From 5de275cdad048a24e3ba017a203944fc9290ba03 Mon Sep 17 00:00:00 2001
From: Isa Aguilar <isa@aguilarstory.com>
Date: Wed, 25 Oct 2023 01:53:25 -0400
Subject: [PATCH] enable retry based on labels

Fixes https://github.com/GalleyBytes/terraform-operator/issues/150

WIP still need to handle
- retry when already in delete stages
- cleanup of status after generation changes
- fix typos in comments and/or descriptions
---
 .../tf.galleybytes.com_terraforms_crd.yaml    | 15 +++++++-
 pkg/apis/tf/v1beta1/terraform_types.go        | 17 +++++++++
 pkg/controllers/terraform_controller.go       | 38 +++++++++++++++++++
 3 files changed, 69 insertions(+), 1 deletion(-)

diff --git a/deploy/crds/tf.galleybytes.com_terraforms_crd.yaml b/deploy/crds/tf.galleybytes.com_terraforms_crd.yaml
index ab416fc..36d655a 100644
--- a/deploy/crds/tf.galleybytes.com_terraforms_crd.yaml
+++ b/deploy/crds/tf.galleybytes.com_terraforms_crd.yaml
@@ -3,7 +3,8 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.12.0
+    controller-gen.kubebuilder.io/version: v0.9.2
+  creationTimestamp: null
   name: terraforms.tf.galleybytes.com
 spec:
   group: tf.galleybytes.com
@@ -2563,6 +2564,18 @@ spec:
                   it, the chance of recycling existing resources is reduced to virtually
                   nil.
                 type: string
+              retryEventReson:
+                description: "RetryEventReason writes a copy of the current resource
+                  label for 'kubernetes.io/change-cause'. The value will be written
+                  as here as the retry reason. \n When '.setup' is is the suffix,
+                  the pipeline will be started from the setup task. \n Example of
+                  starting from setup: \n ```yaml metadata: labels: kubernetes.io/change-cause:
+                  triggered-by-isa_aguilar-20231025T011600.setup ``` \n A default
+                  retry will start from the init task otherwise."
+                type: string
+              retryTimestamp:
+                format: date-time
+                type: string
               stage:
                 description: Stage stores information about the current stage
                 properties:
diff --git a/pkg/apis/tf/v1beta1/terraform_types.go b/pkg/apis/tf/v1beta1/terraform_types.go
index e0e78b3..8aab2c8 100644
--- a/pkg/apis/tf/v1beta1/terraform_types.go
+++ b/pkg/apis/tf/v1beta1/terraform_types.go
@@ -675,6 +675,23 @@ type TerraformStatus struct {
 	// refreshed each generation.
 	// +optional
 	PluginsStarted []TaskName `json:"pluginsStarted,omitempty"`
+
+	// RetryEventReason writes a copy of the current resource label for 'kubernetes.io/change-cause'. The
+	// value will be written as here as the retry reason.
+	//
+	// When '.setup' is is the suffix, the pipeline will be started from the setup task.
+	//
+	// Example of starting from setup:
+	//
+	// ```yaml
+	// metadata:
+	//   labels:
+	//     kubernetes.io/change-cause: triggered-by-isa_aguilar-20231025T011600.setup
+	// ```
+	//
+	// A default retry will start from the init task otherwise.
+	RetryEventReason *string      `json:"retryEventReson,omitempty"`
+	RetryTimestamp   *metav1.Time `json:"retryTimestamp,omitempty"`
 }
 
 type Exported string
diff --git a/pkg/controllers/terraform_controller.go b/pkg/controllers/terraform_controller.go
index cff6fbd..d13c61c 100644
--- a/pkg/controllers/terraform_controller.go
+++ b/pkg/controllers/terraform_controller.go
@@ -588,6 +588,34 @@ func (r *ReconcileTerraform) Reconcile(ctx context.Context, request reconcile.Re
 		return reconcile.Result{}, err
 	}
 
+	if tf.Labels != nil {
+		if label, found := tf.Labels["kubernetes.io/change-cause"]; found {
+			retry := false
+			if tf.Status.RetryEventReason == nil {
+				retry = true
+			} else if *tf.Status.RetryEventReason != label {
+				retry = true
+			}
+
+			if retry {
+				now := metav1.Now()
+				tf.Status.RetryEventReason = &label
+				tf.Status.RetryTimestamp = &now
+				taskType := tfv1beta1.RunInit
+				if strings.HasSuffix(label, ".setup") {
+					taskType = tfv1beta1.RunSetup
+				}
+				stage := newStage(tf, taskType, label, isTaskInterruptable(taskType), tfv1beta1.StateInitializing)
+				tf.Status.Stage = *stage
+				err := r.updateStatusWithRetry(ctx, tf, &tf.Status, reqLogger)
+				if err != nil {
+					reqLogger.V(1).Info(err.Error())
+				}
+				return reconcile.Result{}, nil
+			}
+		}
+	}
+
 	// Final delete by removing finalizers
 	if tf.Status.Phase == tfv1beta1.PhaseDeleted {
 		reqLogger.Info("Remove finalizers")
@@ -765,6 +793,16 @@ func (r *ReconcileTerraform) Reconcile(ctx context.Context, request reconcile.Re
 		return reconcile.Result{}, nil
 	}
 
+	if tf.Status.RetryTimestamp != nil {
+		podSlice := []corev1.Pod{}
+		for _, pod := range pods.Items {
+			if pod.CreationTimestamp.IsZero() || !pod.CreationTimestamp.Before(tf.Status.RetryTimestamp) {
+				podSlice = append(podSlice, pod)
+			}
+		}
+		pods.Items = podSlice
+	}
+
 	if len(pods.Items) == 0 && tf.Status.Stage.State == tfv1beta1.StateInProgress {
 		// This condition is generally met when the user deletes the pod.
 		// Force the state to transition away from in-progress and then