MDEV-27962 Instant DDL downgrades the MDL when table is empty

- Server incorrectly downgrading the MDL after prepare phase when
table is empty. mdl_exclusive_after_prepare is being set in
prepare phase only. But mdl_exclusive_after_prepare condition was
misplaced and checked before prepare phase by
commit d270525dfd and it is now
changed to check after prepare phase.

 - main.innodb_mysql_sync test case was changed to avoid locking
optimization when table is empty.

mysql-test/main/innodb_mysql_sync.result

@@ -131,6 +131,7 @@ connection default;
 DROP DATABASE db1;
 # Test 2: Primary index (implicit), should block writes.
 CREATE TABLE t1(a INT NOT NULL, b INT NOT NULL) engine=innodb;
+INSERT INTO t1 VALUES(1, 2);
 SET DEBUG_SYNC= "alter_table_inplace_after_lock_downgrade SIGNAL manage WAIT_FOR query";
 # Sending:
 ALTER TABLE t1 ADD UNIQUE INDEX(a), LOCK=SHARED;
@@ -139,15 +140,16 @@ SET DEBUG_SYNC= "now WAIT_FOR manage";
 USE test;
 SELECT * FROM t1;
 a	b
+1	2
 # Sending:
-UPDATE t1 SET a=NULL;
+UPDATE t1 SET a=3;
 connection con2;
 # Waiting for SELECT to be blocked by the metadata lock on t1
 SET DEBUG_SYNC= "now SIGNAL query";
 connection default;
 # Reaping: ALTER TABLE t1 ADD UNIQUE INDEX(a)
 connection con1;
-# Reaping: UPDATE t1 SET a=NULL
+# Reaping: UPDATE t1 SET a=3
 # Test 3: Primary index (explicit), should block writes.
 connection default;
 ALTER TABLE t1 DROP INDEX a;
@@ -158,15 +160,16 @@ connection con1;
 SET DEBUG_SYNC= "now WAIT_FOR manage";
 SELECT * FROM t1;
 a	b
+3	2
 # Sending:
-UPDATE t1 SET a=NULL;
+UPDATE t1 SET a=4;
 connection con2;
 # Waiting for SELECT to be blocked by the metadata lock on t1
 SET DEBUG_SYNC= "now SIGNAL query";
 connection default;
 # Reaping: ALTER TABLE t1 ADD PRIMARY KEY (a)
 connection con1;
-# Reaping: UPDATE t1 SET a=NULL
+# Reaping: UPDATE t1 SET a=4
 # Test 4: Secondary unique index, should not block reads.
 connection default;
 SET DEBUG_SYNC= "alter_table_inplace_after_lock_downgrade SIGNAL manage WAIT_FOR query";
@@ -176,6 +179,7 @@ connection con1;
 SET DEBUG_SYNC= "now WAIT_FOR manage";
 SELECT * FROM t1;
 a	b
+4	2
 SET DEBUG_SYNC= "now SIGNAL query";
 connection default;
 # Reaping: ALTER TABLE t1 ADD UNIQUE (b)

mysql-test/main/innodb_mysql_sync.test

@@ -176,6 +176,7 @@ DROP DATABASE db1;
 --echo # Test 2: Primary index (implicit), should block writes.
 
 CREATE TABLE t1(a INT NOT NULL, b INT NOT NULL) engine=innodb;
+INSERT INTO t1 VALUES(1, 2);
 SET DEBUG_SYNC= "alter_table_inplace_after_lock_downgrade SIGNAL manage WAIT_FOR query";
 --echo # Sending:
 --send ALTER TABLE t1 ADD UNIQUE INDEX(a), LOCK=SHARED
@@ -185,13 +186,13 @@ SET DEBUG_SYNC= "now WAIT_FOR manage";
 USE test;
 SELECT * FROM t1;
 --echo # Sending:
---send UPDATE t1 SET a=NULL
+--send UPDATE t1 SET a=3
 
 connection con2;
 --echo # Waiting for SELECT to be blocked by the metadata lock on t1
 let $wait_condition= SELECT COUNT(*)= 1 FROM information_schema.processlist
   WHERE state= 'Waiting for table metadata lock'
-  AND info='UPDATE t1 SET a=NULL';
+  AND info='UPDATE t1 SET a=3';
 --source include/wait_condition.inc
 SET DEBUG_SYNC= "now SIGNAL query";
 
@@ -200,7 +201,7 @@ connection default;
 --reap
 
 connection con1;
---echo # Reaping: UPDATE t1 SET a=NULL
+--echo # Reaping: UPDATE t1 SET a=3
 --reap
 
 --echo # Test 3: Primary index (explicit), should block writes.
@@ -215,13 +216,13 @@ connection con1;
 SET DEBUG_SYNC= "now WAIT_FOR manage";
 SELECT * FROM t1;
 --echo # Sending:
---send UPDATE t1 SET a=NULL
+--send UPDATE t1 SET a=4
 
 connection con2;
 --echo # Waiting for SELECT to be blocked by the metadata lock on t1
 let $wait_condition= SELECT COUNT(*)= 1 FROM information_schema.processlist
   WHERE state= 'Waiting for table metadata lock'
-  AND info='UPDATE t1 SET a=NULL';
+  AND info='UPDATE t1 SET a=4';
 --source include/wait_condition.inc
 SET DEBUG_SYNC= "now SIGNAL query";
 
@@ -230,7 +231,7 @@ connection default;
 --reap
 
 connection con1;
---echo # Reaping: UPDATE t1 SET a=NULL
+--echo # Reaping: UPDATE t1 SET a=4
 --reap
 
 --echo # Test 4: Secondary unique index, should not block reads.

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

@@ -462,12 +462,27 @@ INSERT INTO t1 SET a=0, i=REPEAT('1', 10000);
 ROLLBACK;
 set DEBUG_SYNC='now SIGNAL go';
 connection default;
-disconnect con1;
 SELECT * FROM t1;
 a	b	c	d	e	f	g	h	i
 1	2	3	4	5	6	7	8	test
 DROP TABLE t1;
 SET DEBUG_SYNC=RESET;
+#
+# MDEV-27962 Instant DDL downgrades the MDL when table is empty
+#
+CREATE TABLE t1(f1 INT NOT NULL, f2 INT NOT NULL)ENGINE=InnoDB;
+SET DEBUG_SYNC="alter_table_inplace_after_lock_downgrade SIGNAL try_insert WAIT_FOR alter_progress";
+ALTER TABLE t1 ADD INDEX(f1), ADD INDEX(f2);
+connection con1;
+SET SESSION lock_wait_timeout=1;
+SET DEBUG_SYNC="now WAIT_FOR try_insert";
+INSERT INTO t1 VALUES(1, 2);
+ERROR HY000: Lock wait timeout exceeded; try restarting transaction
+SET DEBUG_SYNC="now SIGNAL alter_progress";
+disconnect con1;
+connection default;
+DROP TABLE t1;
+SET DEBUG_SYNC=reset;
 # End of 10.4 tests
 SET GLOBAL innodb_purge_rseg_truncate_frequency = @save_frequency;
 SELECT variable_value-@old_instant instants

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

@@ -533,11 +533,28 @@ set DEBUG_SYNC='now SIGNAL go';
 connection default;
 reap;
 
-disconnect con1;
 SELECT * FROM t1;
 DROP TABLE t1;
 SET DEBUG_SYNC=RESET;
 
+--echo #
+--echo # MDEV-27962 Instant DDL downgrades the MDL when table is empty
+--echo #
+CREATE TABLE t1(f1 INT NOT NULL, f2 INT NOT NULL)ENGINE=InnoDB;
+SET DEBUG_SYNC="alter_table_inplace_after_lock_downgrade SIGNAL try_insert WAIT_FOR alter_progress";
+send ALTER TABLE t1 ADD INDEX(f1), ADD INDEX(f2);
+connection con1;
+SET SESSION lock_wait_timeout=1;
+SET DEBUG_SYNC="now WAIT_FOR try_insert";
+--error ER_LOCK_WAIT_TIMEOUT
+INSERT INTO t1 VALUES(1, 2);
+SET DEBUG_SYNC="now SIGNAL alter_progress";
+disconnect con1;
+connection default;
+reap;
+DROP TABLE t1;
+SET DEBUG_SYNC=reset;
+
 --echo # End of 10.4 tests
 
 SET GLOBAL innodb_purge_rseg_truncate_frequency = @save_frequency;

sql/sql_table.cc

@@ -7716,16 +7716,15 @@ static bool mysql_inplace_alter_table(THD *thd,
     lock for prepare phase under LOCK TABLES in the same way as when
     exclusive lock is required for duration of the whole statement.
   */
-  if (!ha_alter_info->mdl_exclusive_after_prepare &&
+  if (inplace_supported == HA_ALTER_INPLACE_EXCLUSIVE_LOCK ||
-      (inplace_supported == HA_ALTER_INPLACE_EXCLUSIVE_LOCK ||
+      ((inplace_supported == HA_ALTER_INPLACE_COPY_NO_LOCK ||
-       ((inplace_supported == HA_ALTER_INPLACE_COPY_NO_LOCK ||
         inplace_supported == HA_ALTER_INPLACE_COPY_LOCK ||
         inplace_supported == HA_ALTER_INPLACE_NOCOPY_NO_LOCK ||
         inplace_supported == HA_ALTER_INPLACE_NOCOPY_LOCK ||
         inplace_supported == HA_ALTER_INPLACE_INSTANT) &&
        (thd->locked_tables_mode == LTM_LOCK_TABLES ||
         thd->locked_tables_mode == LTM_PRELOCKED_UNDER_LOCK_TABLES)) ||
-      alter_info->requested_lock == Alter_info::ALTER_TABLE_LOCK_EXCLUSIVE))
+      alter_info->requested_lock == Alter_info::ALTER_TABLE_LOCK_EXCLUSIVE)
   {
     if (wait_while_table_is_used(thd, table, HA_EXTRA_FORCE_REOPEN))
       goto cleanup;
@@ -7822,7 +7821,8 @@ static bool mysql_inplace_alter_table(THD *thd,
     necessary only for prepare phase (unless we are not under LOCK TABLES) and
     user has not explicitly requested exclusive lock.
   */
-  if ((inplace_supported == HA_ALTER_INPLACE_COPY_NO_LOCK ||
+  if (!ha_alter_info->mdl_exclusive_after_prepare &&
+      (inplace_supported == HA_ALTER_INPLACE_COPY_NO_LOCK ||
        inplace_supported == HA_ALTER_INPLACE_COPY_LOCK ||
        inplace_supported == HA_ALTER_INPLACE_NOCOPY_LOCK ||
        inplace_supported == HA_ALTER_INPLACE_NOCOPY_NO_LOCK) &&