Merge lp:~laurynas-biveinis/percona-server/xtradb-preflush-backoff into lp:percona-server/5.6

[Laurynas Biveinis]

http://jenkins.percona.com/job/percona-server-5.6-param/343/

Implement query thread waits at the sync preflush point, blueprint
https://blueprints.launchpad.net/percona-server/+spec/xtradb-preflush-backoff.

Whenever a query thread needs to perform a sync preflush to proceed,
two options are now available:
1) as previously, the query thread may issue a flush list batch itself
and wait for it to complete. This is also used whenever the page
cleaner thread is not running.
2) alternatively the query thread may wait until the flush list flush
is performed by the page cleaner thread. The wait is implemented by
maintaining a wait iteration counter, and a random duration sleep of up
to two to power of loop counter value is performed. To prevent
runaway sleeps, the counter is reset every 16 iterations.

The behavior is controlled by a new system variable
innodb_foreground_preflush with two possible values "sync_preflush"
and "exponential_backoff". Add a sys_vars test for this variable.

[Vadim Tkachenko (vadim-tk)]

LGTM

review: Approve

[Sergei Glushchenko (sergei.glushchenko)]

Approve

review: Approve (g2)

=== added file 'Percona-Server/mysql-test/suite/sys_vars/r/innodb_foreground_preflush_basic.result'
--- Percona-Server/mysql-test/suite/sys_vars/r/innodb_foreground_preflush_basic.result	1970-01-01 00:00:00 +0000
+++ Percona-Server/mysql-test/suite/sys_vars/r/innodb_foreground_preflush_basic.result	2013-10-03 13:54:00 +0000
@@ -0,0 +1,23 @@
+SET @start_value = @@GLOBAL.innodb_foreground_preflush;
+SELECT @@GLOBAL.innodb_foreground_preflush;
+@@GLOBAL.innodb_foreground_preflush
+exponential_backoff
+SELECT @@SESSION.innodb_foreground_preflush;
+ERROR HY000: Variable 'innodb_foreground_preflush' is a GLOBAL variable
+SET GLOBAL innodb_foreground_preflush='sync_preflush';
+SELECT @@GLOBAL.innodb_foreground_preflush;
+@@GLOBAL.innodb_foreground_preflush
+sync_preflush
+SET GLOBAL innodb_foreground_preflush='exponential_backoff';
+SELECT @@GLOBAL.innodb_foreground_preflush;
+@@GLOBAL.innodb_foreground_preflush
+exponential_backoff
+SET GLOBAL innodb_foreground_preflush=1.1;
+ERROR 42000: Incorrect argument type to variable 'innodb_foreground_preflush'
+SET GLOBAL innodb_foreground_preflush=1e1;
+ERROR 42000: Incorrect argument type to variable 'innodb_foreground_preflush'
+SET GLOBAL innodb_foreground_preflush=2;
+ERROR 42000: Variable 'innodb_foreground_preflush' can't be set to the value of '2'
+SET GLOBAL innodb_foreground_preflush='foo';
+ERROR 42000: Variable 'innodb_foreground_preflush' can't be set to the value of 'foo'
+SET GLOBAL innodb_foreground_preflush = @start_value;

=== added file 'Percona-Server/mysql-test/suite/sys_vars/t/innodb_foreground_preflush_basic.test'
--- Percona-Server/mysql-test/suite/sys_vars/t/innodb_foreground_preflush_basic.test	1970-01-01 00:00:00 +0000
+++ Percona-Server/mysql-test/suite/sys_vars/t/innodb_foreground_preflush_basic.test	2013-10-03 13:54:00 +0000
@@ -0,0 +1,30 @@
+--source include/have_innodb.inc
+
+# A dynamic, global variable
+
+SET @start_value = @@GLOBAL.innodb_foreground_preflush;
+
+# Default value
+SELECT @@GLOBAL.innodb_foreground_preflush;
+
+# Global only
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+SELECT @@SESSION.innodb_foreground_preflush;
+
+# Correct values
+SET GLOBAL innodb_foreground_preflush='sync_preflush';
+SELECT @@GLOBAL.innodb_foreground_preflush;
+SET GLOBAL innodb_foreground_preflush='exponential_backoff';
+SELECT @@GLOBAL.innodb_foreground_preflush;
+
+# Incorrect values
+--error ER_WRONG_TYPE_FOR_VAR
+SET GLOBAL innodb_foreground_preflush=1.1;
+--error ER_WRONG_TYPE_FOR_VAR
+SET GLOBAL innodb_foreground_preflush=1e1;
+--error ER_WRONG_VALUE_FOR_VAR
+SET GLOBAL innodb_foreground_preflush=2;
+--error ER_WRONG_VALUE_FOR_VAR
+SET GLOBAL innodb_foreground_preflush='foo';
+
+SET GLOBAL innodb_foreground_preflush = @start_value;

=== modified file 'Percona-Server/storage/innobase/handler/ha_innodb.cc'
--- Percona-Server/storage/innobase/handler/ha_innodb.cc	2013-10-03 10:03:41 +0000
+++ Percona-Server/storage/innobase/handler/ha_innodb.cc	2013-10-03 13:54:00 +0000
@@ -250,6 +250,21 @@
 	NULL
 };
 
+/** Possible values for system variable "innodb_foreground_preflush".  */
+static const char* innodb_foreground_preflush_names[] = {
+	"sync_preflush",
+	"exponential_backoff",
+	NullS
+};
+
+/* Enumeration for innodb_foreground_preflush.  */
+static TYPELIB innodb_foreground_preflush_typelib = {
+	array_elements(innodb_foreground_preflush_names) - 1,
+	"innodb_foreground_preflush_typelib",
+	innodb_foreground_preflush_names,
+	NULL
+};
+
 /* 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
@@ -16720,6 +16735,15 @@
   "established by the buffer pool memory region. Disabled by default.",
   NULL, NULL, FALSE);
 
+static MYSQL_SYSVAR_ENUM(foreground_preflush, srv_foreground_preflush,
+  PLUGIN_VAR_OPCMDARG,
+  "The algorithm InnoDB uses for the query threads at sync preflush.  "
+  "Possible values are "
+  "SYNC_PREFLUSH: perform a sync preflush as Oracle MySQL; "
+  "EXPONENTIAL_BACKOFF: (default) wait for the page cleaner flush.",
+  NULL, NULL, SRV_FOREGROUND_PREFLUSH_EXP_BACKOFF,
+  &innodb_foreground_preflush_typelib);
+
 #ifdef UNIV_LINUX
 
 static MYSQL_SYSVAR_ULONG(sched_priority_cleaner, srv_sched_priority_cleaner,
@@ -17505,6 +17529,7 @@
   MYSQL_SYSVAR(cleaner_eviction_factor),
 #endif /* defined UNIV_DEBUG || defined UNIV_PERF_DEBUG */
   MYSQL_SYSVAR(cleaner_lsn_age_factor),
+  MYSQL_SYSVAR(foreground_preflush),
   MYSQL_SYSVAR(print_all_deadlocks),
   MYSQL_SYSVAR(cmp_per_index_enabled),
   MYSQL_SYSVAR(undo_logs),

=== modified file 'Percona-Server/storage/innobase/include/buf0types.h'
--- Percona-Server/storage/innobase/include/buf0types.h	2013-10-03 07:45:17 +0000
+++ Percona-Server/storage/innobase/include/buf0types.h	2013-10-03 13:54:00 +0000
@@ -107,6 +107,17 @@
 						checkpoint age higher.  */
 };
 
+/** Alternatives for srv_foreground_preflush, set through
+innodb_foreground_preflush variable  */
+enum srv_foreground_preflush_t {
+	SRV_FOREGROUND_PREFLUSH_SYNC_PREFLUSH,	/*!< Original Oracle MySQL 5.6
+						behavior of performing a sync
+						flush list flush  */
+	SRV_FOREGROUND_PREFLUSH_EXP_BACKOFF	/*!< Exponential backoff wait
+						for the page cleaner to flush
+						for us  */
+};
+
 /** Parameters of binary buddy system for compressed pages (buf0buddy.h) */
 /* @{ */
 /** Zip shift value for the smallest page size */

=== modified file 'Percona-Server/storage/innobase/include/srv0srv.h'
--- Percona-Server/storage/innobase/include/srv0srv.h	2013-10-03 10:03:41 +0000
+++ Percona-Server/storage/innobase/include/srv0srv.h	2013-10-03 13:54:00 +0000
@@ -309,6 +309,8 @@
 extern ulint	srv_mem_pool_size;
 extern ulint	srv_lock_table_size;
 
+extern ulint	srv_foreground_preflush;/*!< Query thread preflush algorithm */
+
 extern ulint	srv_cleaner_max_lru_time;/*!< the maximum time limit for a
 					single LRU tail flush iteration by the
 					page cleaner thread */

=== modified file 'Percona-Server/storage/innobase/log/log0log.cc'
--- Percona-Server/storage/innobase/log/log0log.cc	2013-09-26 04:57:55 +0000
+++ Percona-Server/storage/innobase/log/log0log.cc	2013-10-03 13:54:00 +0000
@@ -1716,8 +1716,8 @@
 	lsn_t	new_oldest)	/*!< in: try to advance oldest_modified_lsn
 				at least to this lsn */
 {
-	bool	success;
-	ulint	n_pages;
+	lsn_t	current_oldest;
+	ulint	i;
 
 	if (recv_recovery_on) {
 		/* If the recovery is running, we must first apply all
@@ -1732,21 +1732,45 @@
 		recv_apply_hashed_log_recs(TRUE);
 	}
 
-	success = buf_flush_list(ULINT_MAX, new_oldest, &n_pages);
-
-	buf_flush_wait_batch_end(NULL, BUF_FLUSH_LIST);
-
-	if (!success) {
-		MONITOR_INC(MONITOR_FLUSH_SYNC_WAITS);
-	}
-
-	MONITOR_INC_VALUE_CUMULATIVE(
-		MONITOR_FLUSH_SYNC_TOTAL_PAGE,
-		MONITOR_FLUSH_SYNC_COUNT,
-		MONITOR_FLUSH_SYNC_PAGES,
-		n_pages);
-
-	return(success);
+	if (!buf_page_cleaner_is_active
+	    || (srv_foreground_preflush
+		== SRV_FOREGROUND_PREFLUSH_SYNC_PREFLUSH)) {
+
+		ulint n_pages;
+
+		bool success = buf_flush_list(ULINT_MAX, new_oldest, &n_pages);
+
+		buf_flush_wait_batch_end(NULL, BUF_FLUSH_LIST);
+
+		if (!success) {
+			MONITOR_INC(MONITOR_FLUSH_SYNC_WAITS);
+		}
+
+		MONITOR_INC_VALUE_CUMULATIVE(
+			MONITOR_FLUSH_SYNC_TOTAL_PAGE,
+			MONITOR_FLUSH_SYNC_COUNT,
+			MONITOR_FLUSH_SYNC_PAGES,
+			n_pages);
+
+		return(success);
+	}
+
+	ut_ad(srv_foreground_preflush == SRV_FOREGROUND_PREFLUSH_EXP_BACKOFF);
+
+	current_oldest = buf_pool_get_oldest_modification();
+	i = 0;
+
+	while (buf_page_cleaner_is_active && current_oldest < new_oldest
+	       && current_oldest) {
+
+		os_thread_sleep(ut_rnd_interval(0, 1 << i));
+		i++;
+		i %= 16;
+
+		current_oldest = buf_pool_get_oldest_modification();
+	}
+
+	return(current_oldest >= new_oldest || !current_oldest);
 }
 
 /******************************************************//**

=== modified file 'Percona-Server/storage/innobase/srv/srv0srv.cc'
--- Percona-Server/storage/innobase/srv/srv0srv.cc	2013-10-03 12:42:40 +0000
+++ Percona-Server/storage/innobase/srv/srv0srv.cc	2013-10-03 13:54:00 +0000
@@ -261,6 +261,10 @@
 UNIV_INTERN ulint	srv_mem_pool_size	= ULINT_MAX;
 UNIV_INTERN ulint	srv_lock_table_size	= ULINT_MAX;
 
+/** Query thread preflush algorithm */
+UNIV_INTERN ulint	srv_foreground_preflush
+	= SRV_FOREGROUND_PREFLUSH_EXP_BACKOFF;
+
 /** The maximum time limit for a single LRU tail flush iteration by the page
 cleaner thread */
 UNIV_INTERN ulint	srv_cleaner_max_lru_time = 1000;