my-rook-config/pkg/operator/ceph/cluster/predicate.go

/*
Copyright 2020 The Rook Authors. All rights reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

	http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

// Package cluster to manage a Ceph cluster.
package cluster

import (
	"context"
	"strings"

	"github.com/google/go-cmp/cmp"
	"github.com/google/go-cmp/cmp/cmpopts"
	cephv1 "github.com/rook/rook/pkg/apis/ceph.rook.io/v1"
	"github.com/rook/rook/pkg/clusterd"
	discoverDaemon "github.com/rook/rook/pkg/daemon/discover"
	"github.com/rook/rook/pkg/operator/ceph/cluster/osd/topology"
	opcontroller "github.com/rook/rook/pkg/operator/ceph/controller"
	"github.com/rook/rook/pkg/operator/k8sutil"
	"github.com/rook/rook/pkg/util/log"
	corev1 "k8s.io/api/core/v1"
	"k8s.io/apimachinery/pkg/api/resource"
	v1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	"sigs.k8s.io/controller-runtime/pkg/client"
	"sigs.k8s.io/controller-runtime/pkg/event"
	"sigs.k8s.io/controller-runtime/pkg/predicate"
)

// Always trigger a reconcile when a secret is deleted. This will cause a
// reconciliation failure to happen immediately in hopes of alerting the end
// user to the configuration problem.
var changedOrDeleted = predicate.Or(
	predicate.TypedResourceVersionChangedPredicate[*corev1.Secret]{},
	predicate.TypedFuncs[*corev1.Secret]{
		DeleteFunc: func(e event.TypedDeleteEvent[*corev1.Secret]) bool {
			return true
		},
	},
)

func shouldReconcileChangedNode(objOld, objNew *corev1.Node) bool {
	// do not watch node if only resourceversion got changed
	resourceQtyComparer := cmpopts.IgnoreFields(v1.ObjectMeta{}, "ResourceVersion")
	diff := cmp.Diff(objOld.Spec, objNew.Spec, resourceQtyComparer)

	// do not watch node if only LastHeartbeatTime got changed
	resourceQtyComparer = cmpopts.IgnoreFields(corev1.NodeCondition{}, "LastHeartbeatTime")
	diff1 := cmp.Diff(objOld.Spec, objNew.Spec, resourceQtyComparer)

	if diff == "" && diff1 == "" {
		return false
	}
	return true
}

// topologyLabels is the set of node labels Rook uses to build the OSD CRUSH map
// topology (the hostname, the Kubernetes region/zone labels, and the
// topology.rook.io/* labels). Only changes to one of these labels are relevant
// to Rook; other label or annotation changes on a node must not trigger a
// reconcile.
var topologyLabels = strings.Split(topology.GetDefaultTopologyLabels(), ",")

// nodeTopologyLabelsChanged returns true if the value of any of the OSD topology
// labels Rook cares about differs between the old and new node. This detects a
// topology label being added, removed, or changed.
func nodeTopologyLabelsChanged(objOld, objNew *corev1.Node) bool {
	oldLabels := objOld.GetLabels()
	newLabels := objNew.GetLabels()
	for _, label := range topologyLabels {
		if oldLabels[label] != newLabels[label] {
			return true
		}
	}
	return false
}

// predicateForNodeWatcher is the predicate function to trigger reconcile on Node events
func predicateForNodeWatcher[T *corev1.Node](ctx context.Context, client client.Client, context *clusterd.Context, opNamespace string) predicate.TypedFuncs[T] {
	return predicate.TypedFuncs[T]{
		CreateFunc: func(e event.TypedCreateEvent[T]) bool {
			obj := (*corev1.Node)(e.Object)
			if e.IsInInitialList {
				// Skip node handling while the controller runtime is initializing the cache.
				return false
			}
			clientCluster := newClientCluster(client, obj.GetNamespace(), context)
			return clientCluster.onK8sNode(ctx, obj, opNamespace)
		},

		UpdateFunc: func(e event.TypedUpdateEvent[T]) bool {
			objOld := (*corev1.Node)(e.ObjectOld)
			objNew := (*corev1.Node)(e.ObjectNew)

			// When one of the OSD topology labels Rook cares about (for example
			// the topology.rook.io/* labels) changes, trigger a reconcile
			// directly. Otherwise onK8sNode() would return false for a node that
			// is already an OSD host and the relabel would never be propagated to
			// the CRUSH map.
			if nodeTopologyLabelsChanged(objOld, objNew) {
				return true
			}

			if !shouldReconcileChangedNode(objOld, objNew) {
				return false
			}

			clientCluster := newClientCluster(client, objNew.GetNamespace(), context)
			return clientCluster.onK8sNode(ctx, objNew, opNamespace)
		},

		DeleteFunc: func(e event.TypedDeleteEvent[T]) bool {
			return false
		},

		GenericFunc: func(e event.TypedGenericEvent[T]) bool {
			return false
		},
	}
}

// predicateForHotPlugCMWatcher is the predicate function to trigger reconcile on ConfigMap events (hot-plug)
func predicateForHotPlugCMWatcher[T *corev1.ConfigMap](client client.Client) predicate.TypedFuncs[T] {
	return predicate.TypedFuncs[T]{
		UpdateFunc: func(e event.TypedUpdateEvent[T]) bool {
			objOld := (*corev1.ConfigMap)(e.ObjectOld)
			objNew := (*corev1.ConfigMap)(e.ObjectNew)

			if !isHotPlugCM(objNew) {
				return false
			}

			clientCluster := newClientCluster(client, objNew.GetNamespace(), &clusterd.Context{})
			return clientCluster.onDeviceCMUpdate(objOld, objNew)
		},

		DeleteFunc: func(e event.TypedDeleteEvent[T]) bool {
			// TODO: if the configmap goes away we could retrigger rook-discover DS
			// However at this point the returned bool can only trigger a reconcile of the CephCluster object
			// Definitely non-trivial but nice to have in the future
			return false
		},

		CreateFunc: func(e event.TypedCreateEvent[T]) bool {
			return false
		},

		GenericFunc: func(e event.TypedGenericEvent[T]) bool {
			return false
		},
	}
}

// isHotPlugCM informs whether the object is the cm for hot-plug disk
func isHotPlugCM(cm *corev1.ConfigMap) bool {
	// Get the labels
	labels := cm.GetLabels()

	labelVal, labelKeyExist := labels[k8sutil.AppAttr]
	if labelKeyExist && labelVal == discoverDaemon.AppName {
		return true
	}

	return false
}

// predicateForClusterConfigMapWatcher is the predicate function to trigger reconcile on operator config ConfigMap events
func predicateForClusterConfigMapWatcher[T *corev1.ConfigMap](ctx context.Context, client client.Client) predicate.TypedFuncs[T] {
	return predicate.TypedFuncs[T]{
		UpdateFunc: func(e event.TypedUpdateEvent[T]) bool {
			objNew := (*corev1.ConfigMap)(e.ObjectNew)
			objOld := (*corev1.ConfigMap)(e.ObjectOld)

			if objNew.Name == opcontroller.OperatorSettingConfigMapName {
				if objOld.Data["ROOK_USE_CSI_OPERATOR"] != objNew.Data["ROOK_USE_CSI_OPERATOR"] {
					log.NamespacedInfo(objNew.Namespace, logger, "ROOK_USE_CSI_OPERATOR changed from %q to %q", objOld.Data["ROOK_USE_CSI_OPERATOR"], objNew.Data["ROOK_USE_CSI_OPERATOR"])
					return true
				}
			}

			if isFloatingMonConfigMap(ctx, client, objNew) {
				log.NamespacedInfo(objNew.Namespace, logger, "floating mon ConfigMap %q changed", objNew.Name)
				return true
			}

			return false
		},

		CreateFunc: func(e event.TypedCreateEvent[T]) bool {
			objNew := (*corev1.ConfigMap)(e.Object)
			return objNew.Name == opcontroller.OperatorSettingConfigMapName
		},

		DeleteFunc: func(e event.TypedDeleteEvent[T]) bool {
			return false
		},

		GenericFunc: func(e event.TypedGenericEvent[T]) bool {
			return false
		},
	}
}

// isFloatingMonConfigMap checks if any floating mon configuration refers to the updated ConfigMap
func isFloatingMonConfigMap(ctx context.Context, client client.Client, cm *corev1.ConfigMap) bool {
	clusterList := cephv1.CephClusterList{}
	err := client.List(ctx, &clusterList)
	if err != nil {
		return false
	}

	for _, cluster := range clusterList.Items {
		if cluster.Spec.Mon.FloatingMon.ConfigMapName != "" {
			if cm.GetName() == cluster.Spec.Mon.FloatingMon.ConfigMapName {
				return true
			}
		}
	}

	return false
}

func watchControllerPredicate[T *cephv1.CephCluster](ctx context.Context, c client.Client) predicate.TypedFuncs[T] {
	return predicate.TypedFuncs[T]{
		CreateFunc: func(e event.TypedCreateEvent[T]) bool {
			if opcontroller.DuplicateCephClusters(ctx, c, e.Object, true) {
				return false
			}
			logger.Debug("create event from a CR")
			return true
		},
		DeleteFunc: func(e event.TypedDeleteEvent[T]) bool {
			logger.Debug("delete event from a CR")
			return true
		},
		UpdateFunc: func(e event.TypedUpdateEvent[T]) bool {
			// We still need to check on update event since the user must delete the additional CR
			// Until this is done, the user can still update the CR and the operator will reconcile
			// This should not happen
			if opcontroller.DuplicateCephClusters(ctx, c, e.ObjectOld, true) {
				return false
			}

			// resource.Quantity has non-exportable fields, so we use its comparator method
			resourceQtyComparer := cmp.Comparer(func(x, y resource.Quantity) bool { return x.Cmp(y) == 0 })

			objOld := (*cephv1.CephCluster)(e.ObjectOld)
			objNew := (*cephv1.CephCluster)(e.ObjectNew)

			log.NamespacedDebug(objNew.Namespace, logger, "update event on CephCluster CR")
			// If the user sets the skip-reconcile label, avoid doing anything (including manager reloads)
			// for this CephCluster.
			//
			// However, we still want to trigger a reconcile when the label is added/removed so the
			// controller can early-exit and record a "ReconcileSkipped" event (when label is added),
			// and also resume reconciliation immediately when the label is removed.
			_, oldHasSkipLabel := objOld.GetLabels()[cephv1.SkipReconcileLabelKey]
			_, newHasSkipLabel := objNew.GetLabels()[cephv1.SkipReconcileLabelKey]
			if oldHasSkipLabel != newHasSkipLabel {
				log.NamespacedDebug(objNew.Namespace, logger, "CephCluster %q skip-reconcile label changed (%v -> %v); triggering reconcile", objNew.Name, oldHasSkipLabel, newHasSkipLabel)
				return true
			}
			if newHasSkipLabel {
				log.NamespacedInfo(objNew.Namespace, logger, "object %q matched on update but %q label is set, doing nothing", objNew.Name, cephv1.SkipReconcileLabelKey)
				return false
			}

			// Backwards/alternative label used by some controllers to avoid reconciling an object.
			// If this label is set on the object, let's not reconcile that request.
			if opcontroller.IsDoNotReconcile(objNew.GetLabels()) {
				log.NamespacedDebug(objNew.Namespace, logger, "object %q matched on update but %q label is set, doing nothing", opcontroller.DoNotReconcileLabelName, objNew.Name)
				return false
			}
			diff := cmp.Diff(objOld.Spec, objNew.Spec, resourceQtyComparer)
			if diff != "" {
				log.NamespacedInfo(objNew.Namespace, logger, "CR has changed for %q. diff=%s", objNew.Name, diff)

				if objNew.Spec.CleanupPolicy.HasDataDirCleanPolicy() {
					log.NamespacedInfo(objNew.Namespace, logger, "skipping orchestration for cluster object %q in namespace %q because its cleanup policy is set. not reloading the manager", objNew.GetName(), objNew.GetNamespace())
					return false
				}

				// Stop any ongoing orchestration
				opcontroller.ReloadManager()

				return false

			} else if !objOld.GetDeletionTimestamp().Equal(objNew.GetDeletionTimestamp()) {
				log.NamespacedInfo(objNew.Namespace, logger, "CR %q is going be deleted, cancelling any ongoing orchestration", objNew.Name)

				// Stop any ongoing orchestration
				opcontroller.ReloadManager()

				return false

			} else if objOld.GetGeneration() != objNew.GetGeneration() {
				log.NamespacedDebug(objNew.Namespace, logger, "reconciling CephCluster %q with changed generation", objNew.Name)
				return true
			}

			return false
		},
		GenericFunc: func(e event.TypedGenericEvent[T]) bool {
			return false
		},
	}
}