bk commit into 5.0 tree (jonas:1.2362)

[jonas]

Below is the list of changes that have just been committed into a local
5.0 repository of jonas. When jonas does a push these changes will
be propagated to the main repository and, within 24 hours after the
push, to the public repository.
For information on how to access the public repository
see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html

ChangeSet@stripped, 2007-01-11 21:12:14+01:00, jonas@stripped +5 -0
  Merge joreland@stripped:/home/bk/mysql-5.0-ndb
  into  perch.ndb.mysql.com:/home/jonas/src/mysql-5.0-ndb
  MERGE: 1.2294.4.34

  ndb/include/kernel/signaldata/DumpStateOrd.hpp@stripped, 2007-01-11 21:12:09+01:00,
jonas@stripped +0 -0
    Auto merged
    MERGE: 1.8.1.1

  ndb/src/kernel/blocks/dbdih/DbdihMain.cpp@stripped, 2007-01-11 21:12:09+01:00,
jonas@stripped +0 -0
    Auto merged
    MERGE: 1.63.1.1

  ndb/src/kernel/blocks/qmgr/Qmgr.hpp@stripped, 2007-01-11 21:12:09+01:00,
jonas@stripped +0 -0
    Auto merged
    MERGE: 1.11.1.1

  ndb/src/kernel/blocks/qmgr/QmgrMain.cpp@stripped, 2007-01-11 21:12:09+01:00,
jonas@stripped +0 -0
    Auto merged
    MERGE: 1.36.1.1

  ndb/test/ndbapi/testNodeRestart.cpp@stripped, 2007-01-11 21:12:09+01:00,
jonas@stripped +0 -0
    Auto merged
    MERGE: 1.24.1.2

# This is a BitKeeper patch.  What follows are the unified diffs for the
# set of deltas contained in the patch.  The rest of the patch, the part
# that BitKeeper cares about, is below these diffs.
# User:	jonas
# Host:	perch.ndb.mysql.com
# Root:	/home/jonas/src/mysql-5.0-ndb/RESYNC

--- 1.9/ndb/include/kernel/signaldata/DumpStateOrd.hpp	2007-01-11 21:12:19 +01:00
+++ 1.10/ndb/include/kernel/signaldata/DumpStateOrd.hpp	2007-01-11 21:12:19 +01:00
@@ -67,6 +67,7 @@
     // 100-105 TUP and ACC  
     // 200-240 UTIL
     // 300-305 TRIX
+    QmgrErr935 = 935,
     NdbfsDumpFileStat = 400,
     NdbfsDumpAllFiles = 401,
     NdbfsDumpOpenFiles = 402,

--- 1.64/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp	2007-01-11 21:12:19 +01:00
+++ 1.65/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp	2007-01-11 21:12:19 +01:00
@@ -3554,7 +3554,6 @@
   takeOverPtr.i = takeOverPtrI;
   ptrCheckGuard(takeOverPtr, MAX_NDB_NODES, takeOverRecord);
 
-  releaseTakeOver(takeOverPtrI);
   if ((takeOverPtr.p->toMasterStatus != TakeOverRecord::IDLE) &&
       (takeOverPtr.p->toMasterStatus != TakeOverRecord::TO_WAIT_START_TAKE_OVER)) {
     jam();
@@ -3568,6 +3567,7 @@
   }//if
   setAllowNodeStart(takeOverPtr.p->toStartingNode, true);
   initTakeOver(takeOverPtr);
+  releaseTakeOver(takeOverPtrI);
 }//Dbdih::endTakeOver()
 
 void Dbdih::releaseTakeOver(Uint32 takeOverPtrI)
@@ -4709,6 +4709,7 @@
       break;
     }
     ndbrequire(ok);
+    endTakeOver(takeOverPtr.i);
   }//if
 }//Dbdih::handleTakeOverNewMaster()
 

--- 1.12/ndb/src/kernel/blocks/qmgr/Qmgr.hpp	2007-01-11 21:12:19 +01:00
+++ 1.13/ndb/src/kernel/blocks/qmgr/Qmgr.hpp	2007-01-11 21:12:19 +01:00
@@ -425,6 +425,10 @@
   
   StopReq c_stopReq;
   bool check_multi_node_shutdown(Signal* signal);
+
+#ifdef ERROR_INSERT
+  Uint32 c_error_insert_extra;
+#endif
 };
 
 #endif

--- 1.37/ndb/src/kernel/blocks/qmgr/QmgrMain.cpp	2007-01-11 21:12:19 +01:00
+++ 1.38/ndb/src/kernel/blocks/qmgr/QmgrMain.cpp	2007-01-11 21:12:19 +01:00
@@ -2894,6 +2894,17 @@
     systemErrorLab(signal, __LINE__);
     return;
   }//if
+
+  if (getNodeState().startLevel < NodeState::SL_STARTED)
+  {
+    jam();
+    CRASH_INSERTION(932);
+    char buf[100];
+    BaseString::snprintf(buf, 100, "Node failure during restart");
+    progError(__LINE__, NDBD_EXIT_SR_OTHERNODEFAILED, buf);
+    ndbrequire(false);
+  }
+
   TnoFailedNodes = cnoFailedNodes;
   failReport(signal, failedNodePtr.i, (UintR)ZTRUE, aFailCause);
   if (cpresident == getOwnNodeId()) {
@@ -2980,6 +2991,16 @@
     return;
   }//if
 
+  if (getNodeState().startLevel < NodeState::SL_STARTED)
+  {
+    jam();
+    CRASH_INSERTION(932);
+    char buf[100];
+    BaseString::snprintf(buf, 100, "Node failure during restart");
+    progError(__LINE__, NDBD_EXIT_SR_OTHERNODEFAILED, buf);
+    ndbrequire(false);
+  }
+
   guard0 = cnoPrepFailedNodes - 1;
   arrGuard(guard0, MAX_NDB_NODES);
   for (Tindex = 0; Tindex <= guard0; Tindex++) {
@@ -3157,6 +3178,18 @@
   for (nodePtr.i = 1; nodePtr.i < MAX_NDB_NODES; nodePtr.i++) {
     jam();
     ptrAss(nodePtr, nodeRec);
+
+#ifdef ERROR_INSERT    
+    if (ERROR_INSERTED(935) && nodePtr.i == c_error_insert_extra)
+    {
+      ndbout_c("skipping node %d", c_error_insert_extra);
+      CLEAR_ERROR_INSERT_VALUE;
+      signal->theData[0] = 9999;
+      sendSignalWithDelay(CMVMI_REF, GSN_NDB_TAMPER, signal, 1000, 1);
+      continue;
+    }
+#endif
+
     if (nodePtr.p->phase == ZRUNNING) {
       jam();
       nodePtr.p->sendCommitFailReqStatus = Q_ACTIVE;
@@ -3227,6 +3260,33 @@
   return;
 }//Qmgr::execPREP_FAILREF()
 
+static
+Uint32
+clear_nodes(Uint32 dstcnt, Uint16 dst[], Uint32 srccnt, const Uint16 src[])
+{
+  if (srccnt == 0)
+    return dstcnt;
+  
+  Uint32 pos = 0;
+  for (Uint32 i = 0; i<dstcnt; i++)
+  {
+    Uint32 node = dst[i];
+    for (Uint32 j = 0; j<srccnt; j++)
+    {
+      if (node == dst[j])
+      {
+	node = RNIL;
+	break;
+      }
+    }
+    if (node != RNIL)
+    {
+      dst[pos++] = node;
+    }
+  }
+  return pos;
+}
+
 /*---------------------------------------------------------------------------*/
 /*    THE PRESIDENT IS NOW COMMITTING THE PREVIOUSLY PREPARED NODE FAILURE.  */
 /*---------------------------------------------------------------------------*/
@@ -3314,19 +3374,18 @@
 		   NodeFailRep::SignalLength, JBB);
       }//if
     }//for
-    if (cpresident != getOwnNodeId()) {
-      jam();
-      cnoFailedNodes = cnoCommitFailedNodes - cnoFailedNodes;
-      if (cnoFailedNodes > 0) {
-        jam();
-        guard0 = cnoFailedNodes - 1;
-        arrGuard(guard0 + cnoCommitFailedNodes, MAX_NDB_NODES);
-        for (Tj = 0; Tj <= guard0; Tj++) {
-          jam();
-          cfailedNodes[Tj] = cfailedNodes[Tj + cnoCommitFailedNodes];
-        }//for
-      }//if
-    }//if
+
+    /**
+     * Remove committed nodes from failed/prepared
+     */
+    cnoFailedNodes = clear_nodes(cnoFailedNodes, 
+				 cfailedNodes, 
+				 cnoCommitFailedNodes, 
+				 ccommitFailedNodes);
+    cnoPrepFailedNodes = clear_nodes(cnoPrepFailedNodes, 
+				     cprepFailedNodes,
+				     cnoCommitFailedNodes,
+				     ccommitFailedNodes);
     cnoCommitFailedNodes = 0;
   }//if
   /**-----------------------------------------------------------------------
@@ -4705,6 +4764,14 @@
   default:
     ;
   }//switch
+
+#ifdef ERROR_INSERT
+  if (signal->theData[0] == 935 && signal->getLength() == 2)
+  {
+    SET_ERROR_INSERT_VALUE(935);
+    c_error_insert_extra = signal->theData[1];
+  }
+#endif
 }//Qmgr::execDUMP_STATE_ORD()
 
 void Qmgr::execSET_VAR_REQ(Signal* signal) 

--- 1.25/ndb/test/ndbapi/testNodeRestart.cpp	2007-01-11 21:12:19 +01:00
+++ 1.26/ndb/test/ndbapi/testNodeRestart.cpp	2007-01-11 21:12:19 +01:00
@@ -954,6 +954,96 @@
   return NDBT_OK;
 }
 
+int runBug25364(NDBT_Context* ctx, NDBT_Step* step){
+  int result = NDBT_OK;
+  NdbRestarter restarter;
+  Ndb* pNdb = GETNDB(step);
+  int loops = ctx->getNumLoops();
+  
+  if (restarter.getNumDbNodes() < 4)
+    return NDBT_OK;
+
+  int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+
+  for (; loops; loops --)
+  {
+    int master = restarter.getMasterNodeId();
+    int victim = restarter.getRandomNodeOtherNodeGroup(master, rand());
+    int second = restarter.getRandomNodeSameNodeGroup(victim, rand());
+    
+    int dump[] = { 935, victim } ;
+    if (restarter.dumpStateOneNode(master, dump, 2))
+      return NDBT_FAILED;
+  
+    if (restarter.dumpStateOneNode(master, val2, 2))
+      return NDBT_FAILED;
+  
+    if (restarter.restartOneDbNode(second, false, true, true))
+      return NDBT_FAILED;
+
+    int nodes[2] = { master, second };
+    if (restarter.waitNodesNoStart(nodes, 2))
+      return NDBT_FAILED;
+
+    restarter.startNodes(nodes, 2);
+
+    if (restarter.waitNodesStarted(nodes, 2))
+      return NDBT_FAILED;
+  }
+  
+  return NDBT_OK;
+}
+
+int runBug25554(NDBT_Context* ctx, NDBT_Step* step){
+  
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  NdbRestarter restarter;
+  
+  if (restarter.getNumDbNodes() < 4)
+    return NDBT_OK;
+
+  for (int i = 0; i<loops; i++)
+  {
+    int master = restarter.getMasterNodeId();
+    int node1 = restarter.getRandomNodeOtherNodeGroup(master, rand());
+    restarter.restartOneDbNode(node1, false, true, true);
+
+    int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+  
+    if (restarter.dumpStateOneNode(master, val2, 2))
+      return NDBT_FAILED;
+
+    if (restarter.insertErrorInNode(master, 7141))
+      return NDBT_FAILED;
+
+    if (restarter.waitNodesNoStart(&node1, 1))
+      return NDBT_FAILED;
+
+    if (restarter.dumpStateOneNode(node1, val2, 2))
+      return NDBT_FAILED;
+
+    if (restarter.insertErrorInNode(node1, 932))
+      return NDBT_FAILED;
+
+    if (restarter.startNodes(&node1, 1))
+      return NDBT_FAILED;
+
+    int nodes[] = { master, node1 };
+    if (restarter.waitNodesNoStart(nodes, 2))
+      return NDBT_FAILED;
+
+    if (restarter.startNodes(nodes, 2))
+      return NDBT_FAILED;
+
+    if (restarter.waitClusterStarted())
+      return NDBT_FAILED;
+  }    
+
+  return NDBT_OK;
+}
+
 
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
@@ -1269,6 +1359,12 @@
 }
 TESTCASE("Bug24717", ""){
   INITIALIZER(runBug24717);
+}
+TESTCASE("Bug25364", ""){
+  INITIALIZER(runBug25364);
+}
+TESTCASE("Bug25554", ""){
+  INITIALIZER(runBug25554);
 }
 NDBT_TESTSUITE_END(testNodeRestart);