Merge tulin@bk-internal.mysql.com:/home/bk/mysql-5.0-ndb

into  poseidon.mysql.com:/home/tomas/mysql-5.0-ndb

11 files changed, 197 insertions, 9 deletions

diff --git a/ndb/src/kernel/blocks/ERROR_codes.txt b/ndb/src/kernel/blocks/ERROR_codes.txt
index f7cb49014cb..9c2b441e7be 100644
--- a/ndb/src/kernel/blocks/ERROR_codes.txt
+++ b/ndb/src/kernel/blocks/ERROR_codes.txt
@@ -5,7 +5,7 @@ Next DBACC 3002
 Next DBTUP 4014
 Next DBLQH 5043
 Next DBDICT 6007
-Next DBDIH 7181
+Next DBDIH 7183
 Next DBTC 8039
 Next CMVMI 9000
 Next BACKUP 10022
@@ -489,3 +489,15 @@ Dbdict:
 6003 Crash in participant @ CreateTabReq::Prepare
 6004 Crash in participant @ CreateTabReq::Commit
 6005 Crash in participant @ CreateTabReq::CreateDrop
+
+TUP:
+----
+
+4025: Fail all inserts with out of memory
+4026: Fail one insert with oom
+4027: Fail inserts randomly with oom
+4028: Fail one random insert with oom
+
+NDBCNTR:
+
+1000: Crash insertion on SystemError::CopyFragRef
diff --git a/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp b/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
index 160a5fc135a..44cfa344ab3 100644
--- a/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
+++ b/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
@@ -4811,6 +4811,15 @@ void Dbdih::execMASTER_GCPREQ(Signal* signal)
   } else {
     ndbrequire(failedNodePtr.p->nodeStatus == NodeRecord::DYING);
   }//if
+
+  if (ERROR_INSERTED(7181))
+  {
+    ndbout_c("execGCP_TCFINISHED in MASTER_GCPREQ");
+    CLEAR_ERROR_INSERT_VALUE;
+    signal->theData[1] = coldgcp;
+    execGCP_TCFINISHED(signal);
+  }
+  
   MasterGCPConf::State gcpState;
   switch (cgcpParticipantState) {
   case GCP_PARTICIPANT_READY:
@@ -4877,6 +4886,14 @@ void Dbdih::execMASTER_GCPREQ(Signal* signal)
     masterGCPConf->lcpActive[i] = SYSFILE->lcpActive[i];
   sendSignal(newMasterBlockref, GSN_MASTER_GCPCONF, signal, 
              MasterGCPConf::SignalLength, JBB);
+
+  if (ERROR_INSERTED(7182))
+  {
+    ndbout_c("execGCP_TCFINISHED in MASTER_GCPREQ");
+    CLEAR_ERROR_INSERT_VALUE;
+    signal->theData[1] = coldgcp;
+    execGCP_TCFINISHED(signal);
+  }
 }//Dbdih::execMASTER_GCPREQ()
 
 void Dbdih::execMASTER_GCPCONF(Signal* signal) 
@@ -7549,10 +7566,10 @@ void Dbdih::execGCP_NODEFINISH(Signal* signal)
   } else if (cmasterState == MASTER_TAKE_OVER_GCP) {
     jam();
     //-------------------------------------------------------------
-    // We are currently taking over as master. We will delay the
-    // signal until we have completed the take over gcp handling.
+    // We are currently taking over as master. Ignore
+    // signal in this case since we will discover it in reception of 
+    // MASTER_GCPCONF.
     //-------------------------------------------------------------
-    sendSignalWithDelay(reference(), GSN_GCP_NODEFINISH, signal, 20, 3);
     return;
   } else {
     ndbrequire(cmasterState == MASTER_ACTIVE);
@@ -7699,6 +7716,15 @@ void Dbdih::execGCP_TCFINISHED(Signal* signal)
   Uint32 gci = signal->theData[1];
   ndbrequire(gci == coldgcp);
 
+  if (ERROR_INSERTED(7181) || ERROR_INSERTED(7182))
+  {
+    ndbout_c("killing %d", refToNode(cmasterdihref));
+    signal->theData[0] = 9999;
+    sendSignal(numberToRef(CMVMI, refToNode(cmasterdihref)),
+	       GSN_NDB_TAMPER, signal, 1, JBB);
+    return;
+  }
+
   cgcpParticipantState = GCP_PARTICIPANT_TC_FINISHED;
   signal->theData[0] = cownNodeId;
   signal->theData[1] = coldgcp;
diff --git a/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp b/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
index d73d15f416f..9944adf4ede 100644
--- a/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
+++ b/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
@@ -9634,6 +9634,15 @@ void Dblqh::copyCompletedLab(Signal* signal)
     closeCopyLab(signal);
     return;
   }//if
+
+  if (scanptr.p->scanState == ScanRecord::WAIT_LQHKEY_COPY &&
+      scanptr.p->scanErrorCounter)
+  {
+    jam();
+    closeCopyLab(signal);
+    return;
+  }
+  
   if (scanptr.p->scanState == ScanRecord::WAIT_LQHKEY_COPY) {
     jam();
 /*---------------------------------------------------------------------------*/
@@ -9710,13 +9719,16 @@ void Dblqh::continueCopyAfterBlockedLab(Signal* signal)
 void Dblqh::copyLqhKeyRefLab(Signal* signal) 
 {
   ndbrequire(tcConnectptr.p->transid[1] == signal->theData[4]);
-  tcConnectptr.p->copyCountWords -= signal->theData[3];
+  Uint32 copyWords = signal->theData[3];
   scanptr.i = tcConnectptr.p->tcScanRec;
   c_scanRecordPool.getPtr(scanptr);
   scanptr.p->scanErrorCounter++;
   tcConnectptr.p->errorCode = terrorCode;
-  closeCopyLab(signal);
-  return;
+  
+  LqhKeyConf* conf = (LqhKeyConf*)signal->getDataPtrSend();
+  conf->transId1 = copyWords;
+  conf->transId2 = tcConnectptr.p->transid[1];
+  copyCompletedLab(signal);
 }//Dblqh::copyLqhKeyRefLab()
 
 void Dblqh::closeCopyLab(Signal* signal) 
@@ -9727,6 +9739,7 @@ void Dblqh::closeCopyLab(Signal* signal)
 // Wait until all of those have arrived until we start the
 // close process.
 /*---------------------------------------------------------------------------*/
+    scanptr.p->scanState = ScanRecord::WAIT_LQHKEY_COPY;
     jam();
     return;
   }//if
diff --git a/ndb/src/kernel/blocks/dbtup/DbtupExecQuery.cpp b/ndb/src/kernel/blocks/dbtup/DbtupExecQuery.cpp
index 42b86102dff..72ef9a274ec 100644
--- a/ndb/src/kernel/blocks/dbtup/DbtupExecQuery.cpp
+++ b/ndb/src/kernel/blocks/dbtup/DbtupExecQuery.cpp
@@ -212,6 +212,30 @@ void Dbtup::execTUP_ALLOCREQ(Signal* signal)
 //---------------------------------------------------
   PagePtr pagePtr;
   Uint32 pageOffset;
+
+  if (ERROR_INSERTED(4025))
+  {
+    signal->theData[0] = 827;
+    return;
+  }
+  if (ERROR_INSERTED(4026))
+  {
+    CLEAR_ERROR_INSERT_VALUE;
+    signal->theData[0] = 827;
+    return;
+  }
+  if (ERROR_INSERTED(4027) && (rand() % 100) > 25)
+  {
+    signal->theData[0] = 827;
+    return;
+  }
+  if (ERROR_INSERTED(4028) && (rand() % 100) > 25)
+  {
+    CLEAR_ERROR_INSERT_VALUE;
+    signal->theData[0] = 827;
+    return;
+  }
+  
   if (!allocTh(regFragPtr.p,
                regTabPtr.p,
                NORMAL_PAGE,
diff --git a/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp b/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp
index 0f83c45077b..df8df2d29f3 100644
--- a/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp
+++ b/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp
@@ -65,6 +65,7 @@ void Dbtup::initData()
   undoPage = 0;
   totNoOfPagesAllocated = 0;
   cnoOfAllocatedPages = 0;
+  CLEAR_ERROR_INSERT_VALUE;
   
   // Records with constant sizes
 }//Dbtup::initData()
@@ -568,7 +569,6 @@ void Dbtup::execSTTOR(Signal* signal)
   switch (startPhase) {
   case ZSTARTPHASE1:
     ljam();
-    CLEAR_ERROR_INSERT_VALUE;
     cownref = calcTupBlockRef(0);
     break;
   default:
diff --git a/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp b/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp
index b3f35f12b52..5d34d3f1126 100644
--- a/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp
+++ b/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp
@@ -179,6 +179,7 @@ void Ndbcntr::execSYSTEM_ERROR(Signal* signal)
     break;
 
   case SystemError::CopyFragRefError:
+    CRASH_INSERTION(1000);
     BaseString::snprintf(buf, sizeof(buf), 
 			 "Killed by node %d as "
 			 "copyfrag failed, error: %u",
diff --git a/ndb/src/mgmsrv/ConfigInfo.cpp b/ndb/src/mgmsrv/ConfigInfo.cpp
index f4f643b0f61..0cf37b5f874 100644
--- a/ndb/src/mgmsrv/ConfigInfo.cpp
+++ b/ndb/src/mgmsrv/ConfigInfo.cpp
@@ -565,7 +565,7 @@ const ConfigInfo::ParamInfo ConfigInfo::m_ParamInfo[] = {
     true,
     ConfigInfo::CI_INT,
     "0",
-    "1",
+    "0",
     "2" },
 
   {
diff --git a/ndb/src/mgmsrv/MgmtSrvr.cpp b/ndb/src/mgmsrv/MgmtSrvr.cpp
index 5635d5b4420..3d7eb5bd4f1 100644
--- a/ndb/src/mgmsrv/MgmtSrvr.cpp
+++ b/ndb/src/mgmsrv/MgmtSrvr.cpp
@@ -2494,6 +2494,8 @@ MgmtSrvr::startBackup(Uint32& backupId, int waitCompleted)
 	ndbout_c("I'm not master resending to %d", nodeId);
 #endif
 	do_send = 1; // try again
+	if (!theFacade->get_node_alive(nodeId))
+	  m_master_node = nodeId = 0;
 	continue;
       }
       event.Event = BackupEvent::BackupFailedToStart;
diff --git a/ndb/test/ndbapi/testNodeRestart.cpp b/ndb/test/ndbapi/testNodeRestart.cpp
index a5e787dfd0e..26c8fe35011 100644
--- a/ndb/test/ndbapi/testNodeRestart.cpp
+++ b/ndb/test/ndbapi/testNodeRestart.cpp
@@ -1124,6 +1124,101 @@ runBug26481(NDBT_Context* ctx, NDBT_Step* step)
   return NDBT_OK;
 }
 
+int
+runBug27003(NDBT_Context* ctx, NDBT_Step* step)
+{
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter res;
+  
+  static const int errnos[] = { 4025, 4026, 4027, 4028, 0 };
+
+  int node = res.getRandomNotMasterNodeId(rand());
+  ndbout_c("node: %d", node);
+  if (res.restartOneDbNode(node, false, true, true))
+    return NDBT_FAILED;
+
+  Uint32 pos = 0;
+  for (Uint32 i = 0; i<loops; i++)
+  {
+    while (errnos[pos] != 0)
+    {
+      ndbout_c("Tesing err: %d", errnos[pos]);
+      
+      if (res.waitNodesNoStart(&node, 1))
+	return NDBT_FAILED;
+
+      if (res.insertErrorInNode(node, 1000))
+	return NDBT_FAILED;
+      
+      if (res.insertErrorInNode(node, errnos[pos]))
+	return NDBT_FAILED;
+      
+      int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+      if (res.dumpStateOneNode(node, val2, 2))
+	return NDBT_FAILED;
+      
+      res.startNodes(&node, 1);
+      res.waitNodesStartPhase(&node, 1, 2);
+      pos++;
+    }
+    pos = 0;
+  }
+
+  if (res.waitNodesNoStart(&node, 1))
+    return NDBT_FAILED;
+  
+  res.startNodes(&node, 1);
+  if (res.waitClusterStarted())
+    return NDBT_FAILED;
+  
+  return NDBT_OK;
+}
+
+
+int
+runBug27283(NDBT_Context* ctx, NDBT_Step* step)
+{
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter res;
+
+  if (res.getNumDbNodes() < 2)
+  {
+    return NDBT_OK;
+  }
+
+  static const int errnos[] = { 7181, 7182, 0 };
+  
+  Uint32 pos = 0;
+  for (Uint32 i = 0; i<loops; i++)
+  {
+    while (errnos[pos] != 0)
+    {
+      int master = res.getMasterNodeId();
+      int next = res.getNextMasterNodeId(master);
+      int next2 = res.getNextMasterNodeId(next);
+      
+      int node = (i & 1) ? next : next2;
+      ndbout_c("Tesing err: %d", errnos[pos]);
+      if (res.insertErrorInNode(next, errnos[pos]))
+	return NDBT_FAILED;
+
+      NdbSleep_SecSleep(3);
+      
+      if (res.waitClusterStarted())
+	return NDBT_FAILED;
+      
+      pos++;
+    }
+    pos = 0;
+  }
+  
+  return NDBT_OK;
+}
+
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
 	 "Test that one node at a time can be stopped and then restarted "\
@@ -1451,6 +1546,12 @@ TESTCASE("Bug26457", ""){
 TESTCASE("Bug26481", ""){
   INITIALIZER(runBug26481);
 }
+TESTCASE("Bug27003", ""){
+  INITIALIZER(runBug27003);
+}
+TESTCASE("Bug27283", ""){
+  INITIALIZER(runBug27283);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){
diff --git a/ndb/test/run-test/daily-basic-tests.txt b/ndb/test/run-test/daily-basic-tests.txt
index 06cd0664186..a35b2b3f7c7 100644
--- a/ndb/test/run-test/daily-basic-tests.txt
+++ b/ndb/test/run-test/daily-basic-tests.txt
@@ -425,6 +425,14 @@ max-time: 500
 cmd: testScan
 args: -n Bug24447 T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug27003 T1
+
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug27283 T1
+
 max-time: 500
 cmd: testNodeRestart
 args: -n Bug15587 T1
diff --git a/ndb/test/src/UtilTransactions.cpp b/ndb/test/src/UtilTransactions.cpp
index 3a166f19c92..011cea24af9 100644
--- a/ndb/test/src/UtilTransactions.cpp
+++ b/ndb/test/src/UtilTransactions.cpp
@@ -1381,6 +1381,7 @@ UtilTransactions::compare(Ndb* pNdb, const char* tab_name2, int flags){
       goto error;
     }
   
+    row_count= 0;
     {
       int eof;
       while((eof = pOp->nextResult(true)) == 0)