feat(horaemeta): add node inspector (#1483)

## Rationale
In some extreme scenarios, etcd events may be lost or update failures
may occur, which may cause the mapping between nodes and shards to fail
to be updated correctly. We need to add a cover-up mechanism to ensure
that the mapping relationship is always correct.

## Detailed Changes
* Add `NodeInspector`, it will start when the cluster start, running in
he background and detect the status of nodes.

## Test Plan
Pass CI.

---------

Co-authored-by: xikai.wxk <xikai.wxk@antgroup.com>

horaemeta/server/cluster/cluster.go

@@ -26,6 +26,7 @@ import (
 	"github.com/apache/incubator-horaedb-meta/server/cluster/metadata"
 	"github.com/apache/incubator-horaedb-meta/server/coordinator"
 	"github.com/apache/incubator-horaedb-meta/server/coordinator/eventdispatch"
+	"github.com/apache/incubator-horaedb-meta/server/coordinator/inspector"
 	"github.com/apache/incubator-horaedb-meta/server/coordinator/procedure"
 	"github.com/apache/incubator-horaedb-meta/server/coordinator/scheduler/manager"
 	"github.com/apache/incubator-horaedb-meta/server/id"
@@ -47,6 +48,7 @@ type Cluster struct {
 	procedureFactory *coordinator.Factory
 	procedureManager procedure.Manager
 	schedulerManager manager.SchedulerManager
+	nodeInspector    *inspector.NodeInspector
 }
 
 func NewCluster(logger *zap.Logger, metadata *metadata.ClusterMetadata, client *clientv3.Client, rootPath string) (*Cluster, error) {
@@ -62,12 +64,15 @@ func NewCluster(logger *zap.Logger, metadata *metadata.ClusterMetadata, client *
 
 	schedulerManager := manager.NewManager(logger, procedureManager, procedureFactory, metadata, client, rootPath, metadata.GetTopologyType(), metadata.GetProcedureExecutingBatchSize())
 
+	nodeInspector := inspector.NewNodeInspector(logger, metadata)
+
 	return &Cluster{
 		logger:           logger,
 		metadata:         metadata,
 		procedureFactory: procedureFactory,
 		procedureManager: procedureManager,
 		schedulerManager: schedulerManager,
+		nodeInspector:    nodeInspector,
 	}, nil
 }
 
@@ -78,6 +83,9 @@ func (c *Cluster) Start(ctx context.Context) error {
 	if err := c.schedulerManager.Start(ctx); err != nil {
 		return errors.WithMessage(err, "start scheduler manager")
 	}
+	if err := c.nodeInspector.Start(ctx); err != nil {
+		return errors.WithMessage(err, "start node inspector")
+	}
 	return nil
 }
 
@@ -88,6 +96,9 @@ func (c *Cluster) Stop(ctx context.Context) error {
 	if err := c.schedulerManager.Stop(ctx); err != nil {
 		return errors.WithMessage(err, "stop scheduler manager")
 	}
+	if err := c.nodeInspector.Stop(ctx); err != nil {
+		return errors.WithMessage(err, "stop node inspector")
+	}
 	return nil
 }
 

horaemeta/server/cluster/metadata/cluster_metadata.go

@@ -685,7 +685,7 @@ func (c *ClusterMetadata) UpdateClusterViewByNode(ctx context.Context, shardNode
 	return nil
 }
 
-func (c *ClusterMetadata) DropShardNode(ctx context.Context, shardNodes []storage.ShardNode) error {
+func (c *ClusterMetadata) DropShardNodes(ctx context.Context, shardNodes []storage.ShardNode) error {
 	if err := c.topologyManager.DropShardNodes(ctx, shardNodes); err != nil {
 		return errors.WithMessage(err, "drop shard nodes")
 	}

horaemeta/server/cluster/metadata/cluster_metadata_test.go

@@ -213,7 +213,7 @@ func testShardOperation(ctx context.Context, re *require.Assertions, m *metadata
 	_, err = m.GetShardNodeByTableIDs([]storage.TableID{})
 	re.NoError(err)
 
-	err = m.DropShardNode(ctx, []storage.ShardNode{{
+	err = m.DropShardNodes(ctx, []storage.ShardNode{{
 		ID:        shardNodeResult.NodeShards[0].ShardNode.ID,
 		ShardRole: shardNodeResult.NodeShards[0].ShardNode.ShardRole,
 		NodeName:  shardNodeResult.NodeShards[0].ShardNode.NodeName,

horaemeta/server/coordinator/inspector/node_inspector.go

@@ -0,0 +1,147 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you 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 inspector
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	"github.com/apache/incubator-horaedb-meta/pkg/coderr"
+	"github.com/apache/incubator-horaedb-meta/pkg/log"
+	"github.com/apache/incubator-horaedb-meta/server/cluster/metadata"
+	"github.com/apache/incubator-horaedb-meta/server/storage"
+	"go.uber.org/zap"
+)
+
+var ErrStartAgain = coderr.NewCodeError(coderr.Internal, "try to start again")
+var ErrStopNotStart = coderr.NewCodeError(coderr.Internal, "try to stop a not-started inspector")
+
+const defaultInspectInterval = time.Second * 5
+
+// NodeInspector will inspect node status and remove expired data.
+type NodeInspector struct {
+	logger          *zap.Logger
+	clusterMetadata ClusterMetaDataManipulator
+	interval        time.Duration
+
+	starter sync.Once
+	// After `Start` is called, the following fields will be initialized
+	stopCtx     context.Context
+	bgJobCancel context.CancelFunc
+}
+
+// ClusterMetaDataManipulator provides the snapshot for NodeInspector to check and utilities of drop expired shard nodes.
+type ClusterMetaDataManipulator interface {
+	GetClusterSnapshot() metadata.Snapshot
+	DropShardNodes(context.Context, []storage.ShardNode) error
+}
+
+func NewNodeInspectorWithInterval(logger *zap.Logger, clusterMetadata ClusterMetaDataManipulator, inspectInterval time.Duration) *NodeInspector {
+	return &NodeInspector{
+		logger:          logger,
+		clusterMetadata: clusterMetadata,
+		interval:        inspectInterval,
+		starter:         sync.Once{},
+		stopCtx:         nil,
+		bgJobCancel:     nil,
+	}
+}
+
+func NewNodeInspector(logger *zap.Logger, clusterMetadata ClusterMetaDataManipulator) *NodeInspector {
+	return NewNodeInspectorWithInterval(logger, clusterMetadata, defaultInspectInterval)
+}
+
+func (ni *NodeInspector) Start(ctx context.Context) error {
+	started := false
+	ni.starter.Do(func() {
+		log.Info("node inspector start")
+		started = true
+		ni.stopCtx, ni.bgJobCancel = context.WithCancel(ctx)
+		go func() {
+			for {
+				t := time.NewTimer(ni.interval)
+				select {
+				case <-ni.stopCtx.Done():
+					ni.logger.Info("node inspector is stopped, cancel the bg inspecting")
+					if !t.Stop() {
+						<-t.C
+					}
+					return
+				case <-t.C:
+				}
+
+				ni.inspect(ctx)
+			}
+		}()
+	})
+
+	if !started {
+		return ErrStartAgain
+	}
+
+	return nil
+}
+
+func (ni *NodeInspector) Stop(_ context.Context) error {
+	if ni.bgJobCancel != nil {
+		ni.bgJobCancel()
+		return nil
+	}
+
+	return ErrStopNotStart
+}
+
+func (ni *NodeInspector) inspect(ctx context.Context) {
+	// Get latest cluster snapshot.
+	snapshot := ni.clusterMetadata.GetClusterSnapshot()
+	expiredShardNodes := findExpiredShardNodes(snapshot)
+	if len(expiredShardNodes) == 0 {
+		return
+	}
+
+	// Try to remove useless data if it exists.
+	if err := ni.clusterMetadata.DropShardNodes(ctx, expiredShardNodes); err != nil {
+		log.Error("drop shard node failed", zap.Error(err))
+	}
+}
+
+func findExpiredShardNodes(snapshot metadata.Snapshot) []storage.ShardNode {
+	// In most cases, there is no expired shard nodes so don't pre-allocate the memory here.
+	expiredNodes := make(map[string]struct{}, 0)
+	// Check node status.
+	now := time.Now()
+	for i := range snapshot.RegisteredNodes {
+		node := &snapshot.RegisteredNodes[i]
+		if node.IsExpired(now) {
+			expiredNodes[node.Node.Name] = struct{}{}
+		}
+	}
+
+	expiredShardNodes := make([]storage.ShardNode, 0, len(expiredNodes))
+	for _, shardNode := range snapshot.Topology.ClusterView.ShardNodes {
+		_, ok := expiredNodes[shardNode.NodeName]
+		if ok {
+			expiredShardNodes = append(expiredShardNodes, shardNode)
+		}
+	}
+
+	return expiredShardNodes
+}

horaemeta/server/coordinator/inspector/node_inspector_test.go

@@ -0,0 +1,145 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you 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 inspector
+
+import (
+	"context"
+	"slices"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/apache/incubator-horaedb-meta/server/cluster/metadata"
+	"github.com/apache/incubator-horaedb-meta/server/storage"
+	"github.com/stretchr/testify/assert"
+	"go.uber.org/zap"
+)
+
+type mockClusterMetaDataManipulator struct {
+	snapshot          metadata.Snapshot
+	lock              sync.Mutex
+	droppedShardNodes [][]storage.ShardNode
+}
+
+func newMockClusterMetaDataManipulator(shardNodes []storage.ShardNode, registeredNodes []metadata.RegisteredNode) *mockClusterMetaDataManipulator {
+	var clusterView storage.ClusterView
+	clusterView.ShardNodes = shardNodes
+	topology := metadata.Topology{
+		ShardViewsMapping: nil,
+		ClusterView:       clusterView,
+	}
+
+	snapshot := metadata.Snapshot{
+		Topology:        topology,
+		RegisteredNodes: registeredNodes,
+	}
+	return &mockClusterMetaDataManipulator{
+		snapshot:          snapshot,
+		lock:              sync.Mutex{},
+		droppedShardNodes: make([][]storage.ShardNode, 0),
+	}
+}
+
+func (n *mockClusterMetaDataManipulator) GetClusterSnapshot() metadata.Snapshot {
+	return n.snapshot
+}
+
+func (n *mockClusterMetaDataManipulator) DropShardNodes(_ context.Context, shardNodes []storage.ShardNode) error {
+	n.lock.Lock()
+	defer n.lock.Unlock()
+
+	n.droppedShardNodes = append(n.droppedShardNodes, shardNodes)
+	newShardNodes := make([]storage.ShardNode, 0, 2)
+	for _, node := range n.snapshot.Topology.ClusterView.ShardNodes {
+		dropped := slices.ContainsFunc(shardNodes, func(droppedNode storage.ShardNode) bool {
+			return node.NodeName == droppedNode.NodeName
+		})
+		if !dropped {
+			newShardNodes = append(newShardNodes, node)
+		}
+	}
+	n.snapshot.Topology.ClusterView.ShardNodes = newShardNodes
+	return nil
+}
+
+func (n *mockClusterMetaDataManipulator) CheckDroppedShardNodes(check func(droppedShardNodes [][]storage.ShardNode)) {
+	n.lock.Lock()
+	defer n.lock.Unlock()
+
+	check(n.droppedShardNodes)
+}
+
+func TestStartStopInspector(t *testing.T) {
+	inspector := NewNodeInspector(zap.NewNop(), newMockClusterMetaDataManipulator(nil, nil))
+
+	ctx := context.Background()
+	assert.NoError(t, inspector.Start(ctx))
+	assert.Error(t, inspector.Start(ctx))
+
+	assert.NoError(t, inspector.Stop(ctx))
+}
+
+func TestInspect(t *testing.T) {
+	shardNodes := []storage.ShardNode{
+		{ID: storage.ShardID(0), ShardRole: storage.ShardRoleLeader, NodeName: "192.168.1.102"},
+		{ID: storage.ShardID(1), ShardRole: storage.ShardRoleLeader, NodeName: "192.168.1.102"},
+		{ID: storage.ShardID(2), ShardRole: storage.ShardRoleLeader, NodeName: "192.168.1.103"},
+		{ID: storage.ShardID(3), ShardRole: storage.ShardRoleLeader, NodeName: "192.168.1.103"},
+	}
+	registeredNodes := []metadata.RegisteredNode{
+		{
+			Node: storage.Node{
+				Name:          "192.168.1.102",
+				NodeStats:     storage.NodeStats{Lease: 0, Zone: "", NodeVersion: ""},
+				LastTouchTime: uint64(time.Now().UnixMilli()),
+				State:         storage.NodeStateOnline,
+			},
+			ShardInfos: nil,
+		},
+		{
+			// This node should be outdated.
+			Node: storage.Node{
+				Name:          "192.168.1.103",
+				NodeStats:     storage.NodeStats{Lease: 0, Zone: "", NodeVersion: ""},
+				LastTouchTime: uint64(time.Now().UnixMilli()) - uint64((time.Second * 20)),
+				State:         storage.NodeStateOnline,
+			},
+			ShardInfos: nil,
+		},
+	}
+
+	metadata := newMockClusterMetaDataManipulator(shardNodes, registeredNodes)
+	inspector := NewNodeInspectorWithInterval(zap.NewNop(), metadata, time.Millisecond*100)
+	ctx := context.Background()
+	assert.NoError(t, inspector.Start(ctx))
+
+	// The inspect should be triggered after 200ms.
+	time.Sleep(time.Millisecond * 200)
+
+	// The outdated node should be removed by triggered.
+	metadata.CheckDroppedShardNodes(func(droppedShardNodes [][]storage.ShardNode) {
+		assert.True(t, len(droppedShardNodes) == 1)
+		assert.True(t, len(droppedShardNodes[0]) == 2)
+		assert.Equal(t, droppedShardNodes[0][0], shardNodes[2])
+		assert.Equal(t, droppedShardNodes[0][1], shardNodes[3])
+	})
+
+	assert.NoError(t, inspector.Stop(ctx))
+}

horaemeta/server/coordinator/scheduler/manager/scheduler_manager.go

@@ -203,7 +203,7 @@ func (callback *schedulerWatchCallback) OnShardRegistered(_ context.Context, _ w
 func (callback *schedulerWatchCallback) OnShardExpired(ctx context.Context, event watch.ShardExpireEvent) error {
 	oldLeader := event.OldLeaderNode
 	shardID := event.ShardID
-	return callback.c.DropShardNode(ctx, []storage.ShardNode{
+	return callback.c.DropShardNodes(ctx, []storage.ShardNode{
 		{
 			ID:        shardID,
 			ShardRole: storage.ShardRoleLeader,

horaemeta/server/coordinator/shard_picker_test.go

@@ -83,7 +83,7 @@ func TestLeastTableShardPicker(t *testing.T) {
 	// drop shard node 1, shard 1 should not be picked.
 	for _, shardNode := range snapshot.Topology.ClusterView.ShardNodes {
 		if shardNode.ID == 1 {
-			err = c.GetMetadata().DropShardNode(ctx, []storage.ShardNode{shardNode})
+			err = c.GetMetadata().DropShardNodes(ctx, []storage.ShardNode{shardNode})
 			re.NoError(err)
 		}
 	}

horaemeta/server/service/http/api.go

@@ -230,7 +230,7 @@ func (a *API) dropNodeShards(req *http.Request) apiFuncResult {
 		}
 	}
 
-	if err := c.GetMetadata().DropShardNode(req.Context(), targetShardNodes); err != nil {
+	if err := c.GetMetadata().DropShardNodes(req.Context(), targetShardNodes); err != nil {
 		log.Error("drop node shards failed", zap.Error(err))
 		return errResult(ErrDropNodeShards, err.Error())
 	}