ndb - bug#27283 (wl2325-5.0)

    Handle race condtition between MASTER_GCPCONF and execGCP_NODEFINISH

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

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) 
@@ -7542,10 +7559,10 @@ 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);
@@ -7692,6 +7709,15 @@ 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;

ndb/test/ndbapi/testNodeRestart.cpp

@@ -1178,6 +1178,48 @@ runBug27003(NDBT_Context* ctx, NDBT_Step* step)
 }
 
 
+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 "\
@@ -1508,6 +1550,9 @@ TESTCASE("Bug26481", ""){
 TESTCASE("Bug27003", ""){
   INITIALIZER(runBug27003);
 }
+TESTCASE("Bug27283", ""){
+  INITIALIZER(runBug27283);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -429,6 +429,10 @@ 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