Merge pull request #6442 from taosdata/fix/TD-4599

[TD-4599]<fix>: fix false dnode offline

src/balance/src/bnMain.c

@@ -220,10 +220,6 @@ int32_t bnAllocVnodes(SVgObj *pVgroup) {
 }
 
 static bool bnCheckVgroupReady(SVgObj *pVgroup, SVnodeGid *pRmVnode) {
-  if (pVgroup->lbTime + 5 * tsStatusInterval > tsAccessSquence) {
-    return false;
-  }
-
   int32_t rmVnodeVer = 0;
   for (int32_t i = 0; i < pVgroup->numOfVnodes; ++i) {
     SVnodeGid *pVnode = pVgroup->vnodeGid + i;
@@ -405,7 +401,7 @@ void bnReset() {
     if (pDnode == NULL) break;
 
     // while master change, should reset dnode to offline
-    mInfo("dnode:%d set access:%d to 0", pDnode->dnodeId, pDnode->lastAccess);
+    mInfo("dnode:%d set access:%" PRId64 " to 0", pDnode->dnodeId, pDnode->lastAccess);
     pDnode->lastAccess = 0;
     if (pDnode->status != TAOS_DN_STATUS_DROPPING) {
       pDnode->status = TAOS_DN_STATUS_OFFLINE;
@@ -499,7 +495,7 @@ static bool bnMontiorDropping() {
       if (dnodeIsMasterEp(pDnode->dnodeEp)) continue; 
       if (mnodeGetDnodesNum() <= 1) continue;
 
-      mLInfo("dnode:%d, set to removing state for it offline:%d seconds", pDnode->dnodeId,
+      mLInfo("dnode:%d, set to removing state for it offline:%" PRId64 " seconds", pDnode->dnodeId,
               tsAccessSquence - pDnode->lastAccess);
 
       pDnode->status = TAOS_DN_STATUS_DROPPING;
@@ -574,8 +570,8 @@ void bnCheckStatus() {
       if (pDnode->status != TAOS_DN_STATUS_DROPPING && pDnode->status != TAOS_DN_STATUS_OFFLINE) {
         pDnode->status = TAOS_DN_STATUS_OFFLINE;
         pDnode->offlineReason = TAOS_DN_OFF_STATUS_MSG_TIMEOUT;
-        mInfo("dnode:%d, set to offline state, access seq:%d last seq:%d laststat:%d", pDnode->dnodeId, tsAccessSquence,
-              pDnode->lastAccess, pDnode->status);
+        mInfo("dnode:%d, set to offline state, access seq:%" PRId64 " last seq:%" PRId64 " laststat:%d", pDnode->dnodeId,
+              tsAccessSquence, pDnode->lastAccess, pDnode->status);
         bnSetVgroupOffline(pDnode);
         bnStartTimer(3000);
       }

src/balance/src/bnThread.c

@@ -102,12 +102,12 @@ static void bnProcessTimer(void *handle, void *tmrId) {
   if (tsBnThread.stop) return;
 
   tsBnThread.timer = NULL;
-  tsAccessSquence++;
-
   bnStartTimer(-1);
   bnCheckStatus();
 
   if (handle == NULL) {
+    ++tsAccessSquence;
+
     if (tsAccessSquence % tsBalanceInterval == 0) {
       mDebug("balance function is scheduled by timer");
       bnPostSignal();
@@ -122,8 +122,7 @@ static void bnProcessTimer(void *handle, void *tmrId) {
 void bnStartTimer(int32_t mseconds) {
   if (tsBnThread.stop) return;
 
-  bool updateSoon = (mseconds != -1);
-  if (updateSoon) {
+  if (mseconds != -1) {
     mTrace("balance function will be called after %d ms", mseconds);
     taosTmrReset(bnProcessTimer, mseconds, (void *)(int64_t)mseconds, tsMnodeTmr, &tsBnThread.timer);
   } else {
@@ -132,5 +131,5 @@ void bnStartTimer(int32_t mseconds) {
 }
 
 void bnNotify() {
-  bnStartTimer(500); 
+  bnStartTimer(500);
 }

src/mnode/inc/mnodeDef.h

@@ -48,9 +48,8 @@ typedef struct SDnodeObj {
   int32_t    dnodeId;
   int32_t    openVnodes;
   int64_t    createdTime;
-  int32_t    resever0;         // from dnode status msg, config information
+  int64_t    lastAccess;
   int32_t    customScore;      // config by user
-  uint32_t   lastAccess;
   uint16_t   numOfCores;       // from dnode status msg
   uint16_t   dnodePort;
   char       dnodeFqdn[TSDB_FQDN_LEN];

src/mnode/inc/mnodeDnode.h

@@ -77,7 +77,7 @@ void *  mnodeGetDnodeByEp(char *ep);
 void    mnodeUpdateDnode(SDnodeObj *pDnode);
 int32_t mnodeDropDnode(SDnodeObj *pDnode, void *pMsg);
 
-extern int32_t tsAccessSquence;
+extern int64_t tsAccessSquence;
 
 #ifdef __cplusplus
 }

src/mnode/src/mnodeDnode.c

@@ -39,8 +39,8 @@
 #include "mnodePeer.h"
 #include "mnodeCluster.h"
 
-int32_t tsAccessSquence = 0;
-int64_t         tsDnodeRid = -1;
+int64_t        tsAccessSquence = 0;
+int64_t        tsDnodeRid = -1;
 static void *  tsDnodeSdb = NULL;
 static int32_t tsDnodeUpdateSize = 0;
 extern void *  tsMnodeSdb;
@@ -567,7 +567,7 @@ static int32_t mnodeProcessDnodeStatusMsg(SMnodeMsg *pMsg) {
              mnodeGetClusterId());
       return TSDB_CODE_MND_INVALID_CLUSTER_ID;
     } else {
-      mTrace("dnode:%d, status received, access times %d openVnodes:%d:%d", pDnode->dnodeId, pDnode->lastAccess,
+      mTrace("dnode:%d, status received, access times %" PRId64 " openVnodes:%d:%d", pDnode->dnodeId, pDnode->lastAccess,
              htons(pStatus->openVnodes), pDnode->openVnodes);
     }
   }
@@ -629,9 +629,9 @@ static int32_t mnodeProcessDnodeStatusMsg(SMnodeMsg *pMsg) {
     bnNotify();
   }
 
-  if (!tsEnableBalance) {
-    int32_t numOfMnodes = mnodeGetMnodesNum();
-    if (numOfMnodes < tsNumOfMnodes) bnNotify();
+  int32_t numOfMnodes = mnodeGetMnodesNum();
+  if (numOfMnodes < tsNumOfMnodes && numOfMnodes < mnodeGetOnlineDnodesNum() && !pDnode->isMgmt) {
+    bnNotify();
   }
 
   if (openVnodes != pDnode->openVnodes) {