fix: Avoid system cfg changed can't restart (#2253)

* fix: avoid system cfg changed pod can't restart fix: avoid system cfg changed pod can't restart * delete pdb delete pdb * comment comment * lint lint * init init * Update pkg/controller/pods.go Co-authored-by: Ramon de Klein <[email protected]> * apply suggestion apply suggestion * lint lint * apply suggestoin apply suggestoin --------- Co-authored-by: jiuker <[email protected]> Co-authored-by: Ramon de Klein <[email protected]>
minio · Aug 26, 2024 · 3f1a37b · 3f1a37b
1 parent 2c5289c
commit 3f1a37b
Show file tree

Hide file tree

Showing 3 changed files with 92 additions and 0 deletions.
diff --git a/pkg/controller/main-controller.go b/pkg/controller/main-controller.go
@@ -19,6 +19,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"maps"
 	"net/http"
 	"os"
 	"os/signal"
@@ -1316,6 +1317,29 @@ func (c *Controller) syncHandler(key string) (Result, error) {
 			// return nil so we don't re-queue this work item, this error won't get fixed by reprocessing
 			return WrapResult(Result{}, nil)
 		}
+
+		// check if the system config has changed
+		// if changed, minio request the systemCfg must be the same to restart.
+		expectedSystemCfg, err := c.getSystemCfgFromStatefulSet(ctx, expectedStatefulSet)
+		if err != nil {
+			return WrapResult(Result{}, err)
+		}
+		existingSystemCfg, err := c.getSystemCfgFromStatefulSet(ctx, existingStatefulSet)
+		if err != nil {
+			return WrapResult(Result{}, err)
+		}
+		if !maps.Equal(expectedSystemCfg, existingSystemCfg) {
+			// delete pdb to let all the statefulSet pods get deleted
+			err = c.DeletePDB(ctx, tenant)
+			if err != nil {
+				return Result{}, err
+			}
+			// find all existing statefulSet pods and delete them
+			err = c.DeletePodsByStatefulSet(ctx, existingStatefulSet)
+			if err != nil {
+				return WrapResult(Result{}, err)
+			}
+		}
 	}
 
 	// Handle PVC expansion

diff --git a/pkg/controller/pods.go b/pkg/controller/pods.go
@@ -17,12 +17,17 @@
 package controller
 
 import (
+	"context"
 	"fmt"
 	"time"
 
 	miniov2 "github.com/minio/operator/pkg/apis/minio.min.io/v2"
 	"github.com/minio/operator/pkg/utils"
+	appsv1 "k8s.io/api/apps/v1"
+	corev1 "k8s.io/api/core/v1"
 	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
+	"k8s.io/klog/v2"
+	"sigs.k8s.io/controller-runtime/pkg/client"
 )
 
 // handlePodChange will handle changes in pods and queue it for processing, pods are already filtered by PodInformer
@@ -43,3 +48,25 @@ func (c *Controller) handlePodChange(obj interface{}) {
 	key := fmt.Sprintf("%s/%s", object.GetNamespace(), instanceName)
 	c.healthCheckQueue.AddAfter(key, 1*time.Second)
 }
+
+// DeletePodsByStatefulSet deletes all pods associated with a statefulset
+func (c *Controller) DeletePodsByStatefulSet(ctx context.Context, sts *appsv1.StatefulSet) (err error) {
+	listOpt := &client.ListOptions{
+		Namespace: sts.Namespace,
+	}
+	client.MatchingLabels(sts.Spec.Template.Labels).ApplyToList(listOpt)
+	podList := &corev1.PodList{}
+	if err := c.k8sClient.List(ctx, podList, listOpt); err != nil {
+		return err
+	}
+	for _, item := range podList.Items {
+		if item.DeletionTimestamp == nil {
+			err = c.k8sClient.Delete(ctx, &item)
+			// Ignore Not Found
+			if client.IgnoreNotFound(err) != nil {
+				klog.Infof("unable to restart %s/%s (ignored): %s", item.Namespace, item.Name, err)
+			}
+		}
+	}
+	return
+}
diff --git a/pkg/controller/pools.go b/pkg/controller/pools.go
@@ -20,6 +20,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"strings"
 
 	corev1 "k8s.io/api/core/v1"
 
@@ -92,6 +93,46 @@ func poolSSMatchesSpec(expectedStatefulSet, existingStatefulSet *appsv1.Stateful
 	return true, nil
 }
 
+// getSystemCfgFromStatefulSet gets the MinIO environment variables from a statefulset
+func (c *Controller) getSystemCfgFromStatefulSet(ctx context.Context, sts *appsv1.StatefulSet) (systemCfg map[string]string, err error) {
+	systemCfg = make(map[string]string)
+	for _, container := range sts.Spec.Template.Spec.Containers {
+		if container.Name == miniov2.MinIOServerName {
+			for _, e := range container.Env {
+				if strings.HasPrefix(e.Name, "MINIO_") {
+					switch {
+					case e.Value != "":
+						systemCfg[e.Name] = e.Value
+					case e.ValueFrom != nil && e.ValueFrom.SecretKeyRef != nil:
+						secret, err := c.kubeClientSet.CoreV1().Secrets(sts.Namespace).Get(ctx, e.ValueFrom.SecretKeyRef.Name, metav1.GetOptions{})
+						if err != nil {
+							return nil, err
+						}
+						value, ok := secret.Data[e.ValueFrom.SecretKeyRef.Key]
+						if !ok {
+							return nil, fmt.Errorf("secret %s does not have key %s", e.ValueFrom.SecretKeyRef.Name, e.ValueFrom.SecretKeyRef.Key)
+						}
+						systemCfg[e.Name] = string(value)
+					case e.ValueFrom != nil && e.ValueFrom.ConfigMapKeyRef != nil:
+						configMap, err := c.kubeClientSet.CoreV1().ConfigMaps(sts.Namespace).Get(ctx, e.ValueFrom.ConfigMapKeyRef.Name, metav1.GetOptions{})
+						if err != nil {
+							return nil, err
+						}
+						value, ok := configMap.Data[e.ValueFrom.ConfigMapKeyRef.Key]
+						if !ok {
+							return nil, fmt.Errorf("configmap %s does not have key %s", e.ValueFrom.ConfigMapKeyRef.Name, e.ValueFrom.ConfigMapKeyRef.Key)
+						}
+						systemCfg[e.Name] = value
+					default:
+						return nil, fmt.Errorf("unsupported env var %s", e.Name)
+					}
+				}
+			}
+		}
+	}
+	return
+}
+
 // restartInitializedPool restarts a pool that is assumed to have been initialized
 func (c *Controller) restartInitializedPool(ctx context.Context, tenant *miniov2.Tenant, pool miniov2.Pool, tenantConfiguration map[string][]byte) error {
 	// get a new admin client that points to a pod of an already initialized pool (ie: pool-0)