MDEV-23475 InnoDB performance regression for write-heavy workloads10.3-enterprise-MENT-581

In commit fe39d02f51b96536dccca7ff89faf05e13548877 (MDEV-20638)
we removed some wake-up signaling of the master thread that should
have been there, to ensure a steady log checkpointing workload.

Common sense suggests that the commit omitted some necessary calls
to srv_inc_activity_count(). But, an attempt to add the call to
trx_flush_log_if_needed_low() as well as to reinstate the function
innobase_active_small() did not restore the performance for the
case where sync_binlog=1 is set.

Therefore, we will revert the entire commit in MariaDB Server 10.2.
In MariaDB Server 10.5, adding a srv_inc_activity_count() call to
trx_flush_log_if_needed_low() did restore the performance, so we
will not revert MDEV-20638 across all versions.

10 files changed, 169 insertions, 34 deletions

diff --git a/extra/mariabackup/xtrabackup.cc b/extra/mariabackup/xtrabackup.cc
index f15dc425601..ed957081561 100644
--- a/extra/mariabackup/xtrabackup.cc
+++ b/extra/mariabackup/xtrabackup.cc
@@ -259,6 +259,12 @@ my_bool innobase_locks_unsafe_for_binlog;
 my_bool innobase_rollback_on_timeout;
 my_bool innobase_create_status_file;
 
+/* The following counter is used to convey information to InnoDB
+about server activity: in selects it is not sensible to call
+srv_active_wake_master_thread after each fetch or search, we only do
+it every INNOBASE_WAKE_INTERVAL'th step. */
+
+#define INNOBASE_WAKE_INTERVAL	32
 ulong	innobase_active_counter	= 0;
 
 #ifndef _WIN32
diff --git a/storage/innobase/buf/buf0flu.cc b/storage/innobase/buf/buf0flu.cc
index e1acc7b9095..6f56a09eda2 100644
--- a/storage/innobase/buf/buf0flu.cc
+++ b/storage/innobase/buf/buf0flu.cc
@@ -3116,7 +3116,7 @@ DECLARE_THREAD(buf_flush_page_cleaner_coordinator)(void*)
 		/* The page_cleaner skips sleep if the server is
 		idle and there are no pending IOs in the buffer pool
 		and there is work to do. */
-		if (srv_check_activity(&last_activity)
+		if (srv_check_activity(last_activity)
 		    || buf_get_n_pending_read_ios()
 		    || n_flushed == 0) {
 
@@ -3208,7 +3208,7 @@ DECLARE_THREAD(buf_flush_page_cleaner_coordinator)(void*)
 
 			n_flushed = n_flushed_lru + n_flushed_list;
 
-		} else if (srv_check_activity(&last_activity)) {
+		} else if (srv_check_activity(last_activity)) {
 			ulint	n_to_flush;
 			lsn_t	lsn_limit = 0;
 
diff --git a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc
index b6f39a573ab..64d3fef0fbe 100644
--- a/storage/innobase/handler/ha_innodb.cc
+++ b/storage/innobase/handler/ha_innodb.cc
@@ -449,6 +449,14 @@ TYPELIB innodb_flush_method_typelib = {
 	NULL
 };
 
+/* The following counter is used to convey information to InnoDB
+about server activity: in case of normal DML ops it is not
+sensible to call srv_active_wake_master_thread after each
+operation, we only do it every INNOBASE_WAKE_INTERVAL'th step. */
+
+#define INNOBASE_WAKE_INTERVAL	32
+static ulong	innobase_active_counter	= 0;
+
 /** Allowed values of innodb_change_buffering */
 static const char* innodb_change_buffering_names[] = {
 	"none",		/* IBUF_USE_NONE */
@@ -1846,6 +1854,23 @@ thd_to_trx_id(
 #endif /* WITH_WSREP */
 
 /********************************************************************//**
+Increments innobase_active_counter and every INNOBASE_WAKE_INTERVALth
+time calls srv_active_wake_master_thread. This function should be used
+when a single database operation may introduce a small need for
+server utility activity, like checkpointing. */
+inline
+void
+innobase_active_small(void)
+/*=======================*/
+{
+	innobase_active_counter++;
+
+	if ((innobase_active_counter % INNOBASE_WAKE_INTERVAL) == 0) {
+		srv_active_wake_master_thread();
+	}
+}
+
+/********************************************************************//**
 Converts an InnoDB error code to a MySQL error code and also tells to MySQL
 about a possible transaction rollback inside InnoDB caused by a lock wait
 timeout or a deadlock.
@@ -6460,6 +6485,11 @@ ha_innobase::close()
 
 	MONITOR_INC(MONITOR_TABLE_CLOSE);
 
+	/* Tell InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
 	DBUG_RETURN(0);
 }
 
@@ -8170,6 +8200,8 @@ report_error:
 	}
 
 func_exit:
+	innobase_active_small();
+
 	DBUG_RETURN(error_result);
 }
 
@@ -8856,6 +8888,11 @@ func_exit:
 			error, m_prebuilt->table->flags, m_user_thd);
 	}
 
+	/* Tell InnoDB server that there might be work for
+	utility threads: */
+
+	innobase_active_small();
+
 #ifdef WITH_WSREP
 	if (error == DB_SUCCESS && trx->is_wsrep() &&
 	    wsrep_thd_exec_mode(m_user_thd) == LOCAL_STATE &&
@@ -8914,6 +8951,11 @@ ha_innobase::delete_row(
 
 	innobase_srv_conc_exit_innodb(m_prebuilt);
 
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	innobase_active_small();
+
 #ifdef WITH_WSREP
 	if (error == DB_SUCCESS && trx->is_wsrep()
 	    && wsrep_thd_exec_mode(m_user_thd) == LOCAL_STATE
@@ -12701,6 +12743,7 @@ create_table_info_t::create_table_update_dict()
 	if (m_flags2 & DICT_TF2_FTS) {
 		if (!innobase_fts_load_stopword(innobase_table, NULL, m_thd)) {
 			dict_table_close(innobase_table, FALSE, FALSE);
+			srv_active_wake_master_thread();
 			trx_free(m_trx);
 			DBUG_RETURN(-1);
 		}
@@ -12850,6 +12893,11 @@ ha_innobase::create(
 
 	error = info.create_table_update_dict();
 
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
 	DBUG_RETURN(error);
 }
 
diff --git a/storage/innobase/handler/handler0alter.cc b/storage/innobase/handler/handler0alter.cc
index c5693afe50b..ac36cff6173 100644
--- a/storage/innobase/handler/handler0alter.cc
+++ b/storage/innobase/handler/handler0alter.cc
@@ -9791,6 +9791,11 @@ foreign_fail:
 
 	log_append_on_checkpoint(NULL);
 
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
 	if (fail) {
 		for (inplace_alter_handler_ctx** pctx = ctx_array;
 		     *pctx; pctx++) {
diff --git a/storage/innobase/include/srv0srv.h b/storage/innobase/include/srv0srv.h
index 93755ad7520..6c575733710 100644
--- a/storage/innobase/include/srv0srv.h
+++ b/storage/innobase/include/srv0srv.h
@@ -788,6 +788,19 @@ srv_reset_io_thread_op_info();
 /** Wake up the purge threads if there is work to do. */
 void
 srv_wake_purge_thread_if_not_active();
+/** Wake up the InnoDB master thread if it was suspended (not sleeping). */
+void
+srv_active_wake_master_thread_low();
+
+#define srv_active_wake_master_thread()					\
+	do {								\
+		if (!srv_read_only_mode) {				\
+			srv_active_wake_master_thread_low();		\
+		}							\
+	} while (0)
+/** Wake up the master thread if it is suspended or being suspended. */
+void
+srv_wake_master_thread();
 
 /******************************************************************//**
 Outputs to a file the output of the InnoDB Monitor.
@@ -816,13 +829,13 @@ reading this value as it is only used in heuristics.
 ulint
 srv_get_activity_count(void);
 /*========================*/
-
-/** Check if there has been any activity.
-@param[in,out]  activity_count  recent activity count to be returned
-if there is a change
+/*******************************************************************//**
+Check if there has been any activity.
 @return FALSE if no change in activity counter. */
-bool srv_check_activity(ulint *activity_count);
-
+ibool
+srv_check_activity(
+/*===============*/
+	ulint		old_activity_count);	/*!< old activity count */
 /******************************************************************//**
 Increment the server activity counter. */
 void
diff --git a/storage/innobase/row/row0mysql.cc b/storage/innobase/row/row0mysql.cc
index be221d5cfcd..5dcd596fc98 100644
--- a/storage/innobase/row/row0mysql.cc
+++ b/storage/innobase/row/row0mysql.cc
@@ -3817,7 +3817,7 @@ funct_exit_all_freed:
 
 	trx->op_info = "";
 
-	srv_inc_activity_count();
+	srv_wake_master_thread();
 
 	DBUG_RETURN(err);
 }
diff --git a/storage/innobase/srv/srv0srv.cc b/storage/innobase/srv/srv0srv.cc
index 6af3b89f15c..c5b08cd5eb8 100644
--- a/storage/innobase/srv/srv0srv.cc
+++ b/storage/innobase/srv/srv0srv.cc
@@ -1923,6 +1923,33 @@ srv_get_active_thread_type(void)
 	return(ret);
 }
 
+/** Wake up the InnoDB master thread if it was suspended (not sleeping). */
+void
+srv_active_wake_master_thread_low()
+{
+	ut_ad(!srv_read_only_mode);
+	ut_ad(!srv_sys_mutex_own());
+
+	srv_inc_activity_count();
+
+	if (my_atomic_loadlint(&srv_sys.n_threads_active[SRV_MASTER]) == 0) {
+		srv_slot_t*	slot;
+
+		srv_sys_mutex_enter();
+
+		slot = &srv_sys.sys_threads[SRV_MASTER_SLOT];
+
+		/* Only if the master thread has been started. */
+
+		if (slot->in_use) {
+			ut_a(srv_slot_get_type(slot) == SRV_MASTER);
+			os_event_set(slot->event);
+		}
+
+		srv_sys_mutex_exit();
+	}
+}
+
 /** Wake up the purge threads if there is work to do. */
 void
 srv_wake_purge_thread_if_not_active()
@@ -1937,6 +1964,14 @@ srv_wake_purge_thread_if_not_active()
 	}
 }
 
+/** Wake up the master thread if it is suspended or being suspended. */
+void
+srv_wake_master_thread()
+{
+	srv_inc_activity_count();
+	srv_release_threads(SRV_MASTER, 1);
+}
+
 /*******************************************************************//**
 Get current server activity count. We don't hold srv_sys::mutex while
 reading this value as it is only used in heuristics.
@@ -1948,20 +1983,15 @@ srv_get_activity_count(void)
 	return(srv_sys.activity_count);
 }
 
-/** Check if there has been any activity.
-@param[in,out]  activity_count  recent activity count to be returned
-if there is a change
+/*******************************************************************//**
+Check if there has been any activity.
 @return FALSE if no change in activity counter. */
-bool srv_check_activity(ulint  *activity_count)
+ibool
+srv_check_activity(
+/*===============*/
+	ulint		old_activity_count)	/*!< in: old activity count */
 {
-  ulint new_activity_count= srv_sys.activity_count;
-  if (new_activity_count != *activity_count)
-  {
-    *activity_count= new_activity_count;
-    return true;
-  }
-
-  return false;
+	return(srv_sys.activity_count != old_activity_count);
 }
 
 /********************************************************************//**
@@ -2362,30 +2392,52 @@ DECLARE_THREAD(srv_master_thread)(
 	slot = srv_reserve_slot(SRV_MASTER);
 	ut_a(slot == srv_sys.sys_threads);
 
+loop:
 	while (srv_shutdown_state <= SRV_SHUTDOWN_INITIATED) {
 		srv_master_sleep();
 
 		MONITOR_INC(MONITOR_MASTER_THREAD_SLEEP);
 
-		if (srv_check_activity(&old_activity_count)) {
+		if (srv_check_activity(old_activity_count)) {
+			old_activity_count = srv_get_activity_count();
 			srv_master_do_active_tasks();
 		} else {
 			srv_master_do_idle_tasks();
 		}
 	}
 
-        ut_ad(srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS
-              || srv_shutdown_state == SRV_SHUTDOWN_CLEANUP);
-
-	if (srv_shutdown_state == SRV_SHUTDOWN_CLEANUP
-            && srv_fast_shutdown < 2) {
-	        srv_shutdown(srv_fast_shutdown == 0);
+	switch (srv_shutdown_state) {
+	case SRV_SHUTDOWN_NONE:
+	case SRV_SHUTDOWN_INITIATED:
+		break;
+	case SRV_SHUTDOWN_FLUSH_PHASE:
+	case SRV_SHUTDOWN_LAST_PHASE:
+		ut_ad(0);
+		/* fall through */
+	case SRV_SHUTDOWN_EXIT_THREADS:
+		/* srv_init_abort() must have been invoked */
+	case SRV_SHUTDOWN_CLEANUP:
+		if (srv_shutdown_state == SRV_SHUTDOWN_CLEANUP
+		    && srv_fast_shutdown < 2) {
+			srv_shutdown(srv_fast_shutdown == 0);
+		}
+		srv_suspend_thread(slot);
+		my_thread_end();
+		os_thread_exit();
 	}
 
+	srv_main_thread_op_info = "suspending";
+
 	srv_suspend_thread(slot);
-	my_thread_end();
-	os_thread_exit();
-	OS_THREAD_DUMMY_RETURN;
+
+	/* DO NOT CHANGE THIS STRING. innobase_start_or_create_for_mysql()
+	waits for database activity to die down when converting < 4.1.x
+	databases, and relies on this string being exactly as it is. InnoDB
+	manual also mentions this string in several places. */
+	srv_main_thread_op_info = "waiting for server activity";
+
+	srv_resume_thread(slot);
+	goto loop;
 }
 
 /** @return whether purge should exit due to shutdown */
@@ -2553,13 +2605,15 @@ static ulint srv_do_purge(ulint* n_total_purged
 				++n_use_threads;
 			}
 
-		} else if (srv_check_activity(&old_activity_count)
+		} else if (srv_check_activity(old_activity_count)
 			   && n_use_threads > 1) {
 
 			/* History length same or smaller since last snapshot,
 			use fewer threads. */
 
 			--n_use_threads;
+
+			old_activity_count = srv_get_activity_count();
 		}
 
 		/* Ensure that the purge threads are less than what
diff --git a/storage/innobase/srv/srv0start.cc b/storage/innobase/srv/srv0start.cc
index 859e5a3eec2..318ab49cc9c 100644
--- a/storage/innobase/srv/srv0start.cc
+++ b/storage/innobase/srv/srv0start.cc
@@ -1178,7 +1178,7 @@ srv_shutdown_all_bg_threads()
 			if (srv_start_state_is_set(SRV_START_STATE_MASTER)) {
 				/* c. We wake the master thread so that
 				it exits */
-				srv_inc_activity_count();
+				srv_wake_master_thread();
 			}
 
 			if (srv_start_state_is_set(SRV_START_STATE_PURGE)) {
@@ -2526,7 +2526,7 @@ void srv_shutdown_bg_undo_sources()
 		fts_optimize_shutdown();
 		dict_stats_shutdown();
 		while (row_get_background_drop_list_len_low()) {
-			srv_inc_activity_count();
+			srv_wake_master_thread();
 			os_thread_yield();
 		}
 		srv_undo_sources = false;
diff --git a/storage/innobase/trx/trx0roll.cc b/storage/innobase/trx/trx0roll.cc
index c61a67ec05e..1be444adf07 100644
--- a/storage/innobase/trx/trx0roll.cc
+++ b/storage/innobase/trx/trx0roll.cc
@@ -164,6 +164,9 @@ trx_rollback_to_savepoint_low(
 
 	mem_heap_free(heap);
 
+	/* There might be work for utility threads.*/
+	srv_active_wake_master_thread();
+
 	MONITOR_DEC(MONITOR_TRX_ACTIVE);
 }
 
diff --git a/storage/innobase/trx/trx0trx.cc b/storage/innobase/trx/trx0trx.cc
index e1efc7c895f..1d4202b3033 100644
--- a/storage/innobase/trx/trx0trx.cc
+++ b/storage/innobase/trx/trx0trx.cc
@@ -1470,6 +1470,12 @@ trx_commit_in_memory(
 		}
 
 		trx->commit_lsn = lsn;
+
+		/* Tell server some activity has happened, since the trx
+		does changes something. Background utility threads like
+		master thread, purge thread or page_cleaner thread might
+		have some work to do. */
+		srv_active_wake_master_thread();
 	}
 
 	ut_ad(!trx->rsegs.m_noredo.undo);