Merge pull request #20524 from taosdata/fix/clusterid-isolation

fix: clusterid isolation

include/util/taoserror.h

@@ -223,6 +223,7 @@ int32_t* taosGetErrno();
 // #define TSDB_CODE_MND_NOT_READY              TAOS_DEF_ERROR_CODE(0, 0x033C)  // 2.x
 // #define TSDB_CODE_MND_DNODE_ID_NOT_CONFIGUREDTAOS_DEF_ERROR_CODE(0, 0x033D)  // 2.x
 // #define TSDB_CODE_MND_DNODE_EP_NOT_CONFIGUREDTAOS_DEF_ERROR_CODE(0, 0x033E)  // 2.x
+#define TSDB_CODE_MND_DNODE_DIFF_CLUSTER        TAOS_DEF_ERROR_CODE(0, 0x033F) // internal
 
 // mnode-acct
 #define TSDB_CODE_MND_ACCT_ALREADY_EXIST        TAOS_DEF_ERROR_CODE(0, 0x0340)

source/dnode/mnode/impl/src/mndDnode.c

@@ -25,6 +25,7 @@
 #include "mndUser.h"
 #include "mndVgroup.h"
 #include "tmisce.h"
+#include "mndCluster.h"
 
 #define TSDB_DNODE_VER_NUMBER   1
 #define TSDB_DNODE_RESERVE_SIZE 64
@@ -366,6 +367,14 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
+  int64_t clusterid = mndGetClusterId(pMnode);
+  if (statusReq.clusterId != 0 && statusReq.clusterId != clusterid) {
+    code = TSDB_CODE_MND_DNODE_DIFF_CLUSTER;
+    mWarn("dnode:%d, %s, its clusterid:%" PRId64 " differ from current cluster:%" PRId64 ", code:0x%x",
+          statusReq.dnodeId, statusReq.dnodeEp, statusReq.clusterId, clusterid, code);
+    goto _OVER;
+  }
+
   if (statusReq.dnodeId == 0) {
     pDnode = mndAcquireDnodeByEp(pMnode, statusReq.dnodeEp);
     if (pDnode == NULL) {

source/libs/sync/src/syncMain.c

@@ -37,6 +37,7 @@
 #include "syncVoteMgr.h"
 #include "tglobal.h"
 #include "tref.h"
+#include "syncUtil.h"
 
 static void    syncNodeEqPingTimer(void* param, void* tmrId);
 static void    syncNodeEqElectTimer(void* param, void* tmrId);
@@ -2295,6 +2296,14 @@ int32_t syncNodeOnHeartbeat(SSyncNode* ths, const SRpcMsg* pRpcMsg) {
   int64_t timeDiff = tsMs - pMsg->timeStamp;
   syncLogRecvHeartbeat(ths, pMsg, timeDiff, tbuf);
 
+  if (!syncNodeInRaftGroup(ths, &pMsg->srcId)) {
+    sWarn(
+        "vgId:%d, drop heartbeat msg from dnode:%d, because it come from another cluster:%d, differ from current "
+        "cluster:%d",
+        ths->vgId, DID(&(pMsg->srcId)), CID(&(pMsg->srcId)), CID(&(ths->myRaftId)));
+    return 0;
+  }
+
   SRpcMsg rpcMsg = {0};
   (void)syncBuildHeartbeatReply(&rpcMsg, ths->vgId);
   SyncTerm currentTerm = raftStoreGetTerm(ths);

source/libs/sync/src/syncRequestVote.c

@@ -20,6 +20,7 @@
 #include "syncRaftStore.h"
 #include "syncUtil.h"
 #include "syncVoteMgr.h"
+#include "syncUtil.h"
 
 // TLA+ Spec
 // HandleRequestVoteRequest(i, j, m) ==