From 31273a896728359b3de0d030ed70975fcd1c4efd Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Fri, 5 Feb 2021 09:34:23 +0800
Subject: [PATCH] taosd may crash while run m2.sim

---
 src/sync/src/syncMain.c    | 9 ++++++++-
 src/sync/src/syncRestore.c | 7 +++++++
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/src/sync/src/syncMain.c b/src/sync/src/syncMain.c
index 0f7fb77da8..956ccdc073 100644
--- a/src/sync/src/syncMain.c
+++ b/src/sync/src/syncMain.c
@@ -741,11 +741,14 @@ static SSyncPeer *syncCheckMaster(SSyncNode *pNode) {
       if (pTemp->role != TAOS_SYNC_ROLE_MASTER) continue;
       if (masterIndex < 0) {
         masterIndex = index;
+        sDebug("vgId:%d, peer:%s is master, index:%d", pNode->vgId, pTemp->id, index);
       } else {  // multiple masters, it shall not happen
         if (masterIndex == pNode->selfIndex) {
           sError("%s, peer is master, work as slave instead", pTemp->id);
           nodeRole = TAOS_SYNC_ROLE_SLAVE;
           (*pNode->notifyRoleFp)(pNode->vgId, nodeRole);
+        } else {
+          sError("vgId:%d, peer:%s is master too, masterIndex:%d index:%d", pNode->vgId, pTemp->id, masterIndex, index);
         }
       }
     }
@@ -833,7 +836,7 @@ static void syncCheckRole(SSyncPeer *pPeer, SPeerStatus* peersStatus, int8_t new
   }
 
   if (oldPeerRole != newPeerRole || nodeRole != oldSelfRole) {
-    sDebug("vgId:%d, roles changed, broadcast status", pNode->vgId);
+    sDebug("vgId:%d, roles changed, broadcast status, replica:%d", pNode->vgId, pNode->replica);
     syncBroadcastStatus(pNode);
   }
 
@@ -860,8 +863,12 @@ static void syncRestartPeer(SSyncPeer *pPeer) {
 void syncRestartConnection(SSyncPeer *pPeer) {
   if (pPeer->ip == 0) return;
 
+  if (syncAcquirePeer(pPeer->rid) == NULL) return;
+  
   syncRestartPeer(pPeer);
   syncCheckRole(pPeer, NULL, TAOS_SYNC_ROLE_OFFLINE);
+
+  syncReleasePeer(pPeer);
 }
 
 static void syncProcessSyncRequest(char *msg, SSyncPeer *pPeer) {
diff --git a/src/sync/src/syncRestore.c b/src/sync/src/syncRestore.c
index 76404c751e..2545f4c5ea 100644
--- a/src/sync/src/syncRestore.c
+++ b/src/sync/src/syncRestore.c
@@ -131,6 +131,11 @@ static int32_t syncProcessBufferedFwd(SSyncPeer *pPeer) {
   SRecvBuffer *pRecv = pNode->pRecv;
   int32_t      forwards = 0;
 
+  if (pRecv == NULL) {
+    sError("%s, recv buffer is null, restart connect", pPeer->id);
+    return -1;
+  }
+
   sDebug("%s, number of buffered forwards:%d", pPeer->id, pRecv->forwards);
 
   char *offset = pRecv->buffer;
@@ -179,6 +184,7 @@ int32_t syncSaveIntoBuffer(SSyncPeer *pPeer, SWalHead *pHead) {
 
 static void syncCloseRecvBuffer(SSyncNode *pNode) {
   if (pNode->pRecv) {
+    sDebug("vgId:%d, recv buffer:%p is freed", pNode->vgId, pNode->pRecv);
     tfree(pNode->pRecv->buffer);
   }
 
@@ -203,6 +209,7 @@ static int32_t syncOpenRecvBuffer(SSyncNode *pNode) {
 
   pNode->pRecv = pRecv;
 
+  sDebug("vgId:%d, recv buffer:%p is created", pNode->vgId, pNode->pRecv);
   return 0;
 }
 
-- 
GitLab