BUG#26580041: PUSH OF WL#9644 CAUSED REGRESSION SERVER

              SHUTDOWN

ANALYSIS:
=========
Shutdown time of server with empty database increased
two fold.

During shutdown, we close all the connections to the
server. During that stage, first we signal all threads
that they need to be killed. This will give them some
time to gracefully abort their statements and inform
their clients that the server is about to die. There is
a hardcoded sleep in the code which gives time for the
threads to die if they are not killed already.
Event scheduler is now enabled by default and thus the
event scheduler thread needs to be killed explicitly after
all the client connections are killed. This thread was
killed after the sleep and hence triggered the sleep
everytime there was a shutdown because this thread was
still active. Thus the shutdown time increased.

FIX:
====
The event scheduler is now stopped and the thread is
killed before the sleep after all the client connections
are signalled to die.
Do note that, when we try to stop the event scheduler,
an acknowledgement from the scheduler that it has stopped,
is necessary. Already running events will not be stopped.

Author: karthik-kamath

sql/mysqld.cc

@@ -1712,6 +1712,10 @@ static void close_connections(void)
     set_kill_conn.set_dump_thread_flag();
     thd_manager->do_for_all_thd(&set_kill_conn);
   }
+
+  // Disable the event scheduler
+  Events::stop();
+
   if (thd_manager->get_thd_count() > 0)
     sleep(2);         // Give threads time to die