Merge pull request #5758 from taosdata/fix/TD-3681

[TD-3681]<fix>: fix syncdb precondition checkings

src/mnode/src/mnodeVgroup.c

@@ -994,6 +994,7 @@ void mnodeSendSyncVgroupMsg(SVgObj *pVgroup) {
   mDebug("vgId:%d, send sync all vnodes msg, numOfVnodes:%d db:%s", pVgroup->vgId, pVgroup->numOfVnodes,
          pVgroup->dbName);
   for (int32_t i = 0; i < pVgroup->numOfVnodes; ++i) {
+    if (pVgroup->vnodeGid[i].role != TAOS_SYNC_ROLE_SLAVE) continue;
     SRpcEpSet epSet = mnodeGetEpSetFromIp(pVgroup->vnodeGid[i].pDnode->dnodeEp);
     mDebug("vgId:%d, index:%d, send sync vnode msg to dnode %s", pVgroup->vgId, i,
            pVgroup->vnodeGid[i].pDnode->dnodeEp);

src/sync/src/syncMain.c

@@ -409,23 +409,22 @@ void syncConfirmForward(int64_t rid, uint64_t version, int32_t code, bool force)
   syncReleaseNode(pNode);
 }
 
-#if 1
 void syncRecover(int64_t rid) {
   SSyncPeer *pPeer;
 
   SSyncNode *pNode = syncAcquireNode(rid);
   if (pNode == NULL) return;
 
-  // to do: add a few lines to check if recover is OK
-  // if take this node to unsync state, the whole system may not work
-
   nodeRole = TAOS_SYNC_ROLE_UNSYNCED;
   (*pNode->notifyRoleFp)(pNode->vgId, nodeRole);
-  nodeVersion = 0;
 
   pthread_mutex_lock(&pNode->mutex);
 
+  nodeVersion = 0;
+
   for (int32_t i = 0; i < pNode->replica; ++i) {
+    if (i == pNode->selfIndex) continue;
+
     pPeer = pNode->peerInfo[i];
     if (pPeer->peerFd >= 0) {
       syncRestartConnection(pPeer);
@@ -436,7 +435,6 @@ void syncRecover(int64_t rid) {
 
   syncReleaseNode(pNode);
 }
-#endif
 
 int32_t syncGetNodesRole(int64_t rid, SNodesRole *pNodesRole) {
   SSyncNode *pNode = syncAcquireNode(rid);

src/vnode/src/vnodeMain.c

@@ -99,8 +99,13 @@ int32_t vnodeSync(int32_t vgId) {
     return TSDB_CODE_VND_INVALID_VGROUP_ID;
   }
 
-  if (pVnode->role != TAOS_SYNC_ROLE_MASTER) {
+  if (pVnode->role == TAOS_SYNC_ROLE_SLAVE) {
     vInfo("vgId:%d, vnode will sync, refCount:%d pVnode:%p", pVnode->vgId, pVnode->refCount, pVnode);
+
+    pVnode->version = 0;
+    pVnode->fversion = 0;
+    walResetVersion(pVnode->wal, pVnode->fversion);
+
     syncRecover(pVnode->sync);
   }