Merge mskold@bk-internal.mysql.com:/home/bk/mysql-5.0-ndb

into  mysql.com:/windows/Linux_space/MySQL/mysql-5.0-ndb

mysql-test/r/ndb_backup_print.result

+use test;
+drop table if exists t1,t2,t3,t4,t5,t6,t7,t8,t9,t10;
+Connected to Management Server at: :
+Waiting for completed, this may take several minutes
+Backup  started from node 
+Backup  started from node  completed
+ StartGCP:  StopGCP: 
+ #Records:  #LogRecords: 
+ Data:  bytes Log:  bytes
+create table t1
+(pk int key
+,a1 BIT(1), a2 BIT(5), a3 BIT(33), a4 BIT(63), a5 BIT(64)
+,b1 TINYINT, b2 TINYINT UNSIGNED
+,c1 SMALLINT, c2 SMALLINT UNSIGNED
+,d1 INT, d2 INT UNSIGNED
+,e1 BIGINT, e2 BIGINT UNSIGNED
+,f1 CHAR(1) BINARY, f2 CHAR(32) BINARY, f3 CHAR(255) BINARY
+,g1 VARCHAR(32) BINARY, g2 VARCHAR(255) BINARY, g3 VARCHAR(1000) BINARY
+,h1 BINARY(1), h2 BINARY(8), h3 BINARY(255)
+,i1 VARBINARY(32), i2 VARBINARY(255), i3 VARBINARY(1000)
+) engine ndb;
+insert into t1 values
+(1
+,0x1, 0x17, 0x789a, 0x789abcde, 0xfedc0001
+,127, 255
+,32767, 65535
+,2147483647, 4294967295
+,9223372036854775807, 18446744073709551615
+,'1','12345678901234567890123456789012','123456789'
+  ,'1','12345678901234567890123456789012','123456789'
+  ,0x12,0x123456789abcdef0, 0x012345
+,0x12,0x123456789abcdef0, 0x00123450
+);
+insert into t1 values
+(2
+,0, 0, 0, 0, 0
+,-128, 0
+,-32768, 0
+,-2147483648, 0
+,-9223372036854775808, 0
+,'','',''
+  ,'','',''
+  ,0x0,0x0,0x0
+,0x0,0x0,0x0
+);
+insert into t1 values
+(3
+,NULL,NULL,NULL,NULL,NULL
+,NULL,NULL
+,NULL,NULL
+,NULL,NULL
+,NULL,NULL
+,NULL,NULL,NULL
+,NULL,NULL,NULL
+,NULL,NULL,NULL
+,NULL,NULL,NULL
+);
+Connected to Management Server at: :
+Waiting for completed, this may take several minutes
+Backup  started from node 
+Backup  started from node  completed
+ StartGCP:  StopGCP: 
+ #Records:  #LogRecords: 
+ Data:  bytes Log:  bytes

mysql-test/t/ndb_backup_print.test

+-- source include/have_ndb.inc
+-- source include/ndb_default_cluster.inc
+-- source include/not_embedded.inc
+
+--disable_warnings
+use test;
+drop table if exists t1,t2,t3,t4,t5,t6,t7,t8,t9,t10;
+--enable_warnings
+
+#NO.1 test output of backup
+--exec $NDB_TOOLS_DIR/../src/mgmclient/ndb_mgm -e "start backup" |sed -e 's/[0-9]//g' |sed -e 's/localhost//g' |sed -e 's/\.\.\.*//g'
+
+create table t1
+ (pk int key
+  ,a1 BIT(1), a2 BIT(5), a3 BIT(33), a4 BIT(63), a5 BIT(64)
+  ,b1 TINYINT, b2 TINYINT UNSIGNED
+  ,c1 SMALLINT, c2 SMALLINT UNSIGNED
+  ,d1 INT, d2 INT UNSIGNED
+  ,e1 BIGINT, e2 BIGINT UNSIGNED
+  ,f1 CHAR(1) BINARY, f2 CHAR(32) BINARY, f3 CHAR(255) BINARY
+  ,g1 VARCHAR(32) BINARY, g2 VARCHAR(255) BINARY, g3 VARCHAR(1000) BINARY
+  ,h1 BINARY(1), h2 BINARY(8), h3 BINARY(255)
+  ,i1 VARBINARY(32), i2 VARBINARY(255), i3 VARBINARY(1000)
+ ) engine ndb;
+
+insert into t1 values
+ (1
+  ,0x1, 0x17, 0x789a, 0x789abcde, 0xfedc0001
+  ,127, 255
+  ,32767, 65535
+  ,2147483647, 4294967295
+  ,9223372036854775807, 18446744073709551615
+  ,'1','12345678901234567890123456789012','123456789'
+  ,'1','12345678901234567890123456789012','123456789'
+  ,0x12,0x123456789abcdef0, 0x012345
+  ,0x12,0x123456789abcdef0, 0x00123450
+ );
+
+insert into t1 values
+ (2
+  ,0, 0, 0, 0, 0
+  ,-128, 0
+  ,-32768, 0
+  ,-2147483648, 0
+  ,-9223372036854775808, 0
+  ,'','',''
+  ,'','',''
+  ,0x0,0x0,0x0
+  ,0x0,0x0,0x0
+ );
+
+insert into t1 values
+ (3
+  ,NULL,NULL,NULL,NULL,NULL
+  ,NULL,NULL
+  ,NULL,NULL
+  ,NULL,NULL
+  ,NULL,NULL
+  ,NULL,NULL,NULL
+  ,NULL,NULL,NULL
+  ,NULL,NULL,NULL
+  ,NULL,NULL,NULL
+ );
+
+#NO.2 test output of backup after some simple SQL operations
+--exec $NDB_TOOLS_DIR/../src/mgmclient/ndb_mgm -e "start backup" |sed -e 's/[0-9]//g' |sed -e 's/localhost//g' |sed -e 's/\.\.\.*//g'

ndb/include/debugger/EventLogger.hpp

@@ -175,5 +175,5 @@ private:
   char m_text[MAX_TEXT_LENGTH];
 };
 
-
+extern void getRestartAction(Uint32 action, BaseString &str);
 #endif

ndb/src/common/transporter/TransporterRegistry.cpp

@@ -959,6 +959,7 @@ TransporterRegistry::performReceive()
         {
           Uint32 * ptr;
           Uint32 sz = t->getReceiveData(&ptr);
+          transporter_recv_from(callbackObj, nodeId);
           Uint32 szUsed = unpack(ptr, sz, nodeId, ioStates[nodeId]);
           t->updateReceiveDataPtr(szUsed);
 	}

ndb/src/kernel/blocks/ERROR_codes.txt

 Next QMGR 1
-Next NDBCNTR 1000
+Next NDBCNTR 1002
 Next NDBFS 2000
 Next DBACC 3002
 Next DBTUP 4014
@@ -501,3 +501,4 @@ TUP:
 NDBCNTR:
 
 1000: Crash insertion on SystemError::CopyFragRef
+1001: Delay sending NODE_FAILREP (to own node), until error is cleared

ndb/src/kernel/blocks/dbdict/Dbdict.cpp

@@ -404,6 +404,9 @@ void Dbdict::execFSCLOSECONF(Signal* signal)
   case FsConnectRecord::OPEN_READ_SCHEMA2:
     openSchemaFile(signal, 1, fsPtr.i, false, false);
     break;
+  case FsConnectRecord::OPEN_READ_TAB_FILE2:
+    openTableFile(signal, 1, fsPtr.i, c_readTableRecord.tableId, false);
+    break;
   default:
     jamLine((fsPtr.p->fsState & 0xFFF));
     ndbrequire(false);
@@ -783,8 +786,11 @@ void Dbdict::readTableConf(Signal* signal,
 void Dbdict::readTableRef(Signal* signal,
                           FsConnectRecordPtr fsPtr)
 {
+  /**
+   * First close corrupt file
+   */
   fsPtr.p->fsState = FsConnectRecord::OPEN_READ_TAB_FILE2;
-  openTableFile(signal, 1, fsPtr.i, c_readTableRecord.tableId, false);
+  closeFile(signal, fsPtr.p->filePtr, fsPtr.i);
   return;
 }//Dbdict::readTableRef()
 

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

@@ -4459,12 +4459,18 @@ 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();

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);

ndb/src/kernel/blocks/ndbfs/Ndbfs.cpp

@@ -569,7 +569,7 @@ AsyncFile*
 Ndbfs::createAsyncFile(){
 
   // Check limit of open files
-  if (theFiles.size()+1 ==  m_maxFiles) {
+  if (theFiles.size() ==  m_maxFiles) {
     // Print info about all open files
     for (unsigned i = 0; i < theFiles.size(); i++){
       AsyncFile* file = theFiles[i];

ndb/src/kernel/vm/WatchDog.cpp

@@ -16,6 +16,7 @@
 
 #include <ndb_global.h>
 #include <my_pthread.h>
+#include <sys/times.h>
 
 #include "WatchDog.hpp"
 #include "GlobalData.hpp"
@@ -129,6 +130,13 @@ WatchDog::run(){
         break;
       }//switch
       g_eventLogger.warning("Ndb kernel is stuck in: %s", last_stuck_action);
+      {
+        struct tms my_tms;
+        times(&my_tms);
+        g_eventLogger.info("User time: %llu  System time: %llu",
+                           (Uint64)my_tms.tms_utime,
+                           (Uint64)my_tms.tms_stime);
+      }
       if(alerts == 3){
 	shutdownSystem(last_stuck_action);
       }

ndb/src/mgmclient/CommandInterpreter.cpp

@@ -25,6 +25,7 @@
 
 #include <mgmapi.h>
 #include <util/BaseString.hpp>
+#include <ndbd_exit_codes.h>
 
 class MgmtSrvr;
 
@@ -125,7 +126,7 @@ public:
   int  executeStatus(int processId, const char* parameters, bool all);
   int  executeEventReporting(int processId, const char* parameters, bool all);
   int  executeDumpState(int processId, const char* parameters, bool all);
-  int  executeStartBackup(char * parameters);
+  int  executeStartBackup(char * parameters, bool interactive);
   int  executeAbortBackup(char * parameters);
   int  executeStop(Vector<BaseString> &command_list, unsigned command_pos,
                    int *node_ids, int no_of_nodes);
@@ -768,6 +769,113 @@ CommandInterpreter::printError()
   }
 }
 
+/*
+ * print log event from mgmsrv to console screen
+ */
+static void
+printLogEvent(struct ndb_logevent* event)
+{
+  switch (event->type) {
+    /** 
+     * NDB_MGM_EVENT_CATEGORY_BACKUP
+     */
+    case NDB_LE_BackupStarted:
+      ndbout_c("Backup %d started from node %d", 
+               event->BackupStarted.backup_id, event->BackupStarted.starting_node);
+      break;
+    case NDB_LE_BackupFailedToStart:
+      ndbout_c("Backup request from %d failed to start. Error: %d", 
+               event->BackupFailedToStart.starting_node, event->BackupFailedToStart.error);
+      break;
+    case NDB_LE_BackupCompleted:
+      ndbout_c("Backup %u started from node %u completed\n" 
+               " StartGCP: %u StopGCP: %u\n" 
+               " #Records: %u #LogRecords: %u\n" 
+               " Data: %u bytes Log: %u bytes",
+               event->BackupCompleted.backup_id, event->BackupCompleted.starting_node,
+               event->BackupCompleted.start_gci, event->BackupCompleted.stop_gci,
+               event->BackupCompleted.n_records, event->BackupCompleted.n_log_records,
+               event->BackupCompleted.n_bytes, event->BackupCompleted.n_log_bytes);
+      break;
+    case NDB_LE_BackupAborted:
+      ndbout_c("Backup %d started from %d has been aborted. Error: %d",
+               event->BackupAborted.backup_id, event->BackupAborted.starting_node, 
+               event->BackupAborted.error);
+      break;
+    /** 
+     * NDB_MGM_EVENT_CATEGORY_STARTUP
+     */ 
+    case NDB_LE_NDBStartStarted:
+      ndbout_c("Start initiated (version %d.%d.%d)", 
+               getMajor(event->NDBStartStarted.version),
+               getMinor(event->NDBStartStarted.version),
+               getBuild(event->NDBStartStarted.version));
+      break;
+    case NDB_LE_NDBStartCompleted:
+      ndbout_c("Started (version %d.%d.%d)", 
+               getMajor(event->NDBStartCompleted.version),
+               getMinor(event->NDBStartCompleted.version),
+               getBuild(event->NDBStartCompleted.version));
+      break;
+    case NDB_LE_NDBStopStarted:
+      ndbout_c("%s shutdown initiated", 
+               (event->NDBStopStarted.stoptype == 1 ? "Cluster" : "Node"));
+      break;
+    case NDB_LE_NDBStopCompleted:
+      {
+        BaseString action_str("");
+        BaseString signum_str("");
+        getRestartAction(event->NDBStopCompleted.action, action_str);
+        if (event->NDBStopCompleted.signum)
+          signum_str.appfmt(" Initiated by signal %d.", 
+                            event->NDBStopCompleted.signum);
+        ndbout_c("Node shutdown completed%s.%s", 
+                 action_str.c_str(), 
+                 signum_str.c_str());
+      }
+      break;
+    case NDB_LE_NDBStopForced:
+      {
+        BaseString action_str("");
+        BaseString reason_str("");
+        BaseString sphase_str("");
+        int signum = event->NDBStopForced.signum;
+        int error = event->NDBStopForced.error; 
+        int sphase = event->NDBStopForced.sphase; 
+        int extra = event->NDBStopForced.extra; 
+        getRestartAction(event->NDBStopForced.action, action_str);
+        if (signum)
+          reason_str.appfmt(" Initiated by signal %d.", signum);
+        if (error)
+        {
+          ndbd_exit_classification cl;
+          ndbd_exit_status st;
+          const char *msg = ndbd_exit_message(error, &cl);
+          const char *cl_msg = ndbd_exit_classification_message(cl, &st);
+          const char *st_msg = ndbd_exit_status_message(st);
+          reason_str.appfmt(" Caused by error %d: \'%s(%s). %s\'.", 
+                            error, msg, cl_msg, st_msg);
+          if (extra != 0)
+            reason_str.appfmt(" (extra info %d)", extra);
+        }
+        if (sphase < 255)
+          sphase_str.appfmt(" Occured during startphase %u.", sphase);
+        ndbout_c("Forced node shutdown completed%s.%s%s", 
+                 action_str.c_str(), sphase_str.c_str(), 
+                 reason_str.c_str());
+      }
+      break;
+    case NDB_LE_NDBStopAborted:
+      ndbout_c("Node shutdown aborted");
+      break;
+    /** 
+     * default nothing to print
+     */ 
+    default:
+      break;
+  }
+}
+
 //*****************************************************************************
 //*****************************************************************************
 
@@ -784,27 +892,21 @@ event_thread_run(void* p)
   int filter[] = { 15, NDB_MGM_EVENT_CATEGORY_BACKUP,
 		   1, NDB_MGM_EVENT_CATEGORY_STARTUP,
 		   0 };
-  int fd = ndb_mgm_listen_event(handle, filter);
-  if (fd != NDB_INVALID_SOCKET)
+
+  NdbLogEventHandle log_handle= NULL;
+  struct ndb_logevent log_event;
+
+  log_handle= ndb_mgm_create_logevent_handle(handle, filter);
+  if (log_handle) 
   {
     do_event_thread= 1;
-    char *tmp= 0;
-    char buf[1024];
-    SocketInputStream in(fd,10);
     do {
-      if (tmp == 0) NdbSleep_MilliSleep(10);
-      if((tmp = in.gets(buf, 1024)))
-      {
-	const char ping_token[]= "<PING>";
-	if (memcmp(ping_token,tmp,sizeof(ping_token)-1))
-	  if(tmp && strlen(tmp))
-          {
+      if (ndb_logevent_get_next(log_handle, &log_event, 2000) <= 0)
+        continue;
       Guard g(printmutex);
-            ndbout << tmp;
-          }
-      }
+      printLogEvent(&log_event);
     } while(do_event_thread);
-    NDB_CLOSE_SOCKET(fd);
+    ndb_mgm_destroy_logevent_handle(&log_handle);
   }
   else
   {
@@ -1054,7 +1156,7 @@ CommandInterpreter::execute_impl(const char *_line, bool interactive)
   else if(strcasecmp(firstToken, "START") == 0 &&
 	  allAfterFirstToken != NULL &&
 	  strncasecmp(allAfterFirstToken, "BACKUP", sizeof("BACKUP") - 1) == 0){
-    m_error= executeStartBackup(allAfterFirstToken);
+    m_error= executeStartBackup(allAfterFirstToken, interactive);
     DBUG_RETURN(true);
   }
   else if(strcasecmp(firstToken, "ABORT") == 0 &&
@@ -2531,20 +2633,11 @@ CommandInterpreter::executeEventReporting(int processId,
  * Backup
  *****************************************************************************/
 int
-CommandInterpreter::executeStartBackup(char* parameters)
+CommandInterpreter::executeStartBackup(char* parameters, bool interactive)
 {
   struct ndb_mgm_reply reply;
   unsigned int backupId;
-#if 0
-  int filter[] = { 15, NDB_MGM_EVENT_CATEGORY_BACKUP, 0 };
-  int fd = ndb_mgm_listen_event(m_mgmsrv, filter);
-  if (fd < 0)
-  {
-    ndbout << "Initializing start of backup failed" << endl;
-    printError();
-    return fd;
-  }
-#endif
+
   Vector<BaseString> args;
   {
     BaseString(parameters).split(args);
@@ -2557,25 +2650,20 @@ CommandInterpreter::executeStartBackup(char* parameters)
   int sz= args.size();
 
   int result;
-  if (sz == 2 &&
-      args[1] == "NOWAIT")
+  int flags = 2;
+  if (sz == 2 && args[1] == "NOWAIT")
   {
-    result = ndb_mgm_start_backup(m_mgmsrv, 0, &backupId, &reply);
+    flags = 0;
   }
-  else if (sz == 1 ||
-	   (sz == 3 &&
-	    args[1] == "WAIT" &&
-	    args[2] == "COMPLETED"))
+  else if (sz == 1 || (sz == 3 && args[1] == "WAIT" && args[2] == "COMPLETED"))
   {
+    flags = 2;
     ndbout_c("Waiting for completed, this may take several minutes");
-    result = ndb_mgm_start_backup(m_mgmsrv, 2, &backupId, &reply);
   }
-  else if (sz == 3 &&
-	   args[1] == "WAIT" &&
-	   args[2] == "STARTED")
+  else if (sz == 3 && args[1] == "WAIT" && args[2] == "STARTED")
   {
     ndbout_c("Waiting for started, this may take several minutes");
-    result = ndb_mgm_start_backup(m_mgmsrv, 1, &backupId, &reply);
+    flags = 1;
   }
   else
   {
@@ -2583,48 +2671,80 @@ CommandInterpreter::executeStartBackup(char* parameters)
     return -1;
   }
 
+  NdbLogEventHandle log_handle= NULL;
+  struct ndb_logevent log_event;
+  if (flags == 2 && !interactive)
+  {
+    int filter[] = { 15, NDB_MGM_EVENT_CATEGORY_BACKUP, 0, 0 };
+    log_handle = ndb_mgm_create_logevent_handle(m_mgmsrv, filter);
+    if (!log_handle)
+    {
+      ndbout << "Initializing start of backup failed" << endl;
+      printError();
+      return -1;
+    }
+  }
+  result = ndb_mgm_start_backup(m_mgmsrv, flags, &backupId, &reply);
+
   if (result != 0) {
     ndbout << "Backup failed" << endl;
     printError();
-#if 0
-    close(fd);
-#endif
+
+    if (log_handle) 
+      ndb_mgm_destroy_logevent_handle(&log_handle);
     return result;
   }
-#if 0
-  ndbout_c("Waiting for completed, this may take several minutes");
-  char *tmp;
-  char buf[1024];
+
+  /**
+   * If interactive, event listner thread is already running
+   */
+  if (log_handle && !interactive)
   {
-    SocketInputStream in(fd);
     int count = 0;
+    int retry = 0;
     do {
-      tmp = in.gets(buf, 1024);
-      if(tmp)
+      if (ndb_logevent_get_next(log_handle, &log_event, 60000) > 0)
+      {
+        int print = 0;
+        switch (log_event.type) {
+          case NDB_LE_BackupStarted:
+            if (log_event.BackupStarted.backup_id == backupId)
+              print = 1;
+            break;
+          case NDB_LE_BackupCompleted:
+            if (log_event.BackupCompleted.backup_id == backupId)
+              print = 1;
+            break;
+          case NDB_LE_BackupAborted:
+            if (log_event.BackupAborted.backup_id == backupId)
+              print = 1;
+            break;
+          default:
+            break;
+        }
+        if (print)
         {
-	ndbout << tmp;
-	unsigned int id;
-	if(sscanf(tmp, "%*[^:]: Backup %d ", &id) == 1 && id == backupId){
+          Guard g(m_print_mutex);
+          printLogEvent(&log_event);
           count++;
         }
       }
-    } while(count < 2);
+      else
+      {
+        retry++;
       }
+    } while(count < 2 && retry < 3);
 
-  SocketInputStream in(fd, 10);
-  do {
-    tmp = in.gets(buf, 1024);
-    if(tmp && tmp[0] != 0)
-    {
-      ndbout << tmp;
+    if (retry >= 3)
+      ndbout << "get backup event failed for " << retry << " times" << endl;
+
+    ndb_mgm_destroy_logevent_handle(&log_handle);
   }
-  } while(tmp && tmp[0] != 0);
 
-  close(fd);
-#endif  
   return 0;
 }
 
+
 int
 CommandInterpreter::executeAbortBackup(char* parameters) 
 {

ndb/src/mgmclient/Makefile.am

@@ -21,7 +21,8 @@ libndbmgmclient_la_LIBADD = ../mgmapi/libmgmapi.la \
 			    ../common/logger/liblogger.la \
 			    ../common/portlib/libportlib.la \
 			    ../common/util/libgeneral.la \
-			    ../common/portlib/libportlib.la
+			    ../common/portlib/libportlib.la \
+			    ../common/debugger/libtrace.la
 
 
 ndb_mgm_SOURCES = main.cpp

ndb/test/include/NdbRestarter.hpp

@@ -67,6 +67,8 @@ public:
   int getRandomNodeOtherNodeGroup(int nodeId, int randomNumber);
   int getRandomNotMasterNodeId(int randomNumber);
   
+  NdbMgmHandle handle;
+  
 protected:
 
   int waitClusterState(ndb_mgm_node_status _status,
@@ -89,7 +91,6 @@ protected:
   
   bool connected;
   BaseString addr;
-  NdbMgmHandle handle;
   ndb_mgm_configuration * m_config;
 protected:
   ndb_mgm_configuration * getConfig();

ndb/test/ndbapi/testNdbApi.cpp

@@ -1422,6 +1422,10 @@ TESTCASE("Scan_4006",
   INITIALIZER(runScan_4006);
   FINALIZER(runClearTable);
 }
+TESTCASE("Bug28443", 
+	 ""){ 
+  INITIALIZER(runBug28443);
+}
 TESTCASE("ExecuteAsynch", 
 	 "Check that executeAsync() works (BUG#27495)\n"){ 
   INITIALIZER(runTestExecuteAsynch);

ndb/test/ndbapi/testNodeRestart.cpp

@@ -1219,6 +1219,84 @@ runBug27283(NDBT_Context* ctx, NDBT_Step* step)
   return NDBT_OK;
 }
 
+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 "\
@@ -1552,6 +1630,9 @@ TESTCASE("Bug27003", ""){
 TESTCASE("Bug27283", ""){
   INITIALIZER(runBug27283);
 }
+TESTCASE("Bug28717", ""){
+  INITIALIZER(runBug28717);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

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

@@ -508,6 +508,10 @@ 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