ndb - bug#27003

  Handle random(not in order) LQHKEYREQ failures during node-restart

ndb/src/kernel/blocks/ERROR_codes.txt

@@ -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

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

@@ -9641,6 +9641,15 @@ 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();
 /*---------------------------------------------------------------------------*/
@@ -9717,13 +9726,16 @@ 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) 
@@ -9734,6 +9746,7 @@ 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

ndb/src/kernel/blocks/dbtup/DbtupExecQuery.cpp

@@ -213,6 +213,30 @@ 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,

ndb/src/kernel/blocks/dbtup/DbtupGen.cpp

@@ -66,6 +66,7 @@ void Dbtup::initData()
   undoPage = 0;
   totNoOfPagesAllocated = 0;
   cnoOfAllocatedPages = 0;
+  CLEAR_ERROR_INSERT_VALUE;
   
   // Records with constant sizes
 }//Dbtup::initData()
@@ -570,7 +571,6 @@ void Dbtup::execSTTOR(Signal* signal)
   switch (startPhase) {
   case ZSTARTPHASE1:
     ljam();
-    CLEAR_ERROR_INSERT_VALUE;
     cownref = calcTupBlockRef(0);
     break;
   default:

ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp

@@ -180,6 +180,7 @@ 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",

ndb/test/ndbapi/testNodeRestart.cpp

@@ -1125,6 +1125,59 @@ 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;
+}
+
+
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
 	 "Test that one node at a time can be stopped and then restarted "\
@@ -1452,6 +1505,9 @@ TESTCASE("Bug26457", ""){
 TESTCASE("Bug26481", ""){
   INITIALIZER(runBug26481);
 }
+TESTCASE("Bug27003", ""){
+  INITIALIZER(runBug27003);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -425,6 +425,10 @@ max-time: 500
 cmd: testScan
 args: -n Bug24447 T1
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug27003 T1
+
 max-time: 500
 cmd: testNodeRestart
 args: -n Bug15587 T1