ndb - bug#29167

  Fix case where all node in node group dies before they saved sysfile (wrt gcp)
  and Qmgr incorrectly thinks that "node group is missing"

storage/ndb/src/kernel/blocks/ERROR_codes.txt

@@ -5,7 +5,7 @@ Next DBACC 3002
 Next DBTUP 4029
 Next DBLQH 5045
 Next DBDICT 6007
-Next DBDIH 7183
+Next DBDIH 7184
 Next DBTC 8040
 Next CMVMI 9000
 Next BACKUP 10038
@@ -75,6 +75,8 @@ Delay GCP_SAVEREQ by 10 secs
 
 7180: Crash master during master-take-over in execMASTER_LCPCONF
 
+7183: Crash when receiving COPY_GCIREQ
+
 ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:
 -----------------------------------------------------------------
 

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

@@ -747,6 +747,8 @@ done:
   }
   ndbrequire(ok);
   
+  CRASH_INSERTION(7183);
+  
   /* ----------------------------------------------------------------------- */
   /*     WE START BY TRYING TO OPEN THE FIRST RESTORABLE GCI FILE.           */
   /* ----------------------------------------------------------------------- */
@@ -1230,6 +1232,17 @@ void Dbdih::execDIH_RESTARTREQ(Signal* signal)
 	Uint32 ng = Sysfile::getNodeGroup(i, SYSFILE->nodeGroups);
 	ndbrequire(ng < MAX_NDB_NODES);
 	Uint32 gci = node_gcis[i];
+        if (gci < SYSFILE->lastCompletedGCI[i])
+        {
+          jam();
+          /**
+           * Handle case, where *I* know that node complete GCI
+           *   but node does not...bug#29167
+           *   i.e node died before it wrote own sysfile
+           */
+          gci = SYSFILE->lastCompletedGCI[i];
+        }
+
 	if (gci > node_group_gcis[ng])
 	{
 	  jam();

storage/ndb/test/ndbapi/testSystemRestart.cpp

@@ -1219,6 +1219,48 @@ runBug24664(NDBT_Context* ctx, NDBT_Step* step)
   return result;
 }
 
+int 
+runBug29167(NDBT_Context* ctx, NDBT_Step* step)
+{
+  int result = NDBT_OK;
+  NdbRestarter restarter;
+  Ndb* pNdb = GETNDB(step);
+  const Uint32 nodeCount = restarter.getNumDbNodes();
+
+  if (nodeCount < 2)
+    return NDBT_OK;
+
+  int filter[] = { 15, NDB_MGM_EVENT_CATEGORY_CHECKPOINT, 0 };
+  NdbLogEventHandle handle = 
+    ndb_mgm_create_logevent_handle(restarter.handle, filter);
+
+  struct ndb_logevent event;
+  int master = restarter.getMasterNodeId();
+  do {
+    int node1 = restarter.getRandomNodeOtherNodeGroup(master, rand());
+    int node2 = restarter.getRandomNodeSameNodeGroup(node1, rand());
+    
+    int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };    
+    restarter.dumpStateAllNodes(val2, 2);
+    int dump[] = { DumpStateOrd::DihSetTimeBetweenGcp, 30000 };
+    restarter.dumpStateAllNodes(dump, 2);
+    
+    while(ndb_logevent_get_next(handle, &event, 0) >= 0 &&
+          event.type != NDB_LE_GlobalCheckpointCompleted);
+    
+    CHECK(restarter.insertErrorInAllNodes(932) == 0);
+    
+    CHECK(restarter.insertErrorInNode(node1, 7183) == 0);
+    CHECK(restarter.insertErrorInNode(node2, 7183) == 0);
+
+    CHECK(restarter.waitClusterNoStart() == 0);
+    restarter.startAll();
+    CHECK(restarter.waitClusterStarted() == 0);  
+  } while(false);
+  
+  return result;
+}
+
 NDBT_TESTSUITE(testSystemRestart);
 TESTCASE("SR1", 
 	 "Basic system restart test. Focus on testing restart from REDO log.\n"
@@ -1399,6 +1441,12 @@ TESTCASE("Bug24664",
   STEP(runBug24664);
   FINALIZER(runClearTable);
 }
+TESTCASE("Bug29167", "")
+{
+  INITIALIZER(runWaitStarted);
+  STEP(runBug29167);
+}
+
 NDBT_TESTSUITE_END(testSystemRestart);
 
 int main(int argc, const char** argv){

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

@@ -485,6 +485,10 @@ max-time: 1000
 cmd: testNodeRestart
 args: -n Bug27003 T1
 
+max-time: 300
+cmd: testSystemRestart
+args: -n Bug29167 T1
+
 max-time: 1000
 cmd: testNodeRestart
 args: -n Bug27283 T1