ndb - testNodeRestart -n Bug27466 T1

  make local INCL_NODEREQ/CONF more robust (to remove testprg introduced race, error insert causes 5s delay)

7 files changed, 48 insertions, 19 deletions

diff --git a/storage/ndb/src/kernel/blocks/backup/Backup.cpp b/storage/ndb/src/kernel/blocks/backup/Backup.cpp
index 64e2c41aa69..45501bf50d5 100644
--- a/storage/ndb/src/kernel/blocks/backup/Backup.cpp
+++ b/storage/ndb/src/kernel/blocks/backup/Backup.cpp
@@ -1026,8 +1026,9 @@ Backup::execINCL_NODEREQ(Signal* signal)
       break;
     }//if
   }//for
-  signal->theData[0] = reference();
-  sendSignal(senderRef, GSN_INCL_NODECONF, signal, 1, JBB);
+  signal->theData[0] = inclNode;
+  signal->theData[1] = reference();
+  sendSignal(senderRef, GSN_INCL_NODECONF, signal, 2, JBB);
 }
 
 /*****************************************************************************
diff --git a/storage/ndb/src/kernel/blocks/dbdict/Dbdict.cpp b/storage/ndb/src/kernel/blocks/dbdict/Dbdict.cpp
index 569958a6aa9..7ced078144a 100644
--- a/storage/ndb/src/kernel/blocks/dbdict/Dbdict.cpp
+++ b/storage/ndb/src/kernel/blocks/dbdict/Dbdict.cpp
@@ -3825,8 +3825,9 @@ void Dbdict::execINCL_NODEREQ(Signal* signal)
   c_nodes.getPtr(nodePtr);
   ndbrequire(nodePtr.p->nodeState == NodeRecord::NDB_NODE_DEAD);
   nodePtr.p->nodeState = NodeRecord::NDB_NODE_ALIVE;
-  signal->theData[0] = reference();
-  sendSignal(retRef, GSN_INCL_NODECONF, signal, 1, JBB);
+  signal->theData[0] = nodePtr.i;
+  signal->theData[1] = reference();
+  sendSignal(retRef, GSN_INCL_NODECONF, signal, 2, JBB);
 
   c_aliveNodes.set(nodePtr.i);
 }//execINCL_NODEREQ()
diff --git a/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp b/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
index a081166f2e3..bbacb300089 100644
--- a/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
+++ b/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
@@ -2135,12 +2135,9 @@ void Dbdih::gcpBlockedLab(Signal* signal)
 /*---------------------------------------------------------------------------*/
 void Dbdih::execINCL_NODECONF(Signal* signal) 
 {
-  Uint32 TsendNodeId;
-  Uint32 TstartNode_or_blockref;
-  
   jamEntry();
-  TstartNode_or_blockref = signal->theData[0];
-  TsendNodeId = signal->theData[1];
+  Uint32 TstartNode = signal->theData[0];
+  Uint32 TsendNodeId_or_blockref = signal->theData[1];
 
   Uint32 blocklist[6];
   blocklist[0] = clocallqhblockref;
@@ -2152,9 +2149,21 @@ void Dbdih::execINCL_NODECONF(Signal* signal)
   
   for (Uint32 i = 0; blocklist[i] != 0; i++)
   {
-    if (TstartNode_or_blockref == blocklist[i])
+    if (TsendNodeId_or_blockref == blocklist[i])
     {
       jam();
+
+      if (TstartNode != c_nodeStartSlave.nodeId)
+      {
+        jam();
+        warningEvent("Recevied INCL_NODECONF for %u from %s"
+                     " while %u is starting",
+                     TstartNode,
+                     getBlockName(refToBlock(TsendNodeId_or_blockref)),
+                     c_nodeStartSlave.nodeId);
+        return;
+      }
+      
       if (getNodeStatus(c_nodeStartSlave.nodeId) == NodeRecord::ALIVE && 
 	  blocklist[i+1] != 0)
       {
@@ -2182,10 +2191,21 @@ void Dbdih::execINCL_NODECONF(Signal* signal)
       }
     }
   }
+
+  if (c_nodeStartMaster.startNode != TstartNode)
+  {
+    jam();
+    warningEvent("Recevied INCL_NODECONF for %u from %u"
+                 " while %u is starting",
+                 TstartNode,
+                 TsendNodeId_or_blockref,
+                 c_nodeStartMaster.startNode);
+    return;
+  }
   
   ndbrequire(cmasterdihref = reference());
-  receiveLoopMacro(INCL_NODEREQ, TsendNodeId);
-
+  receiveLoopMacro(INCL_NODEREQ, TsendNodeId_or_blockref);
+  
   CRASH_INSERTION(7128);
   /*-------------------------------------------------------------------------*/
   // Now that we have included the starting node in the node lists in the
@@ -13006,6 +13026,7 @@ void Dbdih::newCrashedReplica(Uint32 nodeId, ReplicaRecordPtr ncrReplicaPtr)
 void Dbdih::nodeResetStart()
 {
   jam();
+  c_nodeStartSlave.nodeId = 0;
   c_nodeStartMaster.startNode = RNIL;
   c_nodeStartMaster.failNr = cfailurenr;
   c_nodeStartMaster.activeState = false;
diff --git a/storage/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp b/storage/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
index 6efa1b1b116..83d38595c1f 100644
--- a/storage/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
+++ b/storage/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp
@@ -494,8 +494,9 @@ void Dblqh::execINCL_NODEREQ(Signal* signal)
       cnodeStatus[i] = ZNODE_UP;
     }//if
   }//for
-  signal->theData[0] = cownref; 
-  sendSignal(retRef, GSN_INCL_NODECONF, signal, 1, JBB);
+  signal->theData[0] = nodeId;
+  signal->theData[1] = cownref; 
+  sendSignal(retRef, GSN_INCL_NODECONF, signal, 2, JBB);
   return;
 }//Dblqh::execINCL_NODEREQ()
 
diff --git a/storage/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp b/storage/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp
index ce20059e663..cffb4b22e4d 100644
--- a/storage/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp
+++ b/storage/ndb/src/kernel/blocks/dbtc/DbtcMain.cpp
@@ -310,9 +310,11 @@ void Dbtc::execINCL_NODEREQ(Signal* signal)
   hostptr.i = signal->theData[1];
   ptrCheckGuard(hostptr, chostFilesize, hostRecord);
   hostptr.p->hostStatus = HS_ALIVE;
-  signal->theData[0] = cownref;
   c_alive_nodes.set(hostptr.i);
 
+  signal->theData[0] = hostptr.i;
+  signal->theData[1] = cownref;
+
   if (ERROR_INSERTED(8039))
   {
     CLEAR_ERROR_INSERT_VALUE;
@@ -321,11 +323,11 @@ void Dbtc::execINCL_NODEREQ(Signal* signal)
     sendSignal(numberToRef(CMVMI, hostptr.i), 
 	       GSN_NDB_TAMPER, signal, 1, JBB);
     signal->theData[0] = save;
-    sendSignalWithDelay(tblockref, GSN_INCL_NODECONF, signal, 5000, 1);
+    sendSignalWithDelay(tblockref, GSN_INCL_NODECONF, signal, 5000, 2);
     return;
   }
 
-  sendSignal(tblockref, GSN_INCL_NODECONF, signal, 1, JBB);
+  sendSignal(tblockref, GSN_INCL_NODECONF, signal, 2, JBB);
 }
 
 void Dbtc::execREAD_NODESREF(Signal* signal) 
diff --git a/storage/ndb/src/kernel/blocks/suma/Suma.cpp b/storage/ndb/src/kernel/blocks/suma/Suma.cpp
index af1334c57fc..113b63a19d3 100644
--- a/storage/ndb/src/kernel/blocks/suma/Suma.cpp
+++ b/storage/ndb/src/kernel/blocks/suma/Suma.cpp
@@ -821,8 +821,9 @@ Suma::execINCL_NODEREQ(Signal* signal){
   ndbrequire(!c_alive_nodes.get(nodeId));
   c_alive_nodes.set(nodeId);
   
-  signal->theData[0] = reference();
-  sendSignal(senderRef, GSN_INCL_NODECONF, signal, 1, JBB);
+  signal->theData[0] = nodeId;
+  signal->theData[1] = reference();
+  sendSignal(senderRef, GSN_INCL_NODECONF, signal, 2, JBB);
 }
 
 void
diff --git a/storage/ndb/test/ndbapi/testNodeRestart.cpp b/storage/ndb/test/ndbapi/testNodeRestart.cpp
index 2a5febb7ae9..751134c43c5 100644
--- a/storage/ndb/test/ndbapi/testNodeRestart.cpp
+++ b/storage/ndb/test/ndbapi/testNodeRestart.cpp
@@ -1590,6 +1590,8 @@ runBug27466(NDBT_Context* ctx, NDBT_Step* step)
       node2 = res.getDbNodeId(rand() % res.getNumDbNodes());
     }
 
+    ndbout_c("nodes %u %u", node1, node2);
+
     if (res.restartOneDbNode(node1, false, true, true))
       return NDBT_FAILED;