Improve TopologyService and HeartbeatService scalability for large clusters

integration-test/src/main/java/org/apache/iotdb/it/env/cluster/config/MppCommonConfig.java

@@ -664,6 +664,12 @@ public CommonConfig setMaxRowsInCteBuffer(int maxRows) {
     return this;
   }
 
+  @Override
+  public CommonConfig setEnableTopologyProbing(boolean enableTopologyProbing) {
+    setProperty("enable_topology_probing", String.valueOf(enableTopologyProbing));
+    return this;
+  }
+
   // For part of the log directory
   public String getClusterConfigStr() {
     return fromConsensusFullNameToAbbr(properties.getProperty(CONFIG_NODE_CONSENSUS_PROTOCOL_CLASS))

integration-test/src/main/java/org/apache/iotdb/itbase/env/CommonConfig.java

@@ -215,4 +215,6 @@ default CommonConfig setDefaultDatabaseLevel(int defaultDatabaseLevel) {
   CommonConfig setCteBufferSize(long cteBufferSize);
 
   CommonConfig setMaxRowsInCteBuffer(int maxRows);
+
+  CommonConfig setEnableTopologyProbing(boolean enableTopologyProbing);
 }

integration-test/src/test/java/org/apache/iotdb/relational/it/query/recent/IoTDBTableAggregationQueryWithNetworkPartitionIT.java

@@ -76,7 +76,8 @@ public static void setUp() throws Exception {
         .setSchemaReplicationFactor(testReplicationFactor)
         .setDataReplicationFactor(testReplicationFactor)
         .setTimePartitionInterval(testTimePartitionInterval)
-        .setDefaultDataRegionGroupNumPerDatabase(testDataRegionGroupPerDatabase);
+        .setDefaultDataRegionGroupNumPerDatabase(testDataRegionGroupPerDatabase)
+        .setEnableTopologyProbing(true);
     EnvFactory.getEnv().initClusterEnvironment(1, 3);
     prepareTableData(createSqls);
   }

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/client/async/CnToDnAsyncRequestType.java

@@ -34,6 +34,7 @@ public enum CnToDnAsyncRequestType {
   SUBMIT_TEST_CONNECTION_TASK,
   SUBMIT_TEST_DN_INTERNAL_CONNECTION_TASK,
   TEST_CONNECTION,
+  PUSH_TOPOLOGY,
 
   // Region Maintenance
   CREATE_DATA_REGION,

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/client/async/CnToDnInternalServiceAsyncRequestManager.java

@@ -100,6 +100,7 @@
 import org.apache.iotdb.mpp.rpc.thrift.TTableDeviceDeletionWithPatternAndFilterReq;
 import org.apache.iotdb.mpp.rpc.thrift.TTableDeviceDeletionWithPatternOrModReq;
 import org.apache.iotdb.mpp.rpc.thrift.TTableDeviceInvalidateCacheReq;
+import org.apache.iotdb.mpp.rpc.thrift.TUpdateClusterTopologyReq;
 import org.apache.iotdb.mpp.rpc.thrift.TUpdateTableReq;
 import org.apache.iotdb.mpp.rpc.thrift.TUpdateTemplateReq;
 import org.apache.iotdb.mpp.rpc.thrift.TUpdateTriggerLocationReq;
@@ -407,6 +408,11 @@ protected void initActionMapBuilder() {
         CnToDnAsyncRequestType.TEST_CONNECTION,
         (req, client, handler) ->
             client.testConnectionEmptyRPC((DataNodeTSStatusRPCHandler) handler));
+    actionMapBuilder.put(
+        CnToDnAsyncRequestType.PUSH_TOPOLOGY,
+        (req, client, handler) ->
+            client.updateClusterTopology(
+                (TUpdateClusterTopologyReq) req, (DataNodeTSStatusRPCHandler) handler));
     actionMapBuilder.put(
         CnToDnAsyncRequestType.INSERT_RECORD,
         (req, client, handler) ->

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/client/async/handlers/rpc/DataNodeAsyncRequestRPCHandler.java

@@ -202,6 +202,7 @@ public static DataNodeAsyncRequestRPCHandler<?> buildHandler(
             dataNodeLocationMap,
             (Map<Integer, TExternalServiceListResp>) responseMap,
             countDownLatch);
+      case PUSH_TOPOLOGY:
       case SET_TTL:
       case CREATE_DATA_REGION:
       case CREATE_SCHEMA_REGION:

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/conf/ConfigNodeConfig.java

@@ -204,6 +204,15 @@ public class ConfigNodeConfig {
   /** Acceptable pause duration for Phi accrual failure detector */
   private long failureDetectorPhiAcceptablePauseInMs = 10000;
 
+  /** Whether to enable topology probing between DataNodes. Supports hot-reload. */
+  private volatile boolean enableTopologyProbing = false;
+
+  /** Base interval in ms for topology probing. */
+  private long topologyProbingBaseIntervalInMs = 5000;
+
+  /** Ratio of probing timeout to probing interval (must be less than 1.0). */
+  private double topologyProbingTimeoutRatio = 0.5;
+
   /** The policy of cluster RegionGroups' leader distribution. */
   private String leaderDistributionPolicy = AbstractLeaderBalancer.CFD_POLICY;
 
@@ -1288,4 +1297,28 @@ public long getFailureDetectorPhiAcceptablePauseInMs() {
   public void setFailureDetectorPhiAcceptablePauseInMs(long failureDetectorPhiAcceptablePauseInMs) {
     this.failureDetectorPhiAcceptablePauseInMs = failureDetectorPhiAcceptablePauseInMs;
   }
+
+  public boolean isEnableTopologyProbing() {
+    return enableTopologyProbing;
+  }
+
+  public void setEnableTopologyProbing(boolean enableTopologyProbing) {
+    this.enableTopologyProbing = enableTopologyProbing;
+  }
+
+  public long getTopologyProbingBaseIntervalInMs() {
+    return topologyProbingBaseIntervalInMs;
+  }
+
+  public void setTopologyProbingBaseIntervalInMs(long topologyProbingBaseIntervalInMs) {
+    this.topologyProbingBaseIntervalInMs = topologyProbingBaseIntervalInMs;
+  }
+
+  public double getTopologyProbingTimeoutRatio() {
+    return topologyProbingTimeoutRatio;
+  }
+
+  public void setTopologyProbingTimeoutRatio(double topologyProbingTimeoutRatio) {
+    this.topologyProbingTimeoutRatio = topologyProbingTimeoutRatio;
+  }
 }

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/conf/ConfigNodeDescriptor.java

@@ -322,6 +322,23 @@ private void loadProperties(TrimProperties properties) throws BadNodeUrlExceptio
                 "failure_detector_phi_acceptable_pause_in_ms",
                 String.valueOf(conf.getFailureDetectorPhiAcceptablePauseInMs()))));
 
+    conf.setEnableTopologyProbing(
+        Boolean.parseBoolean(
+            properties.getProperty(
+                "enable_topology_probing", String.valueOf(conf.isEnableTopologyProbing()))));
+
+    conf.setTopologyProbingBaseIntervalInMs(
+        Long.parseLong(
+            properties.getProperty(
+                "topology_probing_base_interval_in_ms",
+                String.valueOf(conf.getTopologyProbingBaseIntervalInMs()))));
+
+    conf.setTopologyProbingTimeoutRatio(
+        Double.parseDouble(
+            properties.getProperty(
+                "topology_probing_timeout_ratio",
+                String.valueOf(conf.getTopologyProbingTimeoutRatio()))));
+
     String leaderDistributionPolicy =
         properties.getProperty("leader_distribution_policy", conf.getLeaderDistributionPolicy());
     if (AbstractLeaderBalancer.GREEDY_POLICY.equals(leaderDistributionPolicy)
@@ -773,6 +790,8 @@ public void loadHotModifiedProps(TrimProperties properties) {
     ConfigurationFileUtils.updateAppliedProperties(properties, true);
     Optional.ofNullable(properties.getProperty(IoTDBConstant.CLUSTER_NAME))
         .ifPresent(conf::setClusterName);
+    Optional.ofNullable(properties.getProperty("enable_topology_probing"))
+        .ifPresent(v -> conf.setEnableTopologyProbing(Boolean.parseBoolean(v)));
   }
 
   public static ConfigNodeDescriptor getInstance() {

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/consensus/statemachine/ConfigRegionStateMachine.java

@@ -256,6 +256,7 @@ public void notifyNotLeader() {
     configManager.getPipeManager().getPipeRuntimeCoordinator().stopPipeMetaSync();
     configManager.getPipeManager().getPipeRuntimeCoordinator().stopPipeHeartbeat();
     configManager.getSubscriptionManager().getSubscriptionCoordinator().stopSubscriptionMetaSync();
+    configManager.getLoadManager().stopTopologyService();
     configManager.getLoadManager().stopLoadServices();
     configManager.getProcedureManager().stopExecutor();
     configManager.getRetryFailedTasksThread().stopRetryFailedTasksService();
@@ -288,6 +289,10 @@ public void notifyLeaderReady() {
     // Always start load services first
     configManager.getLoadManager().startLoadServices();
 
+    if (CONF.isEnableTopologyProbing()) {
+      configManager.getLoadManager().startTopologyService();
+    }
+
     // Start leader scheduling services
     configManager.getProcedureManager().startExecutor();
     threadPool.submit(

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/manager/ConfigManager.java

@@ -1744,6 +1744,7 @@ public TSStatus setConfiguration(TSetConfigurationReq req) {
       TrimProperties properties = new TrimProperties();
       properties.putAll(req.getConfigs());
 
+      boolean wasTopologyProbingEnabled = CONF.isEnableTopologyProbing();
       if (configurationFileFound) {
         File file = new File(url.getFile());
         try {
@@ -1767,6 +1768,7 @@ public TSStatus setConfiguration(TSetConfigurationReq req) {
         }
         LOGGER.warn(msg);
       }
+      handleTopologyProbingHotReload(wasTopologyProbingEnabled);
       if (currentNodeId == req.getNodeId()) {
         return tsStatus;
       }
@@ -1778,6 +1780,18 @@ public TSStatus setConfiguration(TSetConfigurationReq req) {
     return RpcUtils.squashResponseStatusList(statusList);
   }
 
+  private void handleTopologyProbingHotReload(boolean wasEnabled) {
+    boolean isEnabled = CONF.isEnableTopologyProbing();
+    if (wasEnabled == isEnabled) {
+      return;
+    }
+    if (isEnabled && getConsensusManager().isLeader()) {
+      getLoadManager().startTopologyService();
+    } else if (!isEnabled) {
+      getLoadManager().stopTopologyService();
+    }
+  }
+
   @Override
   public TSStatus startRepairData() {
     TSStatus status = confirmLeader();

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/manager/load/LoadManager.java

@@ -152,7 +152,6 @@ public void startLoadServices() {
     statisticsService.startLoadStatisticsService();
     eventService.startEventService();
     partitionBalancer.setupPartitionBalancer();
-    topologyService.startTopologyService();
   }
 
   public void stopLoadServices() {
@@ -162,6 +161,13 @@ public void stopLoadServices() {
     loadCache.clearHeartbeatCache();
     partitionBalancer.clearPartitionBalancer();
     routeBalancer.clearRegionPriority();
+  }
+
+  public void startTopologyService() {
+    topologyService.startTopologyService();
+  }
+
+  public void stopTopologyService() {
     topologyService.stopTopologyService();
   }
 

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/manager/load/cache/LoadCache.java

@@ -787,7 +787,7 @@ public void updateTopology(Map<Integer, Set<Integer>> latestTopology) {
       for (int fromId : latestTopology.keySet()) {
         for (int toId : latestTopology.keySet()) {
           boolean originReachable =
-              latestTopology.getOrDefault(fromId, Collections.emptySet()).contains(toId);
+              topologyGraph.getOrDefault(fromId, Collections.emptySet()).contains(toId);
           boolean newReachable =
               latestTopology.getOrDefault(fromId, Collections.emptySet()).contains(toId);
           if (originReachable != newReachable) {

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/manager/load/service/HeartbeatService.java

@@ -49,7 +49,6 @@
 import org.slf4j.LoggerFactory;
 
 import java.util.List;
-import java.util.Map;
 import java.util.Optional;
 import java.util.Set;
 import java.util.concurrent.Future;
@@ -168,13 +167,6 @@ protected TDataNodeHeartbeatReq genHeartbeatReq() {
       heartbeatReq.setSpaceQuotaUsage(configManager.getClusterQuotaManager().getSpaceQuotaUsage());
     }
 
-    final Map<Integer, Set<Integer>> topologyMap =
-        configManager.getLoadManager().getLoadCache().getTopology();
-    if (topologyMap != null) {
-      heartbeatReq.setTopology(topologyMap);
-      heartbeatReq.setDataNodes(configManager.getNodeManager().getRegisteredDataNodeLocations());
-    }
-
     // We broadcast region operations list every 100 heartbeat loops
     if (heartbeatCounter.get() % 100 == 0) {
       heartbeatReq.setCurrentRegionOperations(