From 7ee61821e7ca0baf618ad6707d40eddec5dd4a5c Mon Sep 17 00:00:00 2001
From: Minglei Jin <stephenk@gmx.us>
Date: Sun, 28 Feb 2021 12:31:54 +0800
Subject: [PATCH] [TD-3107]<fix>: handle *Too many sync fwd infos* error crash
 with batch create table

---
 src/dnode/src/dnodeMWrite.c |  8 ++++++++
 src/mnode/src/mnodeTable.c  | 17 +++++++++++++++--
 src/mnode/src/mnodeVgroup.c | 13 +++++++++++++
 3 files changed, 36 insertions(+), 2 deletions(-)

diff --git a/src/dnode/src/dnodeMWrite.c b/src/dnode/src/dnodeMWrite.c
index 218d487473..79744e153e 100644
--- a/src/dnode/src/dnodeMWrite.c
+++ b/src/dnode/src/dnodeMWrite.c
@@ -145,6 +145,14 @@ void dnodeSendRpcMWriteRsp(void *pMsg, int32_t code) {
     return;
   }
 
+  dTrace("msg:%p, app:%p type:%s master:%p will be responsed", pWrite, pWrite->rpcMsg.ahandle,
+	 taosMsg[pWrite->rpcMsg.msgType], pWrite->pBatchMasterMsg);
+  if (pWrite->pBatchMasterMsg && pWrite != pWrite->pBatchMasterMsg) {
+    dError("msg:%p, app:%p type:%s master:%p sub message should not response!", pWrite, pWrite->rpcMsg.ahandle,
+	 taosMsg[pWrite->rpcMsg.msgType], pWrite->pBatchMasterMsg);
+    return;
+  }
+
   SRpcMsg rpcRsp = {
     .handle  = pWrite->rpcMsg.handle,
     .pCont   = pWrite->rpcRsp.rsp,
diff --git a/src/mnode/src/mnodeTable.c b/src/mnode/src/mnodeTable.c
index 7f463899ce..4fcedd82e3 100644
--- a/src/mnode/src/mnodeTable.c
+++ b/src/mnode/src/mnodeTable.c
@@ -777,11 +777,11 @@ static int32_t mnodeValidateCreateTableMsg(SCreateTableMsg *pCreateTable, SMnode
   }
 
   if (pCreateTable->numOfTags != 0) {
-    mDebug("msg:%p, app:%p table:%s, create stable msg is received from thandle:%p", pMsg, pMsg->rpcMsg.ahandle,
+    mDebug("msg:%p, app:%p table:%s, batch create stable msg is received from thandle:%p", pMsg, pMsg->rpcMsg.ahandle,
            pCreateTable->tableName, pMsg->rpcMsg.handle);
     return mnodeProcessCreateSuperTableMsg(pMsg);
   } else {
-    mDebug("msg:%p, app:%p table:%s, create ctable msg is received from thandle:%p", pMsg, pMsg->rpcMsg.ahandle,
+    mDebug("msg:%p, app:%p table:%s, batch create ctable msg is received from thandle:%p", pMsg, pMsg->rpcMsg.ahandle,
            pCreateTable->tableName, pMsg->rpcMsg.handle);
     return mnodeProcessCreateChildTableMsg(pMsg);
   }
@@ -1904,6 +1904,19 @@ static int32_t mnodeDoCreateChildTableCb(SMnodeMsg *pMsg, int32_t code) {
            pMsg->rpcMsg.ahandle, pTable->info.tableId, pTable->tid, pTable->uid, tstrerror(code));
     SSdbRow desc = {.type = SDB_OPER_GLOBAL, .pObj = pTable, .pTable = tsChildTableSdb};
     sdbDeleteRow(&desc);
+
+    if (pMsg->pBatchMasterMsg) {
+      ++pMsg->pBatchMasterMsg->successed;
+      if (pMsg->pBatchMasterMsg->successed + pMsg->pBatchMasterMsg->received
+	  >= pMsg->pBatchMasterMsg->expected) {
+	dnodeSendRpcMWriteRsp(pMsg->pBatchMasterMsg, code);
+      }
+
+      mnodeDestroySubMsg(pMsg);
+
+      return TSDB_CODE_MND_ACTION_IN_PROGRESS;
+    }
+
     return code;
   }
 }
diff --git a/src/mnode/src/mnodeVgroup.c b/src/mnode/src/mnodeVgroup.c
index 09e367ed66..67ee11640b 100644
--- a/src/mnode/src/mnodeVgroup.c
+++ b/src/mnode/src/mnodeVgroup.c
@@ -534,6 +534,19 @@ static int32_t mnodeCreateVgroupCb(SMnodeMsg *pMsg, int32_t code) {
            tstrerror(code));
     SSdbRow desc = {.type = SDB_OPER_GLOBAL, .pObj = pVgroup, .pTable = tsVgroupSdb};
     sdbDeleteRow(&desc);
+
+    if (pMsg->pBatchMasterMsg) {
+      ++pMsg->pBatchMasterMsg->received;
+      if (pMsg->pBatchMasterMsg->successed + pMsg->pBatchMasterMsg->received
+          >= pMsg->pBatchMasterMsg->expected) {
+        dnodeSendRpcMWriteRsp(pMsg->pBatchMasterMsg, pMsg->code);
+      }
+
+      mnodeDestroySubMsg(pMsg);
+
+      return TSDB_CODE_MND_ACTION_IN_PROGRESS;
+    }
+
     return code;
   } else {
     mInfo("msg:%p, app:%p vgId:%d, is created in sdb, db:%s replica:%d", pMsg, pMsg->rpcMsg.ahandle, pVgroup->vgId,
-- 
GitLab