Merge pull request #19063 from taosdata/fix/TD-21424_TD-21389_TD-21420

enh: add dropping and creating status for dnode

include/util/taoserror.h

@@ -203,9 +203,15 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_CANT_DROP_LEADER          TAOS_DEF_ERROR_CODE(0, 0x0333)
 #define TSDB_CODE_MND_NO_ENOUGH_DNODES          TAOS_DEF_ERROR_CODE(0, 0x0334)
 #define TSDB_CODE_MND_INVALID_CLUSTER_CFG       TAOS_DEF_ERROR_CODE(0, 0x0335)
+// #define TSDB_CODE_MND_INVALID_DNODE_CFG_...  TAOS_DEF_ERROR_CODE(0, 0x0336)  // 2.x
+// #define TSDB_CODE_MND_BALANCE_ENABLED        TAOS_DEF_ERROR_CODE(0, 0x0337)  // 2.x
 #define TSDB_CODE_MND_VGROUP_NOT_IN_DNODE       TAOS_DEF_ERROR_CODE(0, 0x0338)
 #define TSDB_CODE_MND_VGROUP_ALREADY_IN_DNODE   TAOS_DEF_ERROR_CODE(0, 0x0339)
+// #define TSDB_CODE_MND_DNODE_NOT_FREE         TAOS_DEF_ERROR_CODE(0, 0x033A)  // 2.x
 #define TSDB_CODE_MND_INVALID_CLUSTER_ID        TAOS_DEF_ERROR_CODE(0, 0x033B)
+// #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
 
 // mnode-acct
 #define TSDB_CODE_MND_ACCT_ALREADY_EXIST        TAOS_DEF_ERROR_CODE(0, 0x0340)
@@ -297,6 +303,8 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_VGROUP_UN_CHANGED         TAOS_DEF_ERROR_CODE(0, 0x03B5)
 #define TSDB_CODE_MND_HAS_OFFLINE_DNODE         TAOS_DEF_ERROR_CODE(0, 0x03B6)
 #define TSDB_CODE_MND_INVALID_REPLICA           TAOS_DEF_ERROR_CODE(0, 0x03B7)
+#define TSDB_CODE_MND_DNODE_IN_CREATING         TAOS_DEF_ERROR_CODE(0, 0x03B8)
+#define TSDB_CODE_MND_DNODE_IN_DROPPING         TAOS_DEF_ERROR_CODE(0, 0x03B9)
 
 // mnode-stable-part2
 #define TSDB_CODE_MND_NAME_CONFLICT_WITH_TOPIC  TAOS_DEF_ERROR_CODE(0, 0x03C0)

source/dnode/mgmt/mgmt_dnode/src/dmHandle.c

@@ -41,6 +41,8 @@ static void dmProcessStatusRsp(SDnodeMgmt *pMgmt, SRpcMsg *pRsp) {
              pMgmt->statusSeq);
       pMgmt->pData->dropped = 1;
       dmWriteEps(pMgmt->pData);
+      dInfo("dnode will exit since it is in the dropped state");
+      raise(SIGINT);
     }
   } else {
     SStatusRsp statusRsp = {0};

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

@@ -217,8 +217,18 @@ SDnodeObj *mndAcquireDnode(SMnode *pMnode, int32_t dnodeId) {
   SSdb      *pSdb = pMnode->pSdb;
   SDnodeObj *pDnode = sdbAcquire(pSdb, SDB_DNODE, &dnodeId);
   if (pDnode == NULL) {
+    if (terrno == TSDB_CODE_SDB_OBJ_NOT_THERE) {
       terrno = TSDB_CODE_MND_DNODE_NOT_EXIST;
+    } else if (terrno == TSDB_CODE_SDB_OBJ_CREATING) {
+      terrno = TSDB_CODE_MND_DNODE_IN_CREATING;
+    } else if (terrno == TSDB_CODE_SDB_OBJ_DROPPING) {
+      terrno = TSDB_CODE_MND_DNODE_IN_DROPPING;
+    } else {
+      terrno = TSDB_CODE_APP_ERROR;
+      mFatal("dnode:%d, failed to acquire db since %s", dnodeId, terrstr());
+    }
   }
+
   return pDnode;
 }
 
@@ -254,6 +264,27 @@ static SDnodeObj *mndAcquireDnodeByEp(SMnode *pMnode, char *pEpStr) {
   return NULL;
 }
 
+static SDnodeObj *mndAcquireDnodeAllStatusByEp(SMnode *pMnode, char *pEpStr) {
+  SSdb *pSdb = pMnode->pSdb;
+
+  void *pIter = NULL;
+  while (1) {
+    SDnodeObj *pDnode = NULL;
+    ESdbStatus objStatus = 0;
+    pIter = sdbFetchAll(pSdb, SDB_DNODE, pIter, (void **)&pDnode, &objStatus, true);
+    if (pIter == NULL) break;
+
+    if (strncasecmp(pEpStr, pDnode->ep, TSDB_EP_LEN) == 0) {
+      sdbCancelFetch(pSdb, pIter);
+      return pDnode;
+    }
+
+    sdbRelease(pSdb, pDnode);
+  }
+
+  return NULL;
+}
+
 int32_t mndGetDnodeSize(SMnode *pMnode) {
   SSdb *pSdb = pMnode->pSdb;
   return sdbGetSize(pSdb, SDB_DNODE);
@@ -340,12 +371,22 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
   } else {
     pDnode = mndAcquireDnode(pMnode, statusReq.dnodeId);
     if (pDnode == NULL) {
+      int32_t err = terrno;
       pDnode = mndAcquireDnodeByEp(pMnode, statusReq.dnodeEp);
       if (pDnode != NULL) {
         pDnode->offlineReason = DND_REASON_DNODE_ID_NOT_MATCH;
+        terrno = err;
+        goto _OVER;
       }
-      mError("dnode:%d, %s not exist", statusReq.dnodeId, statusReq.dnodeEp);
+
+      mError("dnode:%d, %s not exist, code:0x%x", statusReq.dnodeId, statusReq.dnodeEp, err);
+      if (err == TSDB_CODE_MND_DNODE_NOT_EXIST) {
+        terrno = err;
         goto _OVER;
+      } else {
+        pDnode = mndAcquireDnodeAllStatusByEp(pMnode, statusReq.dnodeEp);
+        if (pDnode == NULL) goto _OVER;
+      }
     }
   }
 
@@ -967,11 +1008,12 @@ static int32_t mndRetrieveDnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
   SSdb      *pSdb = pMnode->pSdb;
   int32_t    numOfRows = 0;
   int32_t    cols = 0;
+  ESdbStatus objStatus = 0;
   SDnodeObj *pDnode = NULL;
   int64_t    curMs = taosGetTimestampMs();
 
   while (numOfRows < rows) {
-    pShow->pIter = sdbFetch(pSdb, SDB_DNODE, pShow->pIter, (void **)&pDnode);
+    pShow->pIter = sdbFetchAll(pSdb, SDB_DNODE, pShow->pIter, (void **)&pDnode, &objStatus, true);
     if (pShow->pIter == NULL) break;
     bool online = mndIsDnodeOnline(pDnode, curMs);
 
@@ -993,8 +1035,20 @@ static int32_t mndRetrieveDnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)&pDnode->numOfSupportVnodes, false);
 
+    const char *status = "ready";
+    if (objStatus == SDB_STATUS_CREATING) status = "creating";
+    if (objStatus == SDB_STATUS_DROPPING) status = "dropping";
+    if (!online) {
+      if (objStatus == SDB_STATUS_CREATING)
+        status = "creating*";
+      else if (objStatus == SDB_STATUS_DROPPING)
+        status = "dropping*";
+      else
+        status = "offline";
+    }
+
     char b1[9] = {0};
-    STR_TO_VARSTR(b1, online ? "ready" : "offline");
+    STR_TO_VARSTR(b1, status);
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, b1, false);
 

source/util/src/terror.c

@@ -248,6 +248,8 @@ TAOS_DEFINE_ERROR(TSDB_CODE_MND_SINGLE_STB_MODE_DB,       "Database is single st
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_INVALID_SCHEMA_VER,       "Invalid schema version while alter stb")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_STABLE_UID_NOT_MATCH,     "Invalid stable uid while alter stb")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_FIELD_CONFLICT_WITH_TSMA, "Field used by tsma")
+TAOS_DEFINE_ERROR(TSDB_CODE_MND_DNODE_IN_CREATING,        "Dnode in creating status")
+TAOS_DEFINE_ERROR(TSDB_CODE_MND_DNODE_IN_DROPPING,        "Dnode in dropping status")
 
 // mnode-trans
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_TRANS_ALREADY_EXIST,      "Transaction already exists")