconv"
 
+	batchv1 "k8s.io/api/batch/v1"
+	corev1 "k8s.io/api/core/v1"
+
 	slices "golang.org/x/exp/slices"
 	k8s_errors "k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/types"
@@ -35,11 +38,9 @@ import (
 	"github.com/openstack-k8s-operators/lib-common/modules/common/helper"
 	"github.com/openstack-k8s-operators/lib-common/modules/common/util"
 	"github.com/openstack-k8s-operators/lib-common/modules/storage"
-	ansibleeev1 "github.com/openstack-k8s-operators/openstack-ansibleee-operator/api/v1beta1"
 	openstackv1 "github.com/openstack-k8s-operators/openstack-operator/apis/core/v1beta1"
 	dataplanev1 "github.com/openstack-k8s-operators/openstack-operator/apis/dataplane/v1beta1"
 	dataplaneutil "github.com/openstack-k8s-operators/openstack-operator/pkg/dataplane/util"
-	corev1 "k8s.io/api/core/v1"
 )
 
 // Deployer defines a data structure with all of the relevant objects required for a full deployment.
@@ -195,7 +196,7 @@ func (d *Deployer) ConditionalDeploy(
 	}
 
 	if nsConditions.IsFalse(readyCondition) {
-		var ansibleEE *ansibleeev1.OpenStackAnsibleEE
+		var ansibleEE *batchv1.Job
 		_, labelSelector := dataplaneutil.GetAnsibleExecutionNameAndLabels(&foundService, d.Deployment.Name, d.NodeSet.Name)
 		ansibleEE, err = dataplaneutil.GetAnsibleExecution(d.Ctx, d.Helper, d.Deployment, labelSelector)
 		if err != nil {
@@ -213,34 +214,39 @@ func (d *Deployer) ConditionalDeploy(
 				err.Error()))
 		}
 
-		if ansibleEE.Status.JobStatus == ansibleeev1.JobStatusSucceeded {
+		if ansibleEE.Status.Succeeded != 0 {
 			log.Info(fmt.Sprintf("Condition %s ready", readyCondition))
 			nsConditions.Set(condition.TrueCondition(
 				readyCondition,
 				readyMessage))
 		}
 
-		if ansibleEE.Status.JobStatus == ansibleeev1.JobStatusRunning || ansibleEE.Status.JobStatus == ansibleeev1.JobStatusPending {
-			log.Info(fmt.Sprintf("AnsibleEE job is not yet completed: Execution: %s, Status: %s", ansibleEE.Name, ansibleEE.Status.JobStatus))
+		if ansibleEE.Status.Succeeded == 0 || ansibleEE.Status.Active != 0 && ansibleEE.Status.Failed != 0 {
+			log.Info(fmt.Sprintf("AnsibleEE job is not yet completed: Execution: %s, Active pods: %d", ansibleEE.Name, ansibleEE.Status.Active))
 			nsConditions.Set(condition.FalseCondition(
 				readyCondition,
 				condition.RequestedReason,
 				condition.SeverityInfo,
 				readyWaitingMessage))
 		}
 
-		if ansibleEE.Status.JobStatus == ansibleeev1.JobStatusFailed {
-			errorMsg := fmt.Sprintf("execution.name %s execution.namespace %s execution.status.jobstatus: %s", ansibleEE.Name, ansibleEE.Namespace, ansibleEE.Status.JobStatus)
-			ansibleCondition := ansibleEE.Status.Conditions.Get(condition.ReadyCondition)
+		var ansibleCondition *batchv1.JobCondition
+		if ansibleEE.Status.Failed != 0 {
+			errorMsg := fmt.Sprintf("execution.name %s execution.namespace %s failed pods: %d", ansibleEE.Name, ansibleEE.Namespace, ansibleEE.Status.Failed)
+			for _, condition := range ansibleEE.Status.Conditions {
+				if condition.Type == batchv1.JobFailed {
+					ansibleCondition = &condition
+				}
+			}
 			if ansibleCondition.Reason == condition.JobReasonBackoffLimitExceeded {
-				errorMsg = fmt.Sprintf("backoff limit reached for execution.name %s execution.namespace %s execution.status.jobstatus: %s", ansibleEE.Name, ansibleEE.Namespace, ansibleEE.Status.JobStatus)
+				errorMsg = fmt.Sprintf("backoff limit reached for execution.name %s execution.namespace %s execution.condition.message: %s", ansibleEE.Name, ansibleEE.Namespace, ansibleCondition.Message)
 			}
 			log.Info(fmt.Sprintf("Condition %s error", readyCondition))
 			err = fmt.Errorf(errorMsg)
 			nsConditions.Set(condition.FalseCondition(
 				readyCondition,
-				ansibleCondition.Reason,
-				ansibleCondition.Severity,
+				condition.Reason(ansibleCondition.Reason),
+				condition.SeverityError,
 				readyErrorMessage,
 				err.Error()))
 		}

pkg/dataplane/service.go

@@ -47,15 +47,13 @@ type ServiceYAML struct {
 // DeployService service deployment
 func (d *Deployer) DeployService(foundService dataplanev1.OpenStackDataPlaneService) error {
 	err := dataplaneutil.AnsibleExecution(
-		d.Ctx,
 		d.Helper,
 		d.Deployment,
 		&foundService,
 		d.AnsibleSSHPrivateKeySecrets,
 		d.InventorySecrets,
 		d.AeeSpec,
 		d.NodeSet)
-
 	if err != nil {
 		d.Helper.GetLogger().Error(err, fmt.Sprintf("Unable to execute Ansible for %s", foundService.Name))
 		return err
@@ -160,8 +158,9 @@ func EnsureServices(ctx context.Context, helper *helper.Helper, instance *datapl
 // Global Services in multiple NodeSets for a deployment
 func DedupeServices(ctx context.Context, helper *helper.Helper,
 	nodesets []dataplanev1.OpenStackDataPlaneNodeSet,
-	serviceOverride []string) (map[string][]string, error) {
-	var nodeSetServiceMap = make(map[string][]string, 0)
+	serviceOverride []string,
+) (map[string][]string, error) {
+	nodeSetServiceMap := make(map[string][]string, 0)
 	var globalServices []string
 	if len(serviceOverride) != 0 {
 		services, err := dedupe(ctx, helper, serviceOverride, globalServices)
@@ -182,7 +181,8 @@ func DedupeServices(ctx context.Context, helper *helper.Helper,
 }
 
 func dedupe(ctx context.Context, helper *helper.Helper,
-	services []string, globalServices []string) ([]string, error) {
+	services []string, globalServices []string,
+) ([]string, error) {
 	var dedupedServices []string
 	var nodeSetServiceTypes []string
 	for _, svc := range services {