Bug #28899 not possible to set separate watchdog timeout at startup

storage/ndb/include/mgmapi/mgmapi_config_parameters.h

@@ -81,6 +81,8 @@
 #define CFG_DB_BACKUP_WRITE_SIZE          136
 #define CFG_DB_BACKUP_MAX_WRITE_SIZE      139
 
+#define CFG_DB_WATCHDOG_INTERVAL_INITIAL  141
+
 #define CFG_LOG_DESTINATION           147
 
 #define CFG_DB_DISCLESS               148

storage/ndb/include/portlib/NdbTick.h

@@ -37,9 +37,6 @@ NDB_TICKS NdbTick_CurrentMillisecond(void);
  */
 int NdbTick_CurrentMicrosecond(NDB_TICKS * secs, Uint32 * micros);
 
-  /*#define TIME_MEASUREMENT*/
-#ifdef TIME_MEASUREMENT
-
 struct MicroSecondTimer {
   NDB_TICKS seconds;
   NDB_TICKS micro_seconds;
@@ -54,7 +51,6 @@ struct MicroSecondTimer {
 NDB_TICKS NdbTick_getMicrosPassed(struct MicroSecondTimer start,
                             struct MicroSecondTimer stop);
 int NdbTick_getMicroTimer(struct MicroSecondTimer* time_now);
-#endif
 
 #ifdef	__cplusplus
 }

storage/ndb/src/common/portlib/NdbTick.c

@@ -15,7 +15,7 @@
 
 
 #include <ndb_global.h>
-#include "NdbTick.h"
+#include <NdbTick.h>
 
 #define NANOSEC_PER_SEC  1000000000
 #define MICROSEC_PER_SEC 1000000
@@ -71,7 +71,6 @@ NdbTick_CurrentMicrosecond(NDB_TICKS * secs, Uint32 * micros){
 }
 
 #endif
-#ifdef TIME_MEASUREMENT
 int
 NdbTick_getMicroTimer(struct MicroSecondTimer* input_timer)
 {
@@ -102,4 +101,3 @@ NdbTick_getMicrosPassed(struct MicroSecondTimer start,
   }
   return ret_value;
 }
-#endif

storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp

@@ -277,6 +277,14 @@ void Ndbcntr::execSTTOR(Signal* signal)
     break;
   case ZSTART_PHASE_1:
     jam();
+    {
+      Uint32 db_watchdog_interval = 0;
+      const ndb_mgm_configuration_iterator * p = 
+        m_ctx.m_config.getOwnConfigIterator();
+      ndb_mgm_get_int_parameter(p, CFG_DB_WATCHDOG_INTERVAL, &db_watchdog_interval);
+      ndbrequire(db_watchdog_interval);
+      update_watch_dog_timer(db_watchdog_interval);
+    }
     startPhase1Lab(signal);
     break;
   case ZSTART_PHASE_2:

storage/ndb/src/kernel/vm/Configuration.cpp

@@ -443,6 +443,11 @@ Configuration::setupConfiguration(){
 	      "TimeBetweenWatchDogCheck missing");
   }
 
+  if(iter.get(CFG_DB_WATCHDOG_INTERVAL_INITIAL, &_timeBetweenWatchDogCheckInitial)){
+    ERROR_SET(fatal, NDBD_EXIT_INVALID_CONFIG, "Invalid configuration fetched", 
+	      "TimeBetweenWatchDogCheckInitial missing");
+  }
+
   /**
    * Get paths
    */  
@@ -462,9 +467,12 @@ Configuration::setupConfiguration(){
    * Create the watch dog thread
    */
   { 
-    Uint32 t = _timeBetweenWatchDogCheck;
+    if (_timeBetweenWatchDogCheckInitial < _timeBetweenWatchDogCheck)
+      _timeBetweenWatchDogCheckInitial = _timeBetweenWatchDogCheck;
+
+    Uint32 t = _timeBetweenWatchDogCheckInitial;
     t = globalEmulatorData.theWatchDog ->setCheckInterval(t);
-    _timeBetweenWatchDogCheck = t;
+    _timeBetweenWatchDogCheckInitial = t;
   }
   
   ConfigValues* cf = ConfigValuesFactory::extractCurrentSection(iter.m_config);

storage/ndb/src/kernel/vm/Configuration.hpp

@@ -84,6 +84,7 @@ private:
   Uint32 _maxErrorLogs;
   Uint32 _lockPagesInMainMemory;
   Uint32 _timeBetweenWatchDogCheck;
+  Uint32 _timeBetweenWatchDogCheckInitial;
 
   ndb_mgm_configuration * m_ownConfig;
   ndb_mgm_configuration * m_clusterConfig;

storage/ndb/src/kernel/vm/SimulatedBlock.cpp

@@ -19,6 +19,7 @@
 #include <NdbOut.hpp>
 #include <GlobalData.hpp>
 #include <Emulator.hpp>
+#include <WatchDog.hpp>
 #include <ErrorHandlingMacros.hpp>
 #include <TimeQueue.hpp>
 #include <TransporterRegistry.hpp>
@@ -662,7 +663,7 @@ SimulatedBlock::allocRecord(const char * type, size_t s, size_t n, bool clear, U
   void * p = NULL;
   size_t size = n*s;
   Uint64 real_size = (Uint64)((Uint64)n)*((Uint64)s);
-  refresh_watch_dog(); 
+  refresh_watch_dog(9);
   if (real_size > 0){
 #ifdef VM_TRACE_MEM
     ndbout_c("%s::allocRecord(%s, %u, %u) = %llu bytes", 
@@ -696,12 +697,12 @@ SimulatedBlock::allocRecord(const char * type, size_t s, size_t n, bool clear, U
       char * ptr = (char*)p;
       const Uint32 chunk = 128 * 1024;
       while(size > chunk){
-	refresh_watch_dog(); 
+	refresh_watch_dog(9);
 	memset(ptr, 0, chunk);
 	ptr += chunk;
 	size -= chunk;
       }
-      refresh_watch_dog(); 
+      refresh_watch_dog(9);
       memset(ptr, 0, size);
     }
   }
@@ -720,9 +721,16 @@ SimulatedBlock::deallocRecord(void ** ptr,
 }
 
 void
-SimulatedBlock::refresh_watch_dog()
+SimulatedBlock::refresh_watch_dog(Uint32 place)
 {
-  globalData.incrementWatchDogCounter(1);
+  globalData.incrementWatchDogCounter(place);
+}
+
+void
+SimulatedBlock::update_watch_dog_timer(Uint32 interval)
+{
+  extern EmulatorData globalEmulatorData;
+  globalEmulatorData.theWatchDog->setCheckInterval(interval);
 }
 
 void

storage/ndb/src/kernel/vm/SimulatedBlock.hpp

@@ -334,7 +334,8 @@ protected:
    * Refresh Watch Dog in initialising code
    *
    */
-  void refresh_watch_dog();
+  void refresh_watch_dog(Uint32 place = 1);
+  void update_watch_dog_timer(Uint32 interval);
 
   /**
    * Prog error

storage/ndb/src/kernel/vm/WatchDog.cpp

@@ -25,6 +25,8 @@
 #include <ErrorHandlingMacros.hpp>
 #include <EventLogger.hpp>
 
+#include <NdbTick.h>
+
 extern EventLogger g_eventLogger;
 
 extern "C" 
@@ -72,73 +74,115 @@ WatchDog::doStop(){
   }
 }
 
+const char *get_action(Uint32 IPValue)
+{
+  const char *action;
+  switch (IPValue) {
+  case 1:
+    action = "Job Handling";
+    break;
+  case 2:
+    action = "Scanning Timers";
+    break;
+  case 3:
+    action = "External I/O";
+    break;
+  case 4:
+    action = "Print Job Buffers at crash";
+    break;
+  case 5:
+    action = "Checking connections";
+    break;
+  case 6:
+    action = "Performing Send";
+    break;
+  case 7:
+    action = "Polling for Receive";
+    break;
+  case 8:
+    action = "Performing Receive";
+    break;
+  case 9:
+    action = "Allocating memory";
+    break;
+  default:
+    action = "Unknown place";
+    break;
+  }//switch
+  return action;
+}
+
 void 
-WatchDog::run(){
-  unsigned int anIPValue;
-  unsigned int alerts = 0;
+WatchDog::run()
+{
+  unsigned int anIPValue, sleep_time;
   unsigned int oldIPValue = 0;
-  
+  unsigned int theIntervalCheck = theInterval;
+  struct MicroSecondTimer start_time, last_time, now;
+  NdbTick_getMicroTimer(&start_time);
+  last_time = start_time;
+
   // WatchDog for the single threaded NDB
-  while(!theStop){
-    Uint32 tmp  = theInterval / 500;
-    tmp= (tmp ? tmp : 1);
-    
-    while(!theStop && tmp > 0){
-      NdbSleep_MilliSleep(500);
-      tmp--;
-    }
-    
+  while (!theStop)
+  {
+    sleep_time= 100;
+
+    NdbSleep_MilliSleep(sleep_time);
     if(theStop)
       break;
 
+    NdbTick_getMicroTimer(&now);
+    if (NdbTick_getMicrosPassed(last_time, now)/1000 > sleep_time*2)
+    {
+      struct tms my_tms;
+      times(&my_tms);
+      g_eventLogger.info("Watchdog: User time: %llu  System time: %llu",
+                         (Uint64)my_tms.tms_utime,
+                         (Uint64)my_tms.tms_stime);
+      g_eventLogger.warning("Watchdog: Warning overslept %u ms, expected %u ms.",
+                            NdbTick_getMicrosPassed(last_time, now)/1000,
+                            sleep_time);
+    }
+    last_time = now;
+
     // Verify that the IP thread is not stuck in a loop
     anIPValue = *theIPValue;
-    if(anIPValue != 0) {
+    if (anIPValue != 0)
+    {
       oldIPValue = anIPValue;
       globalData.incrementWatchDogCounter(0);
-      alerts = 0;
-    } else {
-      const char *last_stuck_action;
-      alerts++;
-      switch (oldIPValue) {
-      case 1:
-        last_stuck_action = "Job Handling";
-        break;
-      case 2:
-        last_stuck_action = "Scanning Timers";
-        break;
-      case 3:
-        last_stuck_action = "External I/O";
-        break;
-      case 4:
-        last_stuck_action = "Print Job Buffers at crash";
-        break;
-      case 5:
-        last_stuck_action = "Checking connections";
-        break;
-      case 6:
-        last_stuck_action = "Performing Send";
-        break;
-      case 7:
-        last_stuck_action = "Polling for Receive";
-        break;
-      case 8:
-        last_stuck_action = "Performing Receive";
-        break;
-      default:
-        last_stuck_action = "Unknown place";
-        break;
-      }//switch
-      g_eventLogger.warning("Ndb kernel is stuck in: %s", last_stuck_action);
+      NdbTick_getMicroTimer(&start_time);
+      theIntervalCheck = theInterval;
+    }
+    else
+    {
+      int warn = 1;
+      Uint32 elapsed = NdbTick_getMicrosPassed(start_time, now)/1000;
+      /*
+        oldIPValue == 9 indicates malloc going on, this can take some time
+        so only warn if we pass the watchdog interval
+      */
+      if (oldIPValue == 9)
+        if (elapsed < theIntervalCheck)
+          warn = 0;
+        else
+          theIntervalCheck += theInterval;
+
+      if (warn)
       {
-        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);
+        const char *last_stuck_action = get_action(oldIPValue);
+        g_eventLogger.warning("Ndb kernel is stuck in: %s", last_stuck_action);
+        {
+          struct tms my_tms;
+          times(&my_tms);
+          g_eventLogger.info("Watchdog: User time: %llu  System time: %llu",
+                             (Uint64)my_tms.tms_utime,
+                             (Uint64)my_tms.tms_stime);
+        }
+        if (elapsed > 3 * theInterval)
+        {
+          shutdownSystem(last_stuck_action);
+        }
       }
     }
   }

storage/ndb/src/mgmsrv/ConfigInfo.cpp

@@ -571,6 +571,18 @@ const ConfigInfo::ParamInfo ConfigInfo::m_ParamInfo[] = {
     "70",
     STR_VALUE(MAX_INT_RNIL) },
 
+  {
+    CFG_DB_WATCHDOG_INTERVAL_INITIAL,
+    "TimeBetweenWatchDogCheckInitial",
+    DB_TOKEN,
+    "Time between execution checks inside a database node in the early start phases when memory is allocated",
+    ConfigInfo::CI_USED,
+    true,
+    ConfigInfo::CI_INT,
+    "6000",
+    "70",
+    STR_VALUE(MAX_INT_RNIL) },
+
   {
     CFG_DB_STOP_ON_ERROR,
     "StopOnError",