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

into  selena.:H:/MYSQL/src/#13377-mysql-5.0

sql/log.cc

@@ -358,7 +358,7 @@ MYSQL_LOG::MYSQL_LOG()
   :bytes_written(0), last_time(0), query_start(0), name(0),
    file_id(1), open_count(1), log_type(LOG_CLOSED), write_error(0), inited(0),
    need_start_event(1), prepared_xids(0), description_event_for_exec(0),
-   description_event_for_queue(0)
+   description_event_for_queue(0), readers_count(0), reset_pending(false)
 {
   /*
     We don't want to initialize LOCK_Log here as such initialization depends on
@@ -384,7 +384,9 @@ void MYSQL_LOG::cleanup()
     delete description_event_for_exec;
     (void) pthread_mutex_destroy(&LOCK_log);
     (void) pthread_mutex_destroy(&LOCK_index);
+    (void) pthread_mutex_destroy(&LOCK_readers);
     (void) pthread_cond_destroy(&update_cond);
+    (void) pthread_cond_destroy(&reset_cond);
   }
   DBUG_VOID_RETURN;
 }
@@ -429,7 +431,9 @@ void MYSQL_LOG::init_pthread_objects()
   inited= 1;
   (void) pthread_mutex_init(&LOCK_log,MY_MUTEX_INIT_SLOW);
   (void) pthread_mutex_init(&LOCK_index, MY_MUTEX_INIT_SLOW);
+  (void) pthread_mutex_init(&LOCK_readers, MY_MUTEX_INIT_SLOW);
   (void) pthread_cond_init(&update_cond, 0);
+  (void) pthread_cond_init(&reset_cond, 0);
 }
 
 const char *MYSQL_LOG::generate_name(const char *log_name,
@@ -932,6 +936,13 @@ bool MYSQL_LOG::reset_logs(THD* thd)
   */
   pthread_mutex_lock(&LOCK_log);
   pthread_mutex_lock(&LOCK_index);
+
+  /* 
+    we need one more lock to block attempts to open a log while
+    we are waiting untill all log files will be closed
+  */
+  pthread_mutex_lock(&LOCK_readers);
+
   /*
     The following mutex is needed to ensure that no threads call
     'delete thd' as we would then risk missing a 'rollback' from this
@@ -954,6 +965,19 @@ bool MYSQL_LOG::reset_logs(THD* thd)
     goto err;
   }
 
+  reset_pending = true;
+  /* 
+    send update signal just in case so that all reader threads waiting
+    for log update will leave wait condition
+  */
+  signal_update();
+  /* 
+    if there are active readers wait until all of them will 
+    release opened files 
+  */
+  if (readers_count)
+    pthread_cond_wait(&reset_cond, &LOCK_log);
+
   for (;;)
   {
     my_delete(linfo.log_file_name, MYF(MY_WME));
@@ -972,7 +996,10 @@ bool MYSQL_LOG::reset_logs(THD* thd)
   my_free((gptr) save_name, MYF(0));
 
 err:
+  reset_pending = false;
+
   (void) pthread_mutex_unlock(&LOCK_thread_count);
+  pthread_mutex_unlock(&LOCK_readers);
   pthread_mutex_unlock(&LOCK_index);
   pthread_mutex_unlock(&LOCK_log);
   DBUG_RETURN(error);
@@ -2045,6 +2072,10 @@ void MYSQL_LOG::wait_for_update(THD* thd, bool is_slave)
 {
   const char *old_msg;
   DBUG_ENTER("wait_for_update");
+  
+  if (reset_pending)
+    DBUG_VOID_RETURN;
+
   old_msg= thd->enter_cond(&update_cond, &LOCK_log,
                            is_slave ?
                            "Has read all relay log; waiting for the slave I/O "
@@ -2295,6 +2326,32 @@ void MYSQL_LOG::signal_update()
   DBUG_VOID_RETURN;
 }
 
+void MYSQL_LOG::readers_addref()
+{
+  /* 
+    There is no necessity for reference counting on *nix, since it allows to
+    delete opened files, however it is more clean way to wait
+    untill all files will be closed on *nix as well.
+  */
+  DBUG_ENTER("MYSQL_LOG::reader_addref");
+  pthread_mutex_lock(&LOCK_log);
+  pthread_mutex_lock(&LOCK_readers);
+  readers_count++;
+  pthread_mutex_unlock(&LOCK_readers);
+  pthread_mutex_unlock(&LOCK_log);
+  DBUG_VOID_RETURN;
+}
+
+void MYSQL_LOG::readers_release()
+{
+  DBUG_ENTER("MYSQL_LOG::reader_release");
+  pthread_mutex_lock(&LOCK_log);
+  readers_count--;
+  if (!readers_count)
+    pthread_cond_broadcast(&reset_cond);
+  pthread_mutex_unlock(&LOCK_log);
+  DBUG_VOID_RETURN;
+}
 
 #ifdef __NT__
 void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,

sql/slave.cc

@@ -3692,6 +3692,15 @@ err:
   pthread_mutex_lock(&mi->run_lock);
   mi->slave_running = 0;
   mi->io_thd = 0;
+
+  /* Close log file and free buffers */
+  if (mi->rli.cur_log_fd >= 0)
+  {
+    end_io_cache(&mi->rli.cache_buf);
+    my_close(mi->rli.cur_log_fd, MYF(MY_WME));
+    mi->rli.cur_log_fd= -1;
+  }
+
   /* Forget the relay log's format */
   delete mi->rli.relay_log.description_event_for_queue;
   mi->rli.relay_log.description_event_for_queue= 0;
@@ -3908,6 +3917,14 @@ the slave SQL thread with \"SLAVE START\". We stopped at log \
   rli->cached_charset_invalidate();
   rli->save_temporary_tables = thd->temporary_tables;
 
+  /* Close log file and free buffers if it's already open */
+  if (rli->cur_log_fd >= 0)
+  {
+    end_io_cache(&rli->cache_buf);
+    my_close(rli->cur_log_fd, MYF(MY_WME));
+    rli->cur_log_fd = -1;
+  }
+
   /*
     TODO: see if we can do this conditionally in next_event() instead
     to avoid unneeded position re-init

sql/sql_class.h

@@ -189,8 +189,11 @@ class MYSQL_LOG: public TC_LOG
 {
  private:
   /* LOCK_log and LOCK_index are inited by init_pthread_objects() */
-  pthread_mutex_t LOCK_log, LOCK_index;
+  pthread_mutex_t LOCK_log, LOCK_index, LOCK_readers;
   pthread_cond_t update_cond;
+  pthread_cond_t reset_cond;
+  bool reset_pending;
+  int readers_count;
   ulonglong bytes_written;
   time_t last_time,query_start;
   IO_CACHE log_file;
@@ -334,6 +337,9 @@ public:
   int purge_logs_before_date(time_t purge_time);
   int purge_first_log(struct st_relay_log_info* rli, bool included);
   bool reset_logs(THD* thd);
+  inline bool is_reset_pending() { return reset_pending; }
+  void readers_addref();
+  void readers_release();
   void close(uint exiting);
 
   // iterating through the log index file

sql/sql_repl.cc

@@ -372,6 +372,11 @@ void mysql_binlog_send(THD* thd, char* log_ident, my_off_t pos,
     goto err;
   }
 
+  /*
+    Call readers_addref before opening log to track count 
+    of binlog readers
+  */
+  mysql_bin_log.readers_addref();
   if ((file=open_binlog(&log, log_file_name, &errmsg)) < 0)
   {
     my_errno= ER_MASTER_FATAL_ERROR_READING_BINLOG;
@@ -569,7 +574,8 @@ impossible position";
       goto err;
 
     if (!(flags & BINLOG_DUMP_NON_BLOCK) &&
-        mysql_bin_log.is_active(log_file_name))
+        mysql_bin_log.is_active(log_file_name) &&
+        !mysql_bin_log.is_reset_pending())
     {
       /*
 	Block until there is more data in the log
@@ -683,6 +689,13 @@ impossible position";
     {
       bool loop_breaker = 0;
       // need this to break out of the for loop from switch
+
+      // if we are going to switch log file anyway, close current log first
+      end_io_cache(&log);
+      (void) my_close(file, MYF(MY_WME));
+      // decrease reference count of binlog readers
+      mysql_bin_log.readers_release();
+
       thd->proc_info = "Finished reading one binlog; switching to next binlog";
       switch (mysql_bin_log.find_next_log(&linfo, 1)) {
       case LOG_INFO_EOF:
@@ -691,16 +704,25 @@ impossible position";
       case 0:
 	break;
       default:
+	// need following call to do release on err label
+	mysql_bin_log.readers_addref(); 
 	errmsg = "could not find next log";
 	my_errno= ER_MASTER_FATAL_ERROR_READING_BINLOG;
 	goto err;
       }
 
-      if (loop_breaker)
-	break;
+	if (loop_breaker)
+	{
+	  // need following call to do release on end label
+	  mysql_bin_log.readers_addref();
+	  break;
+	}
 
-      end_io_cache(&log);
-      (void) my_close(file, MYF(MY_WME));
+      /*
+        Call readers_addref before opening log to track count 
+        of binlog readers
+      */
+      mysql_bin_log.readers_addref();
 
       /*
         Call fake_rotate_event() in case the previous log (the one which
@@ -733,6 +755,8 @@ end:
 
   end_io_cache(&log);
   (void)my_close(file, MYF(MY_WME));
+  // decrease reference count of binlog readers
+  mysql_bin_log.readers_release();
 
   send_eof(thd);
   thd->proc_info = "Waiting to finalize termination";
@@ -759,6 +783,9 @@ err:
   pthread_mutex_unlock(&LOCK_thread_count);
   if (file >= 0)
     (void) my_close(file, MYF(MY_WME));
+  // decrease reference count of binlog readers
+  mysql_bin_log.readers_release();
+
   my_message(my_errno, errmsg, MYF(0));
   DBUG_VOID_RETURN;
 }