Merge pull request #8760 from BlaineEXE/bp-8708

ceph: retry object health check if creation fails (backport #8708)
rook · Sep 21, 2021 · c4011c9 · c4011c9
2 parents 8ccc3bd + 2de0418
commit c4011c9
Show file tree

Hide file tree

Showing 4 changed files with 194 additions and 177 deletions.
diff --git a/pkg/operator/ceph/object/admin.go b/pkg/operator/ceph/object/admin.go
@@ -150,7 +150,7 @@ func NewMultisiteAdminOpsContext(
 		return nil, errors.Wrapf(err, "failed to create or retrieve rgw admin ops user")
 	}
 
-	httpClient, tlsCert, err := GenObjectStoreHTTPClient(objContext, spec)
+	httpClient, tlsCert, err := genObjectStoreHTTPClientFunc(objContext, spec)
 	if err != nil {
 		return nil, err
 	}

diff --git a/pkg/operator/ceph/object/controller.go b/pkg/operator/ceph/object/controller.go
@@ -442,7 +442,10 @@ func (r *ReconcileCephObjectStore) reconcileCreateObjectStore(cephObjectStore *c
 
 	// Start monitoring
 	if !cephObjectStore.Spec.HealthCheck.Bucket.Disabled {
-		r.startMonitoring(cephObjectStore, objContext, namespacedName)
+		err = r.startMonitoring(cephObjectStore, objContext, namespacedName)
+		if err != nil {
+			return reconcile.Result{}, err
+		}
 	}
 
 	return reconcile.Result{}, nil
@@ -507,22 +510,23 @@ func (r *ReconcileCephObjectStore) reconcileMultisiteCRs(cephObjectStore *cephv1
 	return cephObjectStore.Name, cephObjectStore.Name, cephObjectStore.Name, reconcile.Result{}, nil
 }
 
-func (r *ReconcileCephObjectStore) startMonitoring(objectstore *cephv1.CephObjectStore, objContext *Context, namespacedName types.NamespacedName) {
+func (r *ReconcileCephObjectStore) startMonitoring(objectstore *cephv1.CephObjectStore, objContext *Context, namespacedName types.NamespacedName) error {
 	// Start monitoring object store
 	if r.objectStoreChannels[objectstore.Name].monitoringRunning {
-		logger.Debug("external rgw endpoint monitoring go routine already running!")
-		return
+		logger.Info("external rgw endpoint monitoring go routine already running!")
+		return nil
 	}
 
 	rgwChecker, err := newBucketChecker(r.context, objContext, r.client, namespacedName, &objectstore.Spec)
 	if err != nil {
-		logger.Error(err)
-		return
+		return errors.Wrapf(err, "failed to start rgw health checker for CephObjectStore %q, will re-reconcile", namespacedName.String())
 	}
 
-	logger.Info("starting rgw healthcheck")
+	logger.Infof("starting rgw health checker for CephObjectStore %q", namespacedName.String())
 	go rgwChecker.checkObjectStore(r.objectStoreChannels[objectstore.Name].stopChan)
 
 	// Set the monitoring flag so we don't start more than one go routine
 	r.objectStoreChannels[objectstore.Name].monitoringRunning = true
+
+	return nil
 }