Merge pull request #20300 from taosdata/FIX/TD-23009-main

enh: rotate mnode epSet on sending dnode status req failure

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

@@ -61,6 +61,16 @@ static void dmProcessStatusRsp(SDnodeMgmt *pMgmt, SRpcMsg *pRsp) {
   rpcFreeCont(pRsp->pCont);
 }
 
+void dmEpSetToStr(char *buf, int32_t len, SEpSet *epSet) {
+  int32_t n = 0;
+  n += snprintf(buf + n, len - n, "%s", "{");
+  for (int i = 0; i < epSet->numOfEps; i++) {
+    n += snprintf(buf + n, len - n, "%s:%d%s", epSet->eps[i].fqdn, epSet->eps[i].port,
+                  (i + 1 < epSet->numOfEps ? ", " : ""));
+  }
+  n += snprintf(buf + n, len - n, "%s", "}");
+}
+
 void dmSendStatusReq(SDnodeMgmt *pMgmt) {
   SStatusReq req = {0};
 
@@ -119,11 +129,10 @@ void dmSendStatusReq(SDnodeMgmt *pMgmt) {
   dmGetMnodeEpSet(pMgmt->pData, &epSet);
   rpcSendRecv(pMgmt->msgCb.clientRpc, &epSet, &rpcMsg, &rpcRsp);
   if (rpcRsp.code != 0) {
-    dError("failed to send status req since %s, numOfEps:%d inUse:%d", tstrerror(rpcRsp.code), epSet.numOfEps,
-           epSet.inUse);
-    for (int32_t i = 0; i < epSet.numOfEps; ++i) {
-      dDebug("index:%d, mnode ep:%s:%u", i, epSet.eps[i].fqdn, epSet.eps[i].port);
-    }
+    dmRotateMnodeEpSet(pMgmt->pData);
+    char tbuf[256];
+    dmEpSetToStr(tbuf, sizeof(tbuf), &epSet);
+    dError("failed to send status req since %s, epSet:%s, inUse:%d", tstrerror(rpcRsp.code), tbuf, epSet.inUse);
   }
   dmProcessStatusRsp(pMgmt, &rpcRsp);
 }

source/dnode/mgmt/node_util/inc/dmUtil.h

@@ -166,6 +166,7 @@ int32_t dmReadEps(SDnodeData *pData);
 int32_t dmWriteEps(SDnodeData *pData);
 void    dmUpdateEps(SDnodeData *pData, SArray *pDnodeEps);
 void    dmGetMnodeEpSet(SDnodeData *pData, SEpSet *pEpSet);
+void    dmRotateMnodeEpSet(SDnodeData *pData);
 void    dmGetMnodeEpSetForRedirect(SDnodeData *pData, SRpcMsg *pMsg, SEpSet *pEpSet);
 void    dmSetMnodeEpSet(SDnodeData *pData, SEpSet *pEpSet);
 bool    dmUpdateDnodeInfo(void *pData, int32_t *dnodeId, int64_t *clusterId, char *fqdn, uint16_t *port);

source/dnode/mgmt/node_util/src/dmEps.c

@@ -325,6 +325,28 @@ void dmGetMnodeEpSet(SDnodeData *pData, SEpSet *pEpSet) {
   taosThreadRwlockUnlock(&pData->lock);
 }
 
+static FORCE_INLINE void dmSwapEps(SEp *epLhs, SEp *epRhs) {
+  SEp epTmp;
+
+  epTmp.port = epLhs->port;
+  tstrncpy(epTmp.fqdn, epLhs->fqdn, tListLen(epTmp.fqdn));
+
+  epLhs->port = epRhs->port;
+  tstrncpy(epLhs->fqdn, epRhs->fqdn, tListLen(epLhs->fqdn));
+
+  epRhs->port = epTmp.port;
+  tstrncpy(epRhs->fqdn, epTmp.fqdn, tListLen(epRhs->fqdn));
+}
+
+void dmRotateMnodeEpSet(SDnodeData *pData) {
+  taosThreadRwlockRdlock(&pData->lock);
+  SEpSet *pEpSet = &pData->mnodeEps;
+  for (int i = 1; i < pEpSet->numOfEps; i++) {
+    dmSwapEps(&pEpSet->eps[i - 1], &pEpSet->eps[i]);
+  }
+  taosThreadRwlockUnlock(&pData->lock);
+}
+
 void dmGetMnodeEpSetForRedirect(SDnodeData *pData, SRpcMsg *pMsg, SEpSet *pEpSet) {
   dmGetMnodeEpSet(pData, pEpSet);
   dTrace("msg is redirected, handle:%p num:%d use:%d", pMsg->info.handle, pEpSet->numOfEps, pEpSet->inUse);

source/libs/sync/src/syncRaftStore.c

@@ -139,7 +139,7 @@ int32_t raftStoreWriteFile(SSyncNode *pNode) {
   if (taosRenameFile(file, realfile) != 0) goto _OVER;
 
   code = 0;
-  sInfo("vgId:%d, succeed to write raft store file:%s, len:%d", pNode->vgId, realfile, len);
+  sInfo("vgId:%d, succeed to write raft store file:%s, term:%" PRId64, pNode->vgId, realfile, pStore->currentTerm);
 
 _OVER:
   if (pJson != NULL) tjsonDelete(pJson);