Enhance "remove region" sql to handle the case of DataNode not exist (apache#15728)

liyuheng55555 · web-flow · commit 9d474682cf58 · 2025-06-17T09:50:09.000+08:00
diff --git a/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/manager/ProcedureManager.java b/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/manager/ProcedureManager.java
@@ -24,6 +24,7 @@
 import org.apache.iotdb.common.rpc.thrift.TConsensusGroupType;
 import org.apache.iotdb.common.rpc.thrift.TDataNodeConfiguration;
 import org.apache.iotdb.common.rpc.thrift.TDataNodeLocation;
+import org.apache.iotdb.common.rpc.thrift.TEndPoint;
 import org.apache.iotdb.common.rpc.thrift.TSStatus;
 import org.apache.iotdb.commons.cluster.NodeStatus;
 import org.apache.iotdb.commons.conf.CommonConfig;
@@ -161,6 +162,7 @@
 import org.slf4j.LoggerFactory;
 
 import javax.annotation.Nonnull;
+import javax.annotation.Nullable;
 
 import java.io.IOException;
 import java.nio.ByteBuffer;
@@ -837,15 +839,13 @@ private TSStatus checkExtendRegion(
   private TSStatus checkRemoveRegion(
       TRemoveRegionReq req,
       TConsensusGroupId regionId,
-      TDataNodeLocation targetDataNode,
+      @Nullable TDataNodeLocation targetDataNode,
       TDataNodeLocation coordinator) {
     String failMessage =
         regionOperationCommonCheck(
             regionId,
             targetDataNode,
-            Arrays.asList(
-                new Pair<>("Target DataNode", targetDataNode),
-                new Pair<>("Coordinator", coordinator)),
+            Arrays.asList(new Pair<>("Coordinator", coordinator)),
             req.getModel());
 
     if (configManager
@@ -855,11 +855,12 @@ private TSStatus checkRemoveRegion(
             .getDataNodeLocationsSize()
         == 1) {
       failMessage = String.format("%s only has 1 replica, it cannot be removed", regionId);
-    } else if (configManager
-        .getPartitionManager()
-        .getAllReplicaSets(targetDataNode.getDataNodeId())
-        .stream()
-        .noneMatch(replicaSet -> replicaSet.getRegionId().equals(regionId))) {
+    } else if (targetDataNode != null
+        && configManager
+            .getPartitionManager()
+            .getAllReplicaSets(targetDataNode.getDataNodeId())
+            .stream()
+            .noneMatch(replicaSet -> replicaSet.getRegionId().equals(regionId))) {
       failMessage =
           String.format(
               "Target DataNode %s doesn't contain Region %s", req.getDataNodeId(), regionId);
@@ -1208,6 +1209,23 @@ public TSStatus removeRegion(TRemoveRegionReq req) {
         return status;
       }
 
+      // SPECIAL CASE
+      if (targetDataNode == null) {
+        // If targetDataNode is null, it means the target DataNode does not exist in the
+        // NodeManager.
+        // In this case, simply clean up the partition table once and do nothing else.
+        LOGGER.warn(
+            "Remove region: Target DataNode {} not found, will simply clean up the partition table of region {} and do nothing else.",
+            req.getDataNodeId(),
+            req.getRegionId());
+        this.executor
+            .getEnvironment()
+            .getRegionMaintainHandler()
+            .removeRegionLocation(
+                regionId, buildFakeDataNodeLocation(req.getDataNodeId(), "FakeIpForRemoveRegion"));
+        return new TSStatus(TSStatusCode.SUCCESS_STATUS.getStatusCode());
+      }
+
       // submit procedure
       RemoveRegionPeerProcedure procedure =
           new RemoveRegionPeerProcedure(regionId, coordinator, targetDataNode);
@@ -1219,6 +1237,12 @@ public TSStatus removeRegion(TRemoveRegionReq req) {
     }
   }
 
+  private static TDataNodeLocation buildFakeDataNodeLocation(int dataNodeId, String message) {
+    TEndPoint fakeEndPoint = new TEndPoint(message, -1);
+    return new TDataNodeLocation(
+        dataNodeId, fakeEndPoint, fakeEndPoint, fakeEndPoint, fakeEndPoint, fakeEndPoint);
+  }
+
   // endregion
 
   /**
diff --git a/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/persistence/partition/DatabasePartitionTable.java b/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/persistence/partition/DatabasePartitionTable.java
@@ -539,7 +539,7 @@ void addRegionNewLocation(TConsensusGroupId regionId, TDataNodeLocation node) {
     regionGroup.addRegionLocation(node);
   }
 
-  void removeRegionLocation(TConsensusGroupId regionId, TDataNodeLocation node) {
+  void removeRegionLocation(TConsensusGroupId regionId, int nodeId) {
     RegionGroup regionGroup = regionGroupMap.get(regionId);
     if (regionGroup == null) {
       LOGGER.warn(
@@ -548,16 +548,18 @@ void removeRegionLocation(TConsensusGroupId regionId, TDataNodeLocation node) {
           databaseName);
       return;
     }
-    if (!regionGroup.getReplicaSet().getDataNodeLocations().contains(node)) {
+    if (regionGroup.getReplicaSet().getDataNodeLocations().stream()
+        .map(TDataNodeLocation::getDataNodeId)
+        .noneMatch(id -> id == nodeId)) {
       LOGGER.info(
           "Node is not in region locations when removeRegionOldLocation in {}, "
               + "no need to remove it, node: {}, region: {}",
           databaseName,
-          node,
+          nodeId,
           regionId);
       return;
     }
-    regionGroup.removeRegionLocation(node);
+    regionGroup.removeRegionLocation(nodeId);
   }
 
   /**
diff --git a/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/persistence/partition/PartitionInfo.java b/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/persistence/partition/PartitionInfo.java
@@ -620,7 +620,7 @@ public TSStatus removeRegionLocation(RemoveRegionLocationPlan req) {
         .forEach(
             databasePartitionTable ->
                 databasePartitionTable.removeRegionLocation(
-                    req.getRegionId(), req.getDeprecatedLocation()));
+                    req.getRegionId(), req.getDeprecatedLocation().getDataNodeId()));
     return new TSStatus(TSStatusCode.SUCCESS_STATUS.getStatusCode());
   }
 
diff --git a/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/persistence/partition/RegionGroup.java b/iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/persistence/partition/RegionGroup.java
@@ -95,8 +95,10 @@ public synchronized void addRegionLocation(TDataNodeLocation node) {
     replicaSet.getDataNodeLocations().sort(TDataNodeLocation::compareTo);
   }
 
-  public synchronized void removeRegionLocation(TDataNodeLocation node) {
-    replicaSet.getDataNodeLocations().remove(node);
+  public synchronized void removeRegionLocation(int nodeId) {
+    replicaSet
+        .getDataNodeLocations()
+        .removeIf(tDataNodeLocation -> nodeId == tDataNodeLocation.getDataNodeId());
     replicaSet.getDataNodeLocations().sort(TDataNodeLocation::compareTo);
   }
 

Original file line number	Diff line number	Diff line change
`@@ -620,7 +620,7 @@ public TSStatus removeRegionLocation(RemoveRegionLocationPlan req) {`
`620`	`620`	`.forEach(`
`621`	`621`	`databasePartitionTable ->`
`622`	`622`	`databasePartitionTable.removeRegionLocation(`
`623`		`- req.getRegionId(), req.getDeprecatedLocation()));`
	`623`	`+ req.getRegionId(), req.getDeprecatedLocation().getDataNodeId()));`
`624`	`624`	`return new TSStatus(TSStatusCode.SUCCESS_STATUS.getStatusCode());`
`625`	`625`	`}`
`626`	`626`
Original file line number	Diff line number	Diff line change
`@@ -95,8 +95,10 @@ public synchronized void addRegionLocation(TDataNodeLocation node) {`
`95`	`95`	`replicaSet.getDataNodeLocations().sort(TDataNodeLocation::compareTo);`
`96`	`96`	`}`
`97`	`97`
`98`		`- public synchronized void removeRegionLocation(TDataNodeLocation node) {`
`99`		`- replicaSet.getDataNodeLocations().remove(node);`
	`98`	`+ public synchronized void removeRegionLocation(int nodeId) {`
	`99`	`+ replicaSet`
	`100`	`+ .getDataNodeLocations()`
	`101`	`+ .removeIf(tDataNodeLocation -> nodeId == tDataNodeLocation.getDataNodeId());`
`100`	`102`	`replicaSet.getDataNodeLocations().sort(TDataNodeLocation::compareTo);`
`101`	`103`	`}`
`102`	`104`