Add remediateLastFailure

api/v1alpha2/terraform_types.go

@@ -286,6 +286,12 @@ type Remediation struct {
 	// retries.
 	// +optional
 	Retries int64 `json:"retries,omitempty"`
+
+	// RemediateLastFailure tells the controller to remediate the last failure, when
+	// no retries remain. Defaults to 'false' unless 'Retries' is greater than 0.
+	// +optional
+	RemediateLastFailure *bool `json:"remediateLastFailure,omitempty"`
+
 }
 
 type CloudSpec struct {
@@ -954,6 +960,15 @@ func (in *Terraform) GetRetries() int64 {
 	return in.Spec.Remediation.Retries
 }
 
+// MustRemediateLastFailure returns whether to remediate the last failure when
+// no retries remain.
+func (in *Terraform) MustRemediateLastFailure() bool {
+	if in.Spec.Remediation.RemediateLastFailure == nil {
+		return false
+	}
+	return *in.Spec.Remediation.RemediateLastFailure
+}
+
 func (in *Terraform) GetReconciliationFailures() int64 {
 	return in.Status.ReconciliationFailures
 }

charts/tofu-controller/crds/crds.yaml

@@ -5581,6 +5581,12 @@ spec:
                       retries.
                     format: int64
                     type: integer
+                  remediateLastFailure:
+                    default: false
+                    description: |-
+                      remediateLastFailure instructs the controller to remediate the last failure
+                      when no retries remain. Defaults to false
+                    type: boolean
                 type: object
               retryInterval:
                 description: |-

controllers/tf_controller.go

@@ -443,6 +443,16 @@ func (r *TerraformReconciler) Reconcile(ctx context.Context, req ctrl.Request) (
 			terraform.GetGeneration(),
 		))
 
+		if terraform.MustRemediateLastFailure() {
+			log.Info("RemediateLastFailure is true, reseting retries, requeue after interval", "interval", terraform.Spec.Interval.Duration.String())
+			terraform = infrav1.TerraformResetRetry(terraform)
+			if err := r.patchStatus(ctx, req.NamespacedName, terraform.Status); err != nil {
+				log.Error(err, "unable to update status after maximum number of retries reached")
+				return ctrl.Result{Requeue: true}, err
+			}
+			return ctrl.Result{RequeueAfter: terraform.Spec.Interval.Duration}, nil
+		}
+
 		terraform = infrav1.TerraformReachedLimit(terraform)
 
 		traceLog.Info("Patch the status of the Terraform resource")

docs/References/terraform.md

@@ -702,6 +702,19 @@ before bailing. Defaults to ‘0’, a negative integer denotes unlimite
 retries.</p>
 </td>
 </tr>
+<tr>
+<td>
+<code>remediateLastFailure</code><br>
+<em>
+bool
+</em>
+</td>
+<td>
+<em>(Optional)</em>
+<p>RemediateLastFailure tells the controller to remediate the last failure, when
+no retries remain. Defaults to &lsquo;false&rsquo; unless &lsquo;Retries&rsquo; is greater than 0.</p>
+</td>
+</tr>
 </tbody>
 </table>
 </div>