bug#15587 - ndb

  Fix error in NF during NR


ndb/include/kernel/signaldata/DumpStateOrd.hpp:
  Add dump for ERROR 5002 with specified table
ndb/src/kernel/blocks/ERROR_codes.txt:
  Add dump for ERROR 5002 with specified table
ndb/src/kernel/blocks/dbdih/DbdihMain.cpp:
  Run updateNodeInfo if failed node is not in list of storedReplicas
ndb/src/kernel/blocks/dblqh/Dblqh.hpp:
  Add dump for ERROR 5002 with specified table
ndb/src/kernel/blocks/dblqh/DblqhMain.cpp:
  Add dump for ERROR 5002 with specified table
ndb/test/ndbapi/testNodeRestart.cpp:
  Add testcase for bug#15587
ndb/test/run-test/daily-basic-tests.txt:
  Add testcase for bug#15587

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/src/kernel/blocks/ERROR_codes.txt

@@ -155,6 +155,9 @@ 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

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

@@ -5187,15 +5187,16 @@ 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){
@@ -5211,6 +5212,15 @@ 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

@@ -2881,6 +2881,7 @@ private:
   UintR ctransidHash[1024];
   
   Uint32 c_diskless;
+  Uint32 c_error_insert_table_id;
   
 public:
   /**

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

@@ -3532,6 +3532,7 @@ 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
@@ -18402,8 +18403,12 @@ Dblqh::execDUMP_STATE_ORD(Signal* signal)
     return;
   }
 
-
-
+  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/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 +410,43 @@ 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;
+}
+
+
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
 	 "Test that one node at a time can be stopped and then restarted "\
@@ -671,6 +709,13 @@ TESTCASE("LateCommit",
   STEP(runLateCommit);
   FINALIZER(runClearTable);
 }
+TESTCASE("Bug15587",
+	 "Test bug with NF during NR"){
+  INITIALIZER(runLoadTable);
+  STEP(runScanUpdateUntilStopped);
+  STEP(runBug15587);
+  FINALIZER(runClearTable);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -434,6 +434,10 @@ max-time: 500
 cmd: testScan
 args: -l 100 -n Scan-bug8262 T7
 
+max-time: 500
+cmd: testNodeRestart
+args: -n BugBug15587 T1
+
 # OLD FLEX
 max-time: 500
 cmd: flexBench