MDEV-14505 - Threads_running becomes scalability bottleneck

Instead of updating global counter, calculate Threads_running on the fly.
All threads having command != COM_SLEEP are included.

Behaviour changes:

Previously SHOW STATUS and SHOW GLOBAL STATUS returned the same values
representing global status. Now SHOW STATUS always returns 1 indicating that
current session has 1 thread running.

Previously only threads that were executing dispatch_command() or running events
were accounted by Threads_running. Now it is rough equivalent of
SELECT COUNT(*) FROM INFORMATION_SCHEMA.PROCESSLIST WHERE state!='Sleep'

10 files changed, 44 insertions, 23 deletions

diff --git a/mysql-test/r/status.result b/mysql-test/r/status.result
index 9b82c7896cb..18cde57b295 100644
--- a/mysql-test/r/status.result
+++ b/mysql-test/r/status.result
@@ -409,6 +409,23 @@ Table_open_cache_overflows	5
 FLUSH TABLES;
 FLUSH STATUS;
 SET @@global.table_open_cache= @old_table_open_cache;
+#
+# MDEV-14505 - Threads_running becomes scalability bottleneck
+#
+# Session status for Threads_running is currently always 1.
+SHOW STATUS LIKE 'Threads_running';
+Variable_name	Value
+Threads_running	1
+SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.SESSION_STATUS WHERE VARIABLE_NAME='THREADS_RUNNING';
+VARIABLE_VALUE
+1
+FLUSH STATUS;
+SHOW STATUS LIKE 'Threads_running';
+Variable_name	Value
+Threads_running	1
+SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.SESSION_STATUS WHERE VARIABLE_NAME='THREADS_RUNNING';
+VARIABLE_VALUE
+1
 connection default;
 set @@global.concurrent_insert= @old_concurrent_insert;
 SET GLOBAL log_output = @old_log_output;
diff --git a/mysql-test/t/status.test b/mysql-test/t/status.test
index bff45fca583..92fba9ab0a6 100644
--- a/mysql-test/t/status.test
+++ b/mysql-test/t/status.test
@@ -420,6 +420,16 @@ while ($i)
 enable_query_log;
 SET @@global.table_open_cache= @old_table_open_cache;
 
+--echo #
+--echo # MDEV-14505 - Threads_running becomes scalability bottleneck
+--echo #
+--echo # Session status for Threads_running is currently always 1.
+SHOW STATUS LIKE 'Threads_running';
+SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.SESSION_STATUS WHERE VARIABLE_NAME='THREADS_RUNNING';
+FLUSH STATUS;
+SHOW STATUS LIKE 'Threads_running';
+SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.SESSION_STATUS WHERE VARIABLE_NAME='THREADS_RUNNING';
+
 # Restore global concurrent_insert value. Keep in the end of the test file.
 --connection default
 set @@global.concurrent_insert= @old_concurrent_insert;
diff --git a/sql/event_scheduler.cc b/sql/event_scheduler.cc
index 1572fedff49..d0a2405526a 100644
--- a/sql/event_scheduler.cc
+++ b/sql/event_scheduler.cc
@@ -297,7 +297,6 @@ Event_worker_thread::run(THD *thd, Event_queue_element_for_exec *event)
   DBUG_ENTER("Event_worker_thread::run");
   DBUG_PRINT("info", ("Time is %u, THD: %p", (uint)my_time(0), thd));
 
-  inc_thread_running();
   if (res)
     goto end;
 
@@ -326,7 +325,6 @@ end:
              event->name.str));
 
   delete event;
-  dec_thread_running();
   deinit_event_thread(thd);
 
   DBUG_VOID_RETURN;
diff --git a/sql/mysqld.cc b/sql/mysqld.cc
index f49365c9e9b..1f679ee6e9e 100644
--- a/sql/mysqld.cc
+++ b/sql/mysqld.cc
@@ -492,7 +492,6 @@ uint protocol_version;
 uint lower_case_table_names;
 ulong tc_heuristic_recover= 0;
 int32 thread_count, service_thread_count;
-int32 thread_running;
 int32 slave_open_temp_tables;
 ulong thread_created;
 ulong back_log, connect_timeout, concurrency, server_id;
@@ -8640,7 +8639,7 @@ SHOW_VAR status_vars[]= {
   {"Threads_cached",           (char*) &cached_thread_count,    SHOW_LONG_NOFLUSH},
   {"Threads_connected",        (char*) &connection_count,       SHOW_INT},
   {"Threads_created",	       (char*) &thread_created,		SHOW_LONG_NOFLUSH},
-  {"Threads_running",          (char*) &thread_running,         SHOW_INT},
+  {"Threads_running",          (char*) offsetof(STATUS_VAR, threads_running), SHOW_UINT32_STATUS},
   {"Transactions_multi_engine", (char*) &transactions_multi_engine, SHOW_LONG},
   {"Rpl_transactions_multi_engine", (char*) &rpl_transactions_multi_engine, SHOW_LONG},
   {"Transactions_gtid_foreign_engine", (char*) &transactions_gtid_foreign_engine, SHOW_LONG},
@@ -8817,7 +8816,7 @@ static int mysql_init_variables(void)
   kill_in_progress= 0;
   cleanup_done= 0;
   test_flags= select_errors= dropping_tables= ha_open_options=0;
-  thread_count= thread_running= kill_cached_threads= wake_thread= 0;
+  thread_count= kill_cached_threads= wake_thread= 0;
   service_thread_count= 0;
   slave_open_temp_tables= 0;
   cached_thread_count= 0;
diff --git a/sql/mysqld.h b/sql/mysqld.h
index 7a430d4f00b..1ca193310cc 100644
--- a/sql/mysqld.h
+++ b/sql/mysqld.h
@@ -595,7 +595,6 @@ extern mysql_rwlock_t LOCK_system_variables_hash;
 extern mysql_cond_t COND_thread_count, COND_start_thread;
 extern mysql_cond_t COND_manager;
 extern mysql_cond_t COND_slave_background;
-extern int32 thread_running;
 extern int32 thread_count, service_thread_count;
 
 extern char *opt_ssl_ca, *opt_ssl_capath, *opt_ssl_cert, *opt_ssl_cipher,
@@ -785,16 +784,6 @@ inline void thread_safe_decrement64(int64 *value)
   (void) my_atomic_add64_explicit(value, -1, MY_MEMORY_ORDER_RELAXED);
 }
 
-inline void inc_thread_running()
-{
-  thread_safe_increment32(&thread_running);
-}
-
-inline void dec_thread_running()
-{
-  thread_safe_decrement32(&thread_running);
-}
-
 extern void set_server_version(char *buf, size_t size);
 
 #define current_thd _current_thd()
diff --git a/sql/sql_class.cc b/sql/sql_class.cc
index 9d30ad65be4..30b53807578 100644
--- a/sql/sql_class.cc
+++ b/sql/sql_class.cc
@@ -4207,6 +4207,12 @@ void THD::set_status_var_init()
 {
   bzero((char*) &status_var, offsetof(STATUS_VAR,
                                       last_cleared_system_status_var));
+  /*
+    Session status for Threads_running is always 1. It can only be queried
+    by thread itself via INFORMATION_SCHEMA.SESSION_STATUS or SHOW [SESSION]
+    STATUS. And at this point thread is guaranteed to be running.
+  */
+  status_var.threads_running= 1;
 }
 
 
diff --git a/sql/sql_class.h b/sql/sql_class.h
index 20318d55da1..46dea27dc7a 100644
--- a/sql/sql_class.h
+++ b/sql/sql_class.h
@@ -828,6 +828,7 @@ typedef struct system_status_var
   ulonglong table_open_cache_overflows;
   double last_query_cost;
   double cpu_time, busy_time;
+  uint32_t threads_running;
   /* Don't initialize */
   /* Memory used for thread local storage */
   int64 max_local_memory_used;
diff --git a/sql/sql_parse.cc b/sql/sql_parse.cc
index 7a4c7ca4c56..9ce6c6b8787 100644
--- a/sql/sql_parse.cc
+++ b/sql/sql_parse.cc
@@ -1550,9 +1550,6 @@ bool dispatch_command(enum enum_server_command command, THD *thd,
                        "<?>")));
   bool drop_more_results= 0;
 
-  if (!is_com_multi)
-    inc_thread_running();
-
   /* keep it withing 1 byte */
   compile_time_assert(COM_END == 255);
 
@@ -2410,10 +2407,8 @@ com_multi_end:
   thd->m_digest= NULL;
 
   if (!is_com_multi)
-  {
-    dec_thread_running();
     thd->packet.shrink(thd->variables.net_buffer_length); // Reclaim some memory
-  }
+
   thd->reset_kill_query();  /* Ensure that killed_errmsg is released */
   free_root(thd->mem_root,MYF(MY_KEEP_PREALLOC));
 
diff --git a/sql/sql_plugin.h b/sql/sql_plugin.h
index f100e8a232a..5903029cd68 100644
--- a/sql/sql_plugin.h
+++ b/sql/sql_plugin.h
@@ -24,7 +24,8 @@
 #define SHOW_always_last SHOW_KEY_CACHE_LONG, \
             SHOW_LONG_STATUS, SHOW_DOUBLE_STATUS, \
             SHOW_HAVE, SHOW_MY_BOOL, SHOW_HA_ROWS, SHOW_SYS, \
-            SHOW_LONG_NOFLUSH, SHOW_LONGLONG_STATUS, SHOW_LEX_STRING
+            SHOW_LONG_NOFLUSH, SHOW_LONGLONG_STATUS, SHOW_UINT32_STATUS, \
+            SHOW_LEX_STRING
 #include "mariadb.h"
 #undef SHOW_always_last
 
diff --git a/sql/sql_show.cc b/sql/sql_show.cc
index befffb55e2e..c374425f74a 100644
--- a/sql/sql_show.cc
+++ b/sql/sql_show.cc
@@ -3491,6 +3491,9 @@ const char* get_one_variable(THD *thd,
   case SHOW_MY_BOOL:
     end= strmov(buff, *(my_bool*) value ? "ON" : "OFF");
     break;
+  case SHOW_UINT32_STATUS:
+    value= ((char *) status_var + (intptr) value);
+    /* fall through */
   case SHOW_UINT:
     end= int10_to_str((long) *(uint*) value, buff, 10);
     break;
@@ -3717,6 +3720,8 @@ uint calc_sum_of_all_status(STATUS_VAR *to)
       add_to_status(to, &tmp->status_var);
       to->local_memory_used+= tmp->status_var.local_memory_used;
     }
+    if (tmp->get_command() != COM_SLEEP)
+      to->threads_running++;
   }
   
   mysql_mutex_unlock(&LOCK_thread_count);