MDEV-14477 InnoDB update_time is wrongly updated after partial rollback or internal COMMIT

The non-persistent UPDATE_TIME for InnoDB tables was not being
updated consistently at transaction commit.

If a transaction is partly rolled back so that in the end it will
not modify a table that it intended to modify, the update_time will
be updated nevertheless. This will also happen when InnoDB fails
to write an undo log record for the intended modification.

If a transaction is committed internally in InnoDB, instead of
being committed from the SQL interface, then the trx_t::mod_tables
will not be applied to the update_time of the tables.

trx_t::mod_tables: Replace the std::set<dict_table_t*>
with std::map<dict_table_t*,undo_no_t>, so that the very first
modification within the transaction is identified.

trx_undo_report_row_operation(): Update mod_tables for every operation
after the undo log record was successfully written.

trx_rollback_to_savepoint_low(): After partial rollback, erase from
trx_t::mod_tables any tables for which all changes were rolled back.

trx_commit_in_memory(): Tighten some assertions and simplify conditions.
Invoke trx_update_mod_tables_timestamp() if persistent tables were
affected.

trx_commit_for_mysql(): Remove the call to
trx_update_mod_tables_timestamp(), as it is now invoked at the
lower level, in trx_commit_in_memory().

trx_rollback_finish(): Clear mod_tables before invoking trx_commit(),
because the trx_commit_in_memory() would otherwise wrongly process
mod_tables after a full ROLLBACK.

mysql-test/suite/innodb/r/update_time_wl6658.result

+CREATE TABLE tab1(c1 int,c2 varchar(30), c3 BLOB) ENGINE=InnoDB;
+CREATE TABLE tab1u LIKE tab1;
+CREATE TABLE tab1d LIKE tab1;
+CREATE TABLE tab1i LIKE tab1;
+CREATE TABLE tab3(c1 int,c2 varchar(30)) ENGINE=InnoDB;
+CREATE TABLE tab4(c1 int,c2 varchar(30)) ENGINE=InnoDB;
+CREATE TABLE tab5(c1 int,c2 varchar(30)) ENGINE=InnoDB;
+INSERT INTO tab1u VALUES(1,'Testing the wl6658','Testing the wl6658');
+INSERT INTO tab1d VALUES(1,'Updated','Updated');
+INSERT INTO tab4 VALUES(1,'Test for Update');
+INSERT INTO tab5 VALUES(1,'Test for Delete');
+CREATE TRIGGER test_trig BEFORE INSERT ON tab1
+FOR EACH ROW BEGIN
+INSERT INTO tab3 VALUES(1,'Inserted From Trigger');
+UPDATE tab4 SET c2='Updated from Trigger' WHERE c1=1;
+DELETE FROM tab5;
+END |
+CREATE TABLE tab2(
+id INT NOT NULL,
+store_name VARCHAR(30),
+parts VARCHAR(30),
+store_id INT
+) ENGINE=InnoDB
+PARTITION BY LIST(store_id) (
+PARTITION pNorth VALUES IN (10,20,30),
+PARTITION pEast VALUES IN (40,50,60),
+PARTITION pWest VALUES IN (70,80,100)
+);
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab2';
+update_time
+NULL
+CREATE PROCEDURE proc_wl6658()
+BEGIN
+INSERT INTO tab2 VALUES(1,'ORACLE','NUTT',10);
+INSERT INTO tab2 VALUES(2,'HUAWEI','BOLT',40);
+COMMIT;
+END |
+CALL proc_wl6658;
+SELECT * FROM tab2 ORDER BY id,store_id;
+id	store_name	parts	store_id
+1	ORACLE	NUTT	10
+2	HUAWEI	BOLT	40
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab2';
+COUNT(update_time)
+1
+TRUNCATE TABLE tab2;
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab2';
+COUNT(update_time)
+1
+CREATE TABLE tab7(c1 INT NOT NULL, PRIMARY KEY (c1)) ENGINE=INNODB;
+CREATE TABLE tab8(c1 INT PRIMARY KEY,c2 INT,
+FOREIGN KEY (c2) REFERENCES tab7(c1)  ON DELETE CASCADE )
+ENGINE=INNODB;
+SELECT table_name,update_time
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+table_name	update_time
+tab7	NULL
+tab8	NULL
+INSERT INTO tab7 VALUES(1);
+INSERT INTO tab8 VALUES(1,1);
+SELECT table_name,COUNT(update_time)
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+table_name	COUNT(update_time)
+tab7	1
+tab8	1
+#restart the server
+SELECT table_name,update_time
+FROM information_schema.tables
+WHERE table_name IN ('tab1','tab2','tab3','tab4','tab5','tab7','tab8')
+ORDER BY table_name;
+table_name	update_time
+tab1	NULL
+tab2	NULL
+tab3	NULL
+tab4	NULL
+tab5	NULL
+tab7	NULL
+tab8	NULL
+#case1:
+BEGIN WORK;
+INSERT INTO tab1
+VALUES(1,'Testing the wl6658', 'Testing the wl6658');
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1';
+update_time
+NULL
+COMMIT;
+SELECT * FROM tab1;
+c1	c2	c3
+1	Testing the wl6658	Testing the wl6658
+SELECT * FROM tab3;
+c1	c2
+1	Inserted From Trigger
+SELECT * FROM tab4;
+c1	c2
+1	Updated from Trigger
+SELECT * FROM tab5;
+c1	c2
+SELECT table_name,COUNT(update_time)
+FROM information_schema.tables
+WHERE table_name IN ('tab1','tab3','tab4','tab5')
+GROUP BY table_name ORDER BY table_name;
+table_name	COUNT(update_time)
+tab1	1
+tab3	1
+tab4	1
+tab5	1
+Testcase with UPDATE stmt and transaction
+SELECT * FROM tab1u;
+c1	c2	c3
+1	Testing the wl6658	Testing the wl6658
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1u';
+update_time
+NULL
+#case2:
+START TRANSACTION;
+UPDATE tab1u SET c2='Updated',c3='Updated' WHERE c1=1;
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1u';
+update_time
+NULL
+COMMIT;
+SELECT * FROM tab1u;
+c1	c2	c3
+1	Updated	Updated
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab1u';
+COUNT(update_time)
+1
+SELECT * FROM tab1d;
+c1	c2	c3
+1	Updated	Updated
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1d';
+update_time
+NULL
+#case3:
+START TRANSACTION;
+DELETE FROM tab1d;
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1d';
+update_time
+NULL
+COMMIT;
+SELECT * FROM tab1d;
+c1	c2	c3
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab1d';
+COUNT(update_time)
+1
+SELECT * FROM tab1i;
+c1	c2	c3
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1i';
+update_time
+NULL
+#case4:
+START TRANSACTION;
+INSERT INTO tab1i
+VALUES(1,'Testing the wl6658', 'Testing the wl6658');
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1i';
+update_time
+NULL
+ROLLBACK;
+SELECT * FROM tab1i;
+c1	c2	c3
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1i';
+update_time
+NULL
+BEGIN WORK;
+DELETE FROM tab1i;
+SAVEPOINT A;
+INSERT INTO tab2 VALUES(1,'Oracle','NUTT',10);
+INSERT INTO tab2 VALUES(2,'HUAWEI','BOLT',40);
+SAVEPOINT B;
+INSERT INTO tab2 VALUES(3,'IBM','NAIL',70);
+SAVEPOINT C;
+ROLLBACK to A;
+SELECT * FROM tab2;
+id	store_name	parts	store_id
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab2';
+update_time
+NULL
+#execute DDL instead of commit
+create table tab6(c1 int);
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab2';
+update_time
+NULL
+START TRANSACTION;
+DELETE FROM tab7;
+ROLLBACK;
+SELECT * FROM tab7;
+c1
+1
+SELECT * FROM tab8;
+c1	c2
+1	1
+SELECT table_name,update_time
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+table_name	update_time
+tab7	NULL
+tab8	NULL
+DELETE FROM tab7;
+SELECT * FROM tab7;
+c1
+SELECT * FROM tab8;
+c1	c2
+SELECT table_name,COUNT(update_time)
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+table_name	COUNT(update_time)
+tab7	1
+tab8	1
+#cleanup
+DROP TRIGGER test_trig;
+DROP TABLE tab1,tab1u,tab1d,tab1i,tab2,tab3,tab4,tab5,tab6,tab8,tab7;
+DROP PROCEDURE proc_wl6658;

mysql-test/suite/innodb/t/update_time_wl6658.test

+###################################################################
+#Testing functionality of the WL6658
+#case1: begin work with INSERT with Triggers
+#case2: (tab1u) begin transaction with UPDATE
+#case3: (tab1d) begin transaction with DELETE
+#case4: (tab1i) Rollback & INSERT
+#case5: (tab2) partitioned table and procedures
+#case6: (tab2) SAVEPOINT
+#case7: (tab7,tab8) pk-fk with ON DELETE CASCADE
+###################################################################
+--source include/no_valgrind_without_big.inc
+--source include/have_innodb.inc
+--source include/have_partition.inc
+--source include/not_embedded.inc
+
+CREATE TABLE tab1(c1 int,c2 varchar(30), c3 BLOB) ENGINE=InnoDB;
+CREATE TABLE tab1u LIKE tab1;
+CREATE TABLE tab1d LIKE tab1;
+CREATE TABLE tab1i LIKE tab1;
+CREATE TABLE tab3(c1 int,c2 varchar(30)) ENGINE=InnoDB;
+CREATE TABLE tab4(c1 int,c2 varchar(30)) ENGINE=InnoDB;
+CREATE TABLE tab5(c1 int,c2 varchar(30)) ENGINE=InnoDB;
+
+INSERT INTO tab1u VALUES(1,'Testing the wl6658','Testing the wl6658');
+INSERT INTO tab1d VALUES(1,'Updated','Updated');
+INSERT INTO tab4 VALUES(1,'Test for Update');
+INSERT INTO tab5 VALUES(1,'Test for Delete');
+
+delimiter |;
+
+CREATE TRIGGER test_trig BEFORE INSERT ON tab1
+FOR EACH ROW BEGIN
+    INSERT INTO tab3 VALUES(1,'Inserted From Trigger');
+    UPDATE tab4 SET c2='Updated from Trigger' WHERE c1=1;
+    DELETE FROM tab5;
+END |
+
+delimiter ;|
+
+CREATE TABLE tab2(
+    id INT NOT NULL,
+    store_name VARCHAR(30),
+    parts VARCHAR(30),
+    store_id INT
+) ENGINE=InnoDB
+PARTITION BY LIST(store_id) (
+    PARTITION pNorth VALUES IN (10,20,30),
+    PARTITION pEast VALUES IN (40,50,60),
+    PARTITION pWest VALUES IN (70,80,100)
+);
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab2';
+
+delimiter |;
+
+CREATE PROCEDURE proc_wl6658()
+BEGIN
+INSERT INTO tab2 VALUES(1,'ORACLE','NUTT',10);
+INSERT INTO tab2 VALUES(2,'HUAWEI','BOLT',40);
+COMMIT;
+END |
+
+delimiter ;|
+
+CALL proc_wl6658;
+
+SELECT * FROM tab2 ORDER BY id,store_id;
+
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab2';
+
+TRUNCATE TABLE tab2;
+
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab2';
+
+CREATE TABLE tab7(c1 INT NOT NULL, PRIMARY KEY (c1)) ENGINE=INNODB;
+
+CREATE TABLE tab8(c1 INT PRIMARY KEY,c2 INT,
+FOREIGN KEY (c2) REFERENCES tab7(c1)  ON DELETE CASCADE )
+ENGINE=INNODB;
+
+SELECT table_name,update_time
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+
+INSERT INTO tab7 VALUES(1);
+
+INSERT INTO tab8 VALUES(1,1);
+
+SELECT table_name,COUNT(update_time)
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+
+--echo #restart the server
+--source include/restart_mysqld.inc
+
+SELECT table_name,update_time
+FROM information_schema.tables
+WHERE table_name IN ('tab1','tab2','tab3','tab4','tab5','tab7','tab8')
+ORDER BY table_name;
+
+--echo #case1:
+
+BEGIN WORK;
+
+INSERT INTO tab1
+VALUES(1,'Testing the wl6658', 'Testing the wl6658');
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1';
+
+COMMIT;
+
+SELECT * FROM tab1;
+SELECT * FROM tab3;
+SELECT * FROM tab4;
+SELECT * FROM tab5;
+
+SELECT table_name,COUNT(update_time)
+FROM information_schema.tables
+WHERE table_name IN ('tab1','tab3','tab4','tab5')
+GROUP BY table_name ORDER BY table_name;
+
+--echo Testcase with UPDATE stmt and transaction
+
+SELECT * FROM tab1u;
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1u';
+
+--echo #case2:
+START TRANSACTION;
+
+UPDATE tab1u SET c2='Updated',c3='Updated' WHERE c1=1;
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1u';
+
+COMMIT;
+
+SELECT * FROM tab1u;
+
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab1u';
+
+SELECT * FROM tab1d;
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1d';
+
+--echo #case3:
+START TRANSACTION;
+
+DELETE FROM tab1d;
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1d';
+
+COMMIT;
+
+SELECT * FROM tab1d;
+
+SELECT COUNT(update_time)
+FROM information_schema.tables WHERE table_name='tab1d';
+
+SELECT * FROM tab1i;
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1i';
+
+--echo #case4:
+START TRANSACTION;
+
+INSERT INTO tab1i
+VALUES(1,'Testing the wl6658', 'Testing the wl6658');
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1i';
+
+ROLLBACK;
+
+SELECT * FROM tab1i;
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab1i';
+
+BEGIN WORK;
+DELETE FROM tab1i;
+SAVEPOINT A;
+INSERT INTO tab2 VALUES(1,'Oracle','NUTT',10);
+INSERT INTO tab2 VALUES(2,'HUAWEI','BOLT',40);
+SAVEPOINT B;
+INSERT INTO tab2 VALUES(3,'IBM','NAIL',70);
+SAVEPOINT C;
+ROLLBACK to A;
+
+SELECT * FROM tab2;
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab2';
+
+--echo #execute DDL instead of commit
+create table tab6(c1 int);
+
+SELECT update_time
+FROM information_schema.tables WHERE table_name='tab2';
+
+START TRANSACTION;
+DELETE FROM tab7;
+ROLLBACK;
+
+SELECT * FROM tab7;
+SELECT * FROM tab8;
+
+SELECT table_name,update_time
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+
+DELETE FROM tab7;
+SELECT * FROM tab7;
+SELECT * FROM tab8;
+
+SELECT table_name,COUNT(update_time)
+FROM information_schema.tables WHERE table_name IN ('tab7','tab8')
+GROUP BY table_name ORDER BY table_name;
+
+--echo #cleanup
+DROP TRIGGER test_trig;
+DROP TABLE tab1,tab1u,tab1d,tab1i,tab2,tab3,tab4,tab5,tab6,tab8,tab7;
+DROP PROCEDURE proc_wl6658;

storage/innobase/handler/ha_innodb.cc

@@ -13350,7 +13350,7 @@ ha_innobase::delete_table(
 	     iter != parent_trx->mod_tables.end();
 	     ++iter) {
 
-		dict_table_t*	table_to_drop = *iter;
+		dict_table_t*	table_to_drop = iter->first;
 
 		if (strcmp(norm_name, table_to_drop->name.m_name) == 0) {
 			parent_trx->mod_tables.erase(table_to_drop);

storage/innobase/include/trx0trx.h

@@ -779,12 +779,13 @@ struct trx_lock_t {
 	bool		start_stmt;
 };
 
-/** Type used to store the list of tables that are modified by a given
-transaction. We store pointers to the table objects in memory because
+/** Collection of persistent tables and their first modification
+in a transaction.
+We store pointers to the table objects in memory because
 we know that a table object will not be destroyed while a transaction
 that modified it is running. */
-typedef std::set<
-	dict_table_t*,
+typedef std::map<
+	dict_table_t*, undo_no_t,
 	std::less<dict_table_t*>,
 	ut_allocator<dict_table_t*> >	trx_mod_tables_t;
 

storage/innobase/trx/trx0rec.cc

@@ -1927,16 +1927,6 @@ trx_undo_report_row_operation(
 	} else {
 		ut_ad(!trx->read_only);
 		ut_ad(trx->id);
-		if (UNIV_LIKELY(!clust_entry || clust_entry->info_bits
-				!= REC_INFO_DEFAULT_ROW)) {
-			/* Keep INFORMATION_SCHEMA.TABLES.UPDATE_TIME
-			up-to-date for persistent tables outside
-			instant ADD COLUMN. */
-			trx->mod_tables.insert(index->table);
-		} else {
-			ut_ad(index->is_instant());
-		}
-
 		pundo = &trx->rsegs.m_redo.undo;
 		rseg = trx->rsegs.m_redo.rseg;
 	}
@@ -2022,6 +2012,13 @@ trx_undo_report_row_operation(
 
 			mutex_exit(&trx->undo_mutex);
 
+			if (!is_temp) {
+				trx->mod_tables.insert(
+					trx_mod_tables_t::value_type(
+						index->table,
+						undo->top_undo_no));
+			}
+
 			*roll_ptr = trx_undo_build_roll_ptr(
 				!rec, rseg->id, page_no, offset);
 			return(DB_SUCCESS);

storage/innobase/trx/trx0roll.cc

@@ -126,6 +126,14 @@ trx_rollback_to_savepoint_low(
 		trx_rollback_finish(trx);
 		MONITOR_INC(MONITOR_TRX_ROLLBACK);
 	} else {
+		const undo_no_t limit = savept->least_undo_no;
+		for (trx_mod_tables_t::iterator i = trx->mod_tables.begin();
+		     i != trx->mod_tables.end(); ) {
+			trx_mod_tables_t::iterator j = i++;
+			if (j->second >= limit) {
+				trx->mod_tables.erase(j);
+			}
+		}
 		trx->lock.que_state = TRX_QUE_RUNNING;
 		MONITOR_INC(MONITOR_TRX_ROLLBACK_SAVEPOINT);
 	}
@@ -1154,10 +1162,8 @@ trx_rollback_finish(
 /*================*/
 	trx_t*		trx)	/*!< in: transaction */
 {
-	trx_commit(trx);
-
 	trx->mod_tables.clear();
-
+	trx_commit(trx);
 	trx->lock.que_state = TRX_QUE_RUNNING;
 }
 

storage/innobase/trx/trx0trx.cc

@@ -783,7 +783,9 @@ trx_resurrect_table_locks(
 			}
 
 			if (trx->state == TRX_STATE_PREPARED) {
-				trx->mod_tables.insert(table);
+				trx->mod_tables.insert(
+					trx_mod_tables_t::value_type(table,
+								     0));
 			}
 			lock_table_ix_resurrect(table, trx);
 
@@ -1619,7 +1621,7 @@ trx_update_mod_tables_timestamp(
 		"garbage" in table->update_time is justified because
 		protecting it with a latch here would be too performance
 		intrusive. */
-		(*it)->update_time = now;
+		it->first->update_time = now;
 	}
 
 	trx->mod_tables.clear();
@@ -1725,8 +1727,12 @@ trx_commit_in_memory(
 		trx->state = TRX_STATE_NOT_STARTED;
 
 	} else {
+		const bool rw = trx->rsegs.m_redo.rseg != NULL;
 
-		if (trx->id > 0) {
+		ut_ad(!trx->read_only || !rw);
+		ut_ad(trx->id || !rw);
+
+		if (rw) {
 			/* For consistent snapshot, we need to remove current
 			transaction from running transaction id list for mvcc
 			before doing commit and releasing locks. */
@@ -1741,16 +1747,14 @@ trx_commit_in_memory(
 		ut_ad(trx_state_eq(trx, TRX_STATE_COMMITTED_IN_MEMORY));
 		DEBUG_SYNC_C("after_trx_committed_in_memory");
 
-		if (trx->read_only || trx->rsegs.m_redo.rseg == NULL) {
-
+		if (!rw) {
 			MONITOR_INC(MONITOR_TRX_RO_COMMIT);
 			if (trx->read_view != NULL) {
 				trx_sys->mvcc->view_close(
 					trx->read_view, false);
 			}
-
 		} else {
-			ut_ad(trx->id > 0);
+			trx_update_mod_tables_timestamp(trx);
 			MONITOR_INC(MONITOR_TRX_RW_COMMIT);
 		}
 	}
@@ -2205,10 +2209,6 @@ trx_commit_for_mysql(
 
 		trx->op_info = "committing";
 
-		if (trx->id != 0) {
-			trx_update_mod_tables_timestamp(trx);
-		}
-
 		trx_commit(trx);
 
 		MONITOR_DEC(MONITOR_TRX_ACTIVE);