bug#28717, make sure only master updates activeStatus

  so that othernodes dont get confused after having recevied status from master
  and then tries to update it self


ndb/src/kernel/blocks/ERROR_codes.txt:
  error 1001, delay node_failrep
ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp:
  error 1001, delay node_failrep
ndb/test/ndbapi/testNodeRestart.cpp:
  testcase
ndb/test/run-test/daily-basic-tests.txt:
  testcase

5 files changed, 105 insertions, 2 deletions

diff --git a/ndb/src/kernel/blocks/ERROR_codes.txt b/ndb/src/kernel/blocks/ERROR_codes.txt
index 0bcc99a6334..bf54d583299 100644
--- a/ndb/src/kernel/blocks/ERROR_codes.txt
+++ b/ndb/src/kernel/blocks/ERROR_codes.txt
@@ -1,5 +1,5 @@
 Next QMGR 1
-Next NDBCNTR 1000
+Next NDBCNTR 1002
 Next NDBFS 2000
 Next DBACC 3002
 Next DBTUP 4014
@@ -487,3 +487,8 @@ Dbdict:
 6003 Crash in participant @ CreateTabReq::Prepare
 6004 Crash in participant @ CreateTabReq::Commit
 6005 Crash in participant @ CreateTabReq::CreateDrop
+
+Ndbcntr:
+--------
+
+1001: Delay sending NODE_FAILREP (to own node), until error is cleared
diff --git a/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp b/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
index f24a8e2c7d5..44e2293f318 100644
--- a/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
+++ b/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
@@ -4448,12 +4448,18 @@ void Dbdih::failedNodeLcpHandling(Signal* signal, NodeRecordPtr failedNodePtr)
   jam();
   const Uint32 nodeId = failedNodePtr.i;
 
-  if (c_lcpState.m_participatingLQH.get(failedNodePtr.i)){
+  if (isMaster() && c_lcpState.m_participatingLQH.get(failedNodePtr.i))
+  {
     /*----------------------------------------------------*/
     /*  THE NODE WAS INVOLVED IN A LOCAL CHECKPOINT. WE   */
     /* MUST UPDATE THE ACTIVE STATUS TO INDICATE THAT     */
     /* THE NODE HAVE MISSED A LOCAL CHECKPOINT.           */
     /*----------------------------------------------------*/
+
+    /**
+     * Bug#28717, Only master should do this, as this status is copied
+     *   to other nodes
+     */
     switch (failedNodePtr.p->activeStatus) {
     case Sysfile::NS_Active:
       jam();
diff --git a/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp b/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp
index 26e8f246293..65d80669316 100644
--- a/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp
+++ b/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp
@@ -1375,6 +1375,13 @@ void Ndbcntr::execNODE_FAILREP(Signal* signal)
 {
   jamEntry();
 
+  if (ERROR_INSERTED(1001))
+  {
+    sendSignalWithDelay(reference(), GSN_NODE_FAILREP, signal, 100, 
+                        signal->getLength());
+    return;
+  }
+  
   const NodeFailRep * nodeFail = (NodeFailRep *)&signal->theData[0];
   NdbNodeBitmask allFailed; 
   allFailed.assign(NdbNodeBitmask::Size, nodeFail->theNodes);
diff --git a/ndb/test/ndbapi/testNodeRestart.cpp b/ndb/test/ndbapi/testNodeRestart.cpp
index 9adbfbd46a6..e5ced961b6f 100644
--- a/ndb/test/ndbapi/testNodeRestart.cpp
+++ b/ndb/test/ndbapi/testNodeRestart.cpp
@@ -1045,6 +1045,84 @@ int runBug25554(NDBT_Context* ctx, NDBT_Step* step){
 }
 
 
+int
+runBug28717(NDBT_Context* ctx, NDBT_Step* step)
+{
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  Ndb* pNdb = GETNDB(step);
+  NdbRestarter res;
+
+  if (res.getNumDbNodes() < 4)
+  {
+    return NDBT_OK;
+  }
+
+  int master = res.getMasterNodeId();
+  int node0 = res.getRandomNodeOtherNodeGroup(master, rand());
+  int node1 = res.getRandomNodeSameNodeGroup(node0, rand());
+  
+  ndbout_c("master: %d node0: %d node1: %d", master, node0, node1);
+  
+  if (res.restartOneDbNode(node0, false, true, true))
+  {
+    return NDBT_FAILED;
+  }
+
+  {
+    int filter[] = { 15, NDB_MGM_EVENT_CATEGORY_CHECKPOINT, 0 };
+    NdbLogEventHandle handle = 
+      ndb_mgm_create_logevent_handle(res.handle, filter);
+    
+
+    int dump[] = { DumpStateOrd::DihStartLcpImmediately };
+    struct ndb_logevent event;
+    
+    for (Uint32 i = 0; i<3; i++)
+    {
+      res.dumpStateOneNode(master, dump, 1);
+      while(ndb_logevent_get_next(handle, &event, 0) >= 0 &&
+            event.type != NDB_LE_LocalCheckpointStarted);
+      while(ndb_logevent_get_next(handle, &event, 0) >= 0 &&
+            event.type != NDB_LE_LocalCheckpointCompleted);
+    } 
+  }
+  
+  if (res.waitNodesNoStart(&node0, 1))
+    return NDBT_FAILED;
+  
+  int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+  
+  if (res.dumpStateOneNode(node0, val2, 2))
+    return NDBT_FAILED;
+  
+  if (res.insertErrorInNode(node0, 5010))
+    return NDBT_FAILED;
+  
+  if (res.insertErrorInNode(node1, 1001))
+    return NDBT_FAILED;
+  
+  if (res.startNodes(&node0, 1))
+    return NDBT_FAILED;
+  
+  NdbSleep_SecSleep(3);
+
+  if (res.insertErrorInNode(node1, 0))
+    return NDBT_FAILED;
+
+  if (res.waitNodesNoStart(&node0, 1))
+    return NDBT_FAILED;
+
+  if (res.startNodes(&node0, 1))
+    return NDBT_FAILED;
+
+  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 "\
@@ -1366,6 +1444,9 @@ TESTCASE("Bug25364", ""){
 TESTCASE("Bug25554", ""){
   INITIALIZER(runBug25554);
 }
+TESTCASE("Bug28717", ""){
+  INITIALIZER(runBug28717);
+}
 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 d2c91279d18..5a3947ec1e9 100644
--- a/ndb/test/run-test/daily-basic-tests.txt
+++ b/ndb/test/run-test/daily-basic-tests.txt
@@ -492,6 +492,10 @@ max-time: 1500
 cmd: testDict
 args: -n CreateAndDrop 
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug28717 T1
+
 max-time: 1500
 cmd: testDict
 args: -n CreateAndDropAtRandom -l 200 T1