Merge perch.ndb.mysql.com:/home/jonas/src/mysql-5.0

into  perch.ndb.mysql.com:/home/jonas/src/mysql-5.0-push

ndb/include/kernel/signaldata/DumpStateOrd.hpp

@@ -78,6 +78,8 @@ public:
     LqhDumpAllScanRec = 2301,
     LqhDumpAllActiveScanRec = 2302,
     LqhDumpLcpState = 2303,
+    LqhErrorInsert5042 = 2315,
+
     AccDumpOneScanRec = 2400,
     AccDumpAllScanRec = 2401,
     AccDumpAllActiveScanRec = 2402,

ndb/include/ndb_version.h.in

@@ -57,5 +57,8 @@ char ndb_version_string_buf[NDB_VERSION_STRING_BUF_SZ];
  */
 /*#define NDB_VERSION_ID 0*/
 
+#define NDBD_INCL_NODECONF_VERSION_4 MAKE_VERSION(4,1,17)
+#define NDBD_INCL_NODECONF_VERSION_5 MAKE_VERSION(5,0,18)
+
 #endif
  

ndb/src/kernel/blocks/ERROR_codes.txt

@@ -61,6 +61,8 @@ Insert system error in GCP participant when receiving GCP_SAVEREQ.
 5007:
 Delay GCP_SAVEREQ by 10 secs
 
+7165: Delay INCL_NODE_REQ in starting node yeilding error in GCP_PREPARE
+
 ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:
 -----------------------------------------------------------------
 
@@ -155,11 +157,15 @@ Insert node failure handling when receiving COMPLETEREQ.
 5006:
 Insert node failure handling when receiving ABORTREQ.
 
+5042:
+As 5002, but with specified table (see DumpStateOrd)
+
 These error code can be combined with error codes for testing time-out
 handling in DBTC to ensure that node failures are also well handled in
 time-out handling. They can also be used to test multiple node failure
 handling.
 
+
 ERROR CODES FOR TESTING TIME-OUT HANDLING IN DBLQH
 -------------------------------------------------
 5011:
@@ -198,6 +204,9 @@ Delay execution of ABORTREQ signal 2 seconds to generate time-out.
 
 8050: Send TCKEYREF is operation is non local
 
+5100,5101: Drop ABORT req in primary replica
+           Crash on "next" ABORT
+
 ERROR CODES FOR TESTING TIME-OUT HANDLING IN DBTC
 -------------------------------------------------
 8040:

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

@@ -215,7 +215,7 @@ void Dbdih::sendINCL_NODEREQ(Signal* signal, Uint32 nodeId)
   signal->theData[2] = c_nodeStartMaster.failNr;
   signal->theData[3] = 0;
   signal->theData[4] = currentgcp;  
-  sendSignal(nodeDihRef, GSN_INCL_NODEREQ, signal, 5, JBB);
+  sendSignal(nodeDihRef, GSN_INCL_NODEREQ, signal, 5, JBA);
 }//Dbdih::sendINCL_NODEREQ()
 
 void Dbdih::sendMASTER_GCPREQ(Signal* signal, Uint32 nodeId)
@@ -1863,6 +1863,14 @@ void Dbdih::gcpBlockedLab(Signal* signal)
   // global checkpoint id and the correct state. We do not wait for any reply
   // since the starting node will not send any.
   /*-------------------------------------------------------------------------*/
+  Uint32 startVersion = getNodeInfo(c_nodeStartMaster.startNode).m_version;
+  
+  if ((getMajor(startVersion) == 4 && startVersion >= NDBD_INCL_NODECONF_VERSION_4) ||
+      (getMajor(startVersion) == 5 && startVersion >= NDBD_INCL_NODECONF_VERSION_5))
+  {
+    c_INCL_NODEREQ_Counter.setWaitingFor(c_nodeStartMaster.startNode);
+  }
+  
   sendINCL_NODEREQ(signal, c_nodeStartMaster.startNode);
 }//Dbdih::gcpBlockedLab()
 
@@ -2065,6 +2073,13 @@ void Dbdih::execINCL_NODEREQ(Signal* signal)
   jamEntry();
   Uint32 retRef = signal->theData[0];
   Uint32 nodeId = signal->theData[1];
+  if (nodeId == getOwnNodeId() && ERROR_INSERTED(7165))
+  {
+    CLEAR_ERROR_INSERT_VALUE;
+    sendSignalWithDelay(reference(), GSN_INCL_NODEREQ, signal, 5000, signal->getLength());
+    return;
+  }
+  
   Uint32 tnodeStartFailNr = signal->theData[2];
   currentgcp = signal->theData[4];
   CRASH_INSERTION(7127);
@@ -2092,6 +2107,15 @@ void Dbdih::execINCL_NODEREQ(Signal* signal)
     // id's and the lcp status.
     /*-----------------------------------------------------------------------*/
     CRASH_INSERTION(7171);
+    Uint32 masterVersion = getNodeInfo(refToNode(cmasterdihref)).m_version;
+    
+    if ((NDB_VERSION_MAJOR == 4 && masterVersion >= NDBD_INCL_NODECONF_VERSION_4) ||
+	(NDB_VERSION_MAJOR == 5 && masterVersion >= NDBD_INCL_NODECONF_VERSION_5))
+    {
+      signal->theData[0] = getOwnNodeId();
+      signal->theData[1] = getOwnNodeId();
+      sendSignal(cmasterdihref, GSN_INCL_NODECONF, signal, 2, JBB);
+    }
     return;
   }//if
   if (getNodeStatus(nodeId) != NodeRecord::STARTING) {
@@ -3741,8 +3765,16 @@ void Dbdih::execNODE_FAILREP(Signal* signal)
   /*------------------------------------------------------------------------*/
   // Verify that a starting node has also crashed. Reset the node start record.
   /*-------------------------------------------------------------------------*/
-  if (c_nodeStartMaster.startNode != RNIL) {
-    ndbrequire(getNodeStatus(c_nodeStartMaster.startNode)!= NodeRecord::ALIVE);
+  if (false && c_nodeStartMaster.startNode != RNIL && getNodeStatus(c_nodeStartMaster.startNode) == NodeRecord::ALIVE)
+  {
+    BlockReference cntrRef = calcNdbCntrBlockRef(c_nodeStartMaster.startNode);
+    SystemError * const sysErr = (SystemError*)&signal->theData[0];
+    sysErr->errorCode = SystemError::StartInProgressError;
+    sysErr->errorRef = reference();
+    sysErr->data1= 0;
+    sysErr->data2= __LINE__;
+    sendSignal(cntrRef, GSN_SYSTEM_ERROR, signal,  SystemError::SignalLength, JBA);
+    nodeResetStart();  
   }//if
 
   /*--------------------------------------------------*/
@@ -5189,15 +5221,16 @@ void Dbdih::removeNodeFromTable(Signal* signal,
     /**
      * For each of replica record
      */
-    Uint32 replicaNo = 0;
+    bool found = false;
     ReplicaRecordPtr replicaPtr;
     for(replicaPtr.i = fragPtr.p->storedReplicas; replicaPtr.i != RNIL;
-        replicaPtr.i = replicaPtr.p->nextReplica, replicaNo++) {
+        replicaPtr.i = replicaPtr.p->nextReplica) {
       jam();
 
       ptrCheckGuard(replicaPtr, creplicaFileSize, replicaRecord);
       if(replicaPtr.p->procNode == nodeId){
         jam();
+	found = true;
 	noOfRemovedReplicas++;
 	removeNodeFromStored(nodeId, fragPtr, replicaPtr);
 	if(replicaPtr.p->lcpOngoingFlag){
@@ -5213,6 +5246,15 @@ void Dbdih::removeNodeFromTable(Signal* signal,
 	}
       }
     }
+    if (!found)
+    {
+      jam();
+      /**
+       * Run updateNodeInfo to remove any dead nodes from list of activeNodes
+       *  see bug#15587
+       */
+      updateNodeInfo(fragPtr);
+    }
     noOfRemainingLcpReplicas += fragPtr.p->noLcpReplicas;
   }
   

ndb/src/kernel/blocks/dblqh/Dblqh.hpp

@@ -2885,6 +2885,7 @@ private:
   UintR ctransidHash[1024];
   
   Uint32 c_diskless;
+  Uint32 c_error_insert_table_id;
   
 public:
   /**

ndb/src/kernel/blocks/dblqh/DblqhMain.cpp

@@ -3543,6 +3543,7 @@ void Dblqh::execLQHKEYREQ(Signal* signal)
     jam();
     regTcPtr->activeCreat = ZTRUE;
     CRASH_INSERTION(5002);
+    CRASH_INSERTION2(5042, tabptr.i == c_error_insert_table_id);
   } else {
     regTcPtr->activeCreat = ZFALSE;
   }//if
@@ -5880,12 +5881,21 @@ void Dblqh::execABORT(Signal* signal)
     warningReport(signal, 8);
     return;
   }//if
+  
+  TcConnectionrec * const regTcPtr = tcConnectptr.p;
+
+  if (ERROR_INSERTED(5100))
+  {
+    SET_ERROR_INSERT_VALUE(5101);
+    return;
+  }
+  CRASH_INSERTION2(5101, regTcPtr->nextReplica != ZNIL);
+  
 /* ------------------------------------------------------------------------- */
 /*A GUIDING DESIGN PRINCIPLE IN HANDLING THESE ERROR SITUATIONS HAVE BEEN    */
 /*KEEP IT SIMPLE. THUS WE RATHER INSERT A WAIT AND SET THE ABORT_STATE TO    */
 /*ACTIVE RATHER THAN WRITE NEW CODE TO HANDLE EVERY SPECIAL SITUATION.       */
 /* ------------------------------------------------------------------------- */
-  TcConnectionrec * const regTcPtr = tcConnectptr.p;
   if (regTcPtr->nextReplica != ZNIL) {
 /* ------------------------------------------------------------------------- */
 // We will immediately send the ABORT message also to the next LQH node in line.
@@ -18522,6 +18532,12 @@ Dblqh::execDUMP_STATE_ORD(Signal* signal)
       
     }
   }
+
+  if (dumpState->args[0] == DumpStateOrd::LqhErrorInsert5042 && signal->getLength() == 2)
+  {
+    c_error_insert_table_id = dumpState->args[1];
+    SET_ERROR_INSERT_VALUE(5042);
+  }
 }//Dblqh::execDUMP_STATE_ORD()
 
 void Dblqh::execSET_VAR_REQ(Signal* signal) 

ndb/src/kernel/blocks/dbtc/DbtcMain.cpp

@@ -6183,7 +6183,6 @@ void Dbtc::timeOutFoundLab(Signal* signal, Uint32 TapiConPtr)
 	<< " - place: " << c_apiConTimer_line[apiConnectptr.i]);
   switch (apiConnectptr.p->apiConnectstate) {
   case CS_STARTED:
-    ndbrequire(c_apiConTimer_line[apiConnectptr.i] != 3615);
     if(apiConnectptr.p->lqhkeyreqrec == apiConnectptr.p->lqhkeyconfrec){
       jam();
       /*
@@ -6443,8 +6442,8 @@ void Dbtc::sendAbortedAfterTimeout(Signal* signal, int Tcheck)
 	warningEvent(buf);
 	ndbout_c(buf);
 	ndbrequire(false);
+	releaseAbortResources(signal);
       }
-      releaseAbortResources(signal);
       return;
     }//if
     TloopCount++;

ndb/test/ndbapi/testNodeRestart.cpp

@@ -21,6 +21,7 @@
 #include <NdbRestarter.hpp>
 #include <NdbRestarts.hpp>
 #include <Vector.hpp>
+#include <signaldata/DumpStateOrd.hpp>
 
 
 int runLoadTable(NDBT_Context* ctx, NDBT_Step* step){
@@ -409,6 +410,132 @@ int runLateCommit(NDBT_Context* ctx, NDBT_Step* step){
   return NDBT_OK;
 }
 
+int runBug15587(NDBT_Context* ctx, NDBT_Step* step){
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter restarter;
+  
+  Uint32 tableId = ctx->getTab()->getTableId();
+  int dump[2] = { DumpStateOrd::LqhErrorInsert5042, 0 };
+  dump[1] = tableId;
+
+  int nodeId = restarter.getDbNodeId(1);
+
+  ndbout << "Restart node " << nodeId << endl; 
+  
+  if (restarter.restartOneDbNode(nodeId,
+				 /** initial */ false, 
+				 /** nostart */ true,
+				 /** abort   */ true))
+    return NDBT_FAILED;
+  
+  if (restarter.waitNodesNoStart(&nodeId, 1))
+    return NDBT_FAILED; 
+   
+  if (restarter.dumpStateOneNode(nodeId, dump, 2))
+    return NDBT_FAILED;
+
+  if (restarter.startNodes(&nodeId, 1))
+    return NDBT_FAILED;
+
+  if (restarter.waitNodesStarted(&nodeId, 1))
+    return NDBT_FAILED;
+  
+  ctx->stopTest();
+  return NDBT_OK;
+}
+
+int runBug15632(NDBT_Context* ctx, NDBT_Step* step){
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter restarter;
+  
+  int nodeId = restarter.getDbNodeId(1);
+
+  ndbout << "Restart node " << nodeId << endl; 
+  
+  if (restarter.restartOneDbNode(nodeId,
+				 /** initial */ false, 
+				 /** nostart */ true,
+				 /** abort   */ true))
+    return NDBT_FAILED;
+  
+  if (restarter.waitNodesNoStart(&nodeId, 1))
+    return NDBT_FAILED; 
+   
+  if (restarter.insertErrorInNode(nodeId, 7165))
+    return NDBT_FAILED;
+  
+  if (restarter.startNodes(&nodeId, 1))
+    return NDBT_FAILED;
+
+  if (restarter.waitNodesStarted(&nodeId, 1))
+    return NDBT_FAILED;
+
+  if (restarter.restartOneDbNode(nodeId,
+				 /** initial */ false, 
+				 /** nostart */ true,
+				 /** abort   */ true))
+    return NDBT_FAILED;
+  
+  if (restarter.waitNodesNoStart(&nodeId, 1))
+    return NDBT_FAILED; 
+   
+  if (restarter.insertErrorInNode(nodeId, 7171))
+    return NDBT_FAILED;
+  
+  if (restarter.startNodes(&nodeId, 1))
+    return NDBT_FAILED;
+  
+  if (restarter.waitNodesStarted(&nodeId, 1))
+    return NDBT_FAILED;
+  
+  ctx->stopTest();
+  return NDBT_OK;
+}
+
+int runBug15685(NDBT_Context* ctx, NDBT_Step* step){
+
+  Ndb* pNdb = GETNDB(step);
+  HugoOperations hugoOps(*ctx->getTab());
+  NdbRestarter restarter;
+
+  HugoTransactions hugoTrans(*ctx->getTab());
+  if (hugoTrans.loadTable(GETNDB(step), 10) != 0){
+    return NDBT_FAILED;
+  }
+
+  if(hugoOps.startTransaction(pNdb) != 0)
+    goto err;
+  
+  if(hugoOps.pkUpdateRecord(pNdb, 0, 1, rand()) != 0)
+    goto err;
+
+  if(hugoOps.execute_NoCommit(pNdb) != 0)
+    goto err;
+
+  if (restarter.insertErrorInAllNodes(5100))
+    return NDBT_FAILED;
+  
+  hugoOps.execute_Rollback(pNdb);
+
+  if (restarter.waitClusterStarted() != 0)
+    goto err;
+
+  if (restarter.insertErrorInAllNodes(0))
+    return NDBT_FAILED;
+  
+  ctx->stopTest();
+  return NDBT_OK;
+  
+err:
+  ctx->stopTest();
+  return NDBT_FAILED;
+}
+
+
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
 	 "Test that one node at a time can be stopped and then restarted "\
@@ -558,6 +685,8 @@ TESTCASE("RestartNFDuringNR",
   INITIALIZER(runCheckAllNodesStarted);
   INITIALIZER(runLoadTable);
   STEP(runRestarts);
+  STEP(runPkUpdateUntilStopped);
+  STEP(runScanUpdateUntilStopped);
   FINALIZER(runScanReadVerify);
   FINALIZER(runClearTable);
 }
@@ -647,6 +776,8 @@ TESTCASE("RestartNodeDuringLCP",
   INITIALIZER(runCheckAllNodesStarted);
   INITIALIZER(runLoadTable);
   STEP(runRestarts);
+  STEP(runPkUpdateUntilStopped);
+  STEP(runScanUpdateUntilStopped);
   FINALIZER(runScanReadVerify);
   FINALIZER(runClearTable);
 }
@@ -671,6 +802,24 @@ TESTCASE("LateCommit",
   STEP(runLateCommit);
   FINALIZER(runClearTable);
 }
+TESTCASE("Bug15587",
+	 "Test bug with NF during NR"){
+  INITIALIZER(runLoadTable);
+  STEP(runScanUpdateUntilStopped);
+  STEP(runBug15587);
+  FINALIZER(runClearTable);
+}
+TESTCASE("Bug15632",
+	 "Test bug with NF during NR"){
+  INITIALIZER(runLoadTable);
+  STEP(runBug15632);
+  FINALIZER(runClearTable);
+}
+TESTCASE("Bug15685",
+	 "Test bug with NF during abort"){
+  STEP(runBug15685);
+  FINALIZER(runClearTable);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -413,6 +413,27 @@ max-time: 500
 cmd: testScan
 args: -n ScanParallelism
 
+max-time: 500
+cmd: testNodeRestart
+args: -n Bug15587 T1
+
+max-time: 500
+cmd: testNodeRestart
+args: -n Bug15632 T1
+
+max-time: 500
+cmd: testNodeRestart
+args: -n Bug15685 T1
+
+# OLD FLEX
+max-time: 500
+cmd: flexBench
+args: -c 25 -t 10
+
+max-time: 500
+cmd: flexHammer
+args: -r 5 -t 32
+
 #
 # DICT TESTS
 max-time: 1500