ndb - bug#25554

  fix bug when master failure during nr
  (recommit against 5.0)

ndb/src/kernel/blocks/dbdih/DbdihMain.cpp

@@ -3555,7 +3555,6 @@ void Dbdih::endTakeOver(Uint32 takeOverPtrI)
   takeOverPtr.i = takeOverPtrI;
   ptrCheckGuard(takeOverPtr, MAX_NDB_NODES, takeOverRecord);
 
-  releaseTakeOver(takeOverPtrI);
   if ((takeOverPtr.p->toMasterStatus != TakeOverRecord::IDLE) &&
       (takeOverPtr.p->toMasterStatus != TakeOverRecord::TO_WAIT_START_TAKE_OVER)) {
     jam();
@@ -3569,6 +3568,7 @@ void Dbdih::endTakeOver(Uint32 takeOverPtrI)
   }//if
   setAllowNodeStart(takeOverPtr.p->toStartingNode, true);
   initTakeOver(takeOverPtr);
+  releaseTakeOver(takeOverPtrI);
 }//Dbdih::endTakeOver()
 
 void Dbdih::releaseTakeOver(Uint32 takeOverPtrI)
@@ -4710,6 +4710,7 @@ void Dbdih::handleTakeOverNewMaster(Signal* signal, Uint32 takeOverPtrI)
       break;
     }
     ndbrequire(ok);
+    endTakeOver(takeOverPtr.i);
   }//if
 }//Dbdih::handleTakeOverNewMaster()
 

ndb/src/kernel/blocks/qmgr/QmgrMain.cpp

@@ -2847,6 +2847,17 @@ void Qmgr::failReportLab(Signal* signal, Uint16 aFailedNode,
     systemErrorLab(signal, __LINE__);
     return;
   }//if
+
+  if (getNodeState().startLevel < NodeState::SL_STARTED)
+  {
+    jam();
+    CRASH_INSERTION(932);
+    char buf[100];
+    BaseString::snprintf(buf, 100, "Node failure during restart");
+    progError(__LINE__, NDBD_EXIT_SR_OTHERNODEFAILED, buf);
+    ndbrequire(false);
+  }
+
   TnoFailedNodes = cnoFailedNodes;
   failReport(signal, failedNodePtr.i, (UintR)ZTRUE, aFailCause);
   if (cpresident == getOwnNodeId()) {
@@ -2933,6 +2944,16 @@ void Qmgr::execPREP_FAILREQ(Signal* signal)
     return;
   }//if
 
+  if (getNodeState().startLevel < NodeState::SL_STARTED)
+  {
+    jam();
+    CRASH_INSERTION(932);
+    char buf[100];
+    BaseString::snprintf(buf, 100, "Node failure during restart");
+    progError(__LINE__, NDBD_EXIT_SR_OTHERNODEFAILED, buf);
+    ndbrequire(false);
+  }
+
   guard0 = cnoPrepFailedNodes - 1;
   arrGuard(guard0, MAX_NDB_NODES);
   for (Tindex = 0; Tindex <= guard0; Tindex++) {

ndb/test/ndbapi/testNodeRestart.cpp

@@ -995,6 +995,56 @@ int runBug25364(NDBT_Context* ctx, NDBT_Step* step){
   return NDBT_OK;
 }
 
+int runBug25554(NDBT_Context* ctx, NDBT_Step* step){
+  
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter restarter;
+  
+  if (restarter.getNumDbNodes() < 4)
+    return NDBT_OK;
+
+  for (int i = 0; i<loops; i++)
+  {
+    int master = restarter.getMasterNodeId();
+    int node1 = restarter.getRandomNodeOtherNodeGroup(master, rand());
+    restarter.restartOneDbNode(node1, false, true, true);
+
+    int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+  
+    if (restarter.dumpStateOneNode(master, val2, 2))
+      return NDBT_FAILED;
+
+    if (restarter.insertErrorInNode(master, 7141))
+      return NDBT_FAILED;
+
+    if (restarter.waitNodesNoStart(&node1, 1))
+      return NDBT_FAILED;
+
+    if (restarter.dumpStateOneNode(node1, val2, 2))
+      return NDBT_FAILED;
+
+    if (restarter.insertErrorInNode(node1, 932))
+      return NDBT_FAILED;
+
+    if (restarter.startNodes(&node1, 1))
+      return NDBT_FAILED;
+
+    int nodes[] = { master, node1 };
+    if (restarter.waitNodesNoStart(nodes, 2))
+      return NDBT_FAILED;
+
+    if (restarter.startNodes(nodes, 2))
+      return NDBT_FAILED;
+
+    if (restarter.waitClusterStarted())
+      return NDBT_FAILED;
+  }    
+
+  return NDBT_OK;
+}
+
 
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
@@ -1314,6 +1364,9 @@ TESTCASE("Bug24717", ""){
 TESTCASE("Bug25364", ""){
   INITIALIZER(runBug25364);
 }
+TESTCASE("Bug25554", ""){
+  INITIALIZER(runBug25554);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

ndb/test/run-test/daily-basic-tests.txt

@@ -473,6 +473,10 @@ max-time: 1000
 cmd: testNodeRestart
 args: -n Bug25364 T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug25554 T1
+
 # OLD FLEX
 max-time: 500
 cmd: flexBench