MDEV-20605 Awaken transaction can miss inserted by other transaction records...

MDEV-20605 Awaken transaction can miss inserted by other transaction records due to wrong persistent cursor restoration

sel_restore_position_for_mysql() moves forward persistent cursor
position after btr_pcur_restore_position() call if cursor relative position
is BTR_PCUR_ON and the cursor points to the record with NOT the same field
values as in a stored record(and some other not important for this case
conditions).

It was done because btr_pcur_restore_position() sets
page_cur_mode_t mode  to PAGE_CUR_LE for cursor->rel_pos ==  BTR_PCUR_ON
before opening cursor. So we are searching for the record less or equal
to stored one. And if the found record is not equal to stored one, then
it is less and we need to move cursor forward.

But there can be a situation when the stored record was purged, but the
new one with the same key but different value was inserted while
row_search_mvcc() was suspended. In this case, when the thread is
awaken, it will invoke sel_restore_position_for_mysql(), which, in turns,
invoke btr_pcur_restore_position(), which will return false because found
record don't match stored record, and
sel_restore_position_for_mysql() will move forward cursor position.

The above can lead to the case when awaken row_search_mvcc() do not see
records inserted by other transactions while it slept. The mtr test case
shows the example how it can be.

The fix is to return special value from persistent cursor restoring
function which would notify its caller that uniq fields of restored
record and stored record are the same, and in this case
sel_restore_position_for_mysql() don't move cursor forward.

Delete-marked records are correctly processed in row_search_mvcc().
Non-unique secondary indexes are "uniquified" by adding the PK, the
index->n_uniq should then be index->n_fields. So there is no need in
additional checks in the fix.

If transaction's readview can't see the changes made in secondary index
record, it requests clustered index record in row_search_mvcc() to check
its transaction id and get the correspondent record version. After this
row_search_mvcc() commits mtr to preserve clustered index latching
order, and starts mtr. Between those mtr commit and start secondary
index pages are unlatched, and purge has the ability to remove stored in
the cursor record, what causes rows duplication in result set for
non-locking reads, as cursor position is restored to the previously
visited record.

To solve this the changes are just switched off for non-locking reads,
it's quite simple solution, besides the changes don't make sense for
non-locking reads.

The more complex and effective from performance perspective solution is
to create mtr savepoint before clustered record requesting and rolling
back to that savepoint after that. See MDEV-27557.

One more solution is to have per-record transaction id for secondary
indexes. See MDEV-17598.

If any of those is implemented, just remove select_lock_type argument in
sel_restore_position_for_mysql().

mysql-test/suite/innodb/r/cursor-restore-locking.result

+CREATE TABLE t (a int PRIMARY KEY, b int NOT NULL UNIQUE) engine = InnoDB;
+connect prevent_purge,localhost,root,,;
+start transaction with consistent snapshot;
+connect con_del_1,localhost,root,,;
+INSERT INTO t VALUES (20,20);
+SET DEBUG_SYNC = 'innodb_row_search_for_mysql_exit SIGNAL first_del_row_search_mvcc_finished WAIT_FOR first_del_cont';
+DELETE FROM t WHERE b = 20;
+connect con_ins_1,localhost,root,,;
+SET DEBUG_SYNC = 'now WAIT_FOR first_del_row_search_mvcc_finished';
+SET DEBUG_SYNC = 'lock_wait_suspend_thread_enter SIGNAL first_ins_locked';
+SET DEBUG_SYNC = 'ib_after_row_insert SIGNAL first_ins_row_inserted WAIT_FOR first_ins_cont';
+INSERT INTO t VALUES(10, 20);
+connect con_del_2,localhost,root,,;
+SET DEBUG_SYNC = 'now WAIT_FOR first_ins_locked';
+SET DEBUG_SYNC = 'lock_wait_suspend_thread_enter SIGNAL second_del_locked';
+DELETE FROM t WHERE b = 20;
+connection default;
+SET DEBUG_SYNC = 'now WAIT_FOR second_del_locked';
+SET DEBUG_SYNC = 'now SIGNAL first_del_cont';
+SET DEBUG_SYNC = 'now WAIT_FOR first_ins_row_inserted';
+connection con_del_1;
+connection default;
+disconnect prevent_purge;
+InnoDB		0 transactions not purged
+SET DEBUG_SYNC = 'now SIGNAL first_ins_cont';
+connection con_del_2;
+connection con_ins_1;
+connection default;
+INSERT INTO t VALUES(30, 20);
+disconnect con_ins_1;
+disconnect con_del_1;
+disconnect con_del_2;
+connection default;
+SET DEBUG_SYNC = 'RESET';
+DROP TABLE t;

mysql-test/suite/innodb/r/cursor-restore-non-locking-read.result

+SET @saved_frequency = @@GLOBAL.innodb_purge_rseg_truncate_frequency;
+SET GLOBAL innodb_purge_rseg_truncate_frequency=1;
+CREATE TABLE t1 (pk int PRIMARY KEY, c int UNIQUE) ENGINE=InnoDB;
+INSERT INTO t1 VALUES (10,10),(20,20),(30,30);
+connect prevent_purge,localhost,root,,;
+start transaction with consistent snapshot;
+UPDATE t1 SET c=300 WHERE pk = 30;
+connection default;
+DELETE FROM t1 WHERE pk = 10;
+INSERT INTO t1 VALUES(5,10);
+SET DEBUG_SYNC = "row_search_clust_unlatched SIGNAL unlatched WAIT_FOR cont";
+SELECT pk FROM t1 FORCE INDEX (c);
+connect con1,localhost,root,,;
+SET DEBUG_SYNC = "now WAIT_FOR unlatched";
+disconnect prevent_purge;
+InnoDB		1 transactions not purged
+SET DEBUG_SYNC = 'now SIGNAL cont';
+disconnect con1;
+connection default;
+pk
+5
+20
+30
+SET DEBUG_SYNC = 'RESET';
+DROP TABLE t1;
+SET GLOBAL innodb_purge_rseg_truncate_frequency = @saved_frequency;

mysql-test/suite/innodb/t/cursor-restore-locking.test

+--source include/have_innodb.inc
+--source include/count_sessions.inc
+source include/have_debug.inc;
+source include/have_debug_sync.inc;
+
+CREATE TABLE t (a int PRIMARY KEY, b int NOT NULL UNIQUE) engine = InnoDB;
+
+--connect(prevent_purge,localhost,root,,)
+start transaction with consistent snapshot;
+
+--connect(con_del_1,localhost,root,,)
+INSERT INTO t VALUES (20,20);
+SET DEBUG_SYNC = 'innodb_row_search_for_mysql_exit SIGNAL first_del_row_search_mvcc_finished WAIT_FOR first_del_cont';
+--send DELETE FROM t WHERE b = 20
+
+--connect(con_ins_1,localhost,root,,)
+SET DEBUG_SYNC = 'now WAIT_FOR first_del_row_search_mvcc_finished';
+# It's supposed the following INSERT will be suspended just after
+# lock_wait_suspend_thread_enter syncpoint, and will be awaken
+# after the previous DELETE commits. ib_after_row_insert will be executed
+# after the INSERT is woken up. The previous DELETE will wait for
+# first_del_cont signal before commit, and this signal will be sent later.
+# So it's safe to use two signals in a row here, it's guaranted the first
+# signal will be received before the second signal is sent.
+SET DEBUG_SYNC = 'lock_wait_suspend_thread_enter SIGNAL first_ins_locked';
+SET DEBUG_SYNC = 'ib_after_row_insert SIGNAL first_ins_row_inserted WAIT_FOR first_ins_cont';
+--send INSERT INTO t VALUES(10, 20)
+
+--connect(con_del_2,localhost,root,,)
+SET DEBUG_SYNC = 'now WAIT_FOR first_ins_locked';
+SET DEBUG_SYNC = 'lock_wait_suspend_thread_enter SIGNAL second_del_locked';
+###############################################################################
+# This DELETE is locked by the previous DELETE, after that DELETE is
+# committed, it will still be locked by the next INSERT on delete-marked
+# heap_no 2 record. After that INSERT inserted the record with heap_no 3,
+# and after heap_no 2 record is purged, this DELETE will be unlocked and
+# must restore persistent cursor position at heap_no 3 record, as it has the
+# same secondary key value as former heap_no 2 record. Then it must be blocked
+# by the previous INSERT, and after the INSERT is committed, it must
+# delete the record, inserted by the previous INSERT, and the last INSERT(see
+# below) must be finished without error. But instead this DELETE restores
+# persistent cursor position to supremum, as a result, it does not delete the
+# record, inserted by the previous INSERT, and the last INSERT is finished with
+# duplicate key check error.
+###############################################################################
+--send DELETE FROM t WHERE b = 20
+
+--connection default
+SET DEBUG_SYNC = 'now WAIT_FOR second_del_locked';
+SET DEBUG_SYNC = 'now SIGNAL first_del_cont';
+SET DEBUG_SYNC = 'now WAIT_FOR first_ins_row_inserted';
+--connection con_del_1
+--reap
+
+--connection default
+--disconnect prevent_purge
+--source include/wait_all_purged.inc
+SET DEBUG_SYNC = 'now SIGNAL first_ins_cont';
+
+--connection con_del_2
+--reap
+
+--connection con_ins_1
+--reap
+
+--connection default
+###############################################################################
+# Duplicate key error is expected if the bug is not fixed.
+###############################################################################
+INSERT INTO t VALUES(30, 20);
+
+--disconnect con_ins_1
+--disconnect con_del_1
+--disconnect con_del_2
+--connection default
+
+SET DEBUG_SYNC = 'RESET';
+DROP TABLE t;
+--source include/wait_until_count_sessions.inc

mysql-test/suite/innodb/t/cursor-restore-non-locking-read.test

+--source include/have_innodb.inc
+--source include/have_debug.inc
+--source include/have_debug_sync.inc
+--source include/count_sessions.inc
+
+SET @saved_frequency = @@GLOBAL.innodb_purge_rseg_truncate_frequency;
+SET GLOBAL innodb_purge_rseg_truncate_frequency=1;
+
+CREATE TABLE t1 (pk int PRIMARY KEY, c int UNIQUE) ENGINE=InnoDB;
+
+INSERT INTO t1 VALUES (10,10),(20,20),(30,30);
+
+--connect(prevent_purge,localhost,root,,)
+start transaction with consistent snapshot;
+# We need this to update page's transaction id for secondary index.
+UPDATE t1 SET c=300 WHERE pk = 30;
+
+--connection default
+DELETE FROM t1 WHERE pk = 10;
+INSERT INTO t1 VALUES(5,10);
+SET DEBUG_SYNC = "row_search_clust_unlatched SIGNAL unlatched WAIT_FOR cont";
+# With the above sync point row_search_mvcc() will be blocked on delete-marked
+# record (10,10) in secondary index just after all page latches are released.
+# After this record is purged, row_searc_mvcc() will be unblocked, and cursor
+# will be restored to the secondary index record (10,5). As the unique field is
+# the same as in the cursor's stored record, and the bug is not fixed, there
+# value 5 will be doubled in the result set.
+--send SELECT pk FROM t1 FORCE INDEX (c)
+
+--connect(con1,localhost,root,,)
+SET DEBUG_SYNC = "now WAIT_FOR unlatched";
+--disconnect prevent_purge
+let $wait_all_purged= 1;
+--source include/wait_all_purged.inc
+SET DEBUG_SYNC = 'now SIGNAL cont';
+--disconnect con1
+
+--connection default
+--reap
+
+SET DEBUG_SYNC = 'RESET';
+DROP TABLE t1;
+SET GLOBAL innodb_purge_rseg_truncate_frequency = @saved_frequency;
+--source include/wait_until_count_sessions.inc

storage/innobase/btr/btr0cur.cc

@@ -7193,11 +7193,9 @@ struct btr_blob_log_check_t {
 			buf_block_buf_fix_dec(m_pcur->btr_cur.page_cur.block);
 		} else {
 			ut_ad(m_pcur->rel_pos == BTR_PCUR_ON);
-			bool ret = btr_pcur_restore_position(
-				BTR_MODIFY_LEAF | BTR_MODIFY_EXTERNAL,
-				m_pcur, m_mtr);
-
-			ut_a(ret);
+			ut_a(btr_pcur_restore_position(
+			      BTR_MODIFY_LEAF | BTR_MODIFY_EXTERNAL, m_pcur,
+			      m_mtr) == btr_pcur_t::SAME_ALL);
 		}
 
 		*m_block	= btr_pcur_get_block(m_pcur);

storage/innobase/fts/fts0fts.cc

@@ -3502,7 +3502,6 @@ fts_add_doc_by_id(
 				get_doc, clust_index, doc_pcur, offsets, &doc);
 
 			if (doc.found) {
-				ibool	success MY_ATTRIBUTE((unused));
 
 				btr_pcur_store_position(doc_pcur, &mtr);
 				mtr_commit(&mtr);
@@ -3556,12 +3555,10 @@ fts_add_doc_by_id(
 				mtr_start(&mtr);
 
 				if (i < num_idx - 1) {
-
-					success = btr_pcur_restore_position(
-						BTR_SEARCH_LEAF, doc_pcur,
-						&mtr);
-
-					ut_ad(success);
+					ut_d(auto status=)
+					  btr_pcur_restore_position(
+					      BTR_SEARCH_LEAF, doc_pcur, &mtr);
+					ut_ad(status == btr_pcur_t::SAME_ALL);
 				}
 			}
 

storage/innobase/ibuf/ibuf0ibuf.cc

@@ -4055,7 +4055,8 @@ ibuf_restore_pos(
 	ut_ad(mode == BTR_MODIFY_LEAF
 	      || BTR_LATCH_MODE_WITHOUT_INTENTION(mode) == BTR_MODIFY_TREE);
 
-	if (UNIV_LIKELY(btr_pcur_restore_position(mode, pcur, mtr))) {
+	if (UNIV_LIKELY(btr_pcur_restore_position(mode, pcur, mtr) ==
+	      btr_pcur_t::SAME_ALL)) {
 		return true;
 	}
 

storage/innobase/include/btr0pcur.h

@@ -253,31 +253,9 @@ btr_pcur_store_position(
 /*====================*/
 	btr_pcur_t*	cursor, /*!< in: persistent cursor */
 	mtr_t*		mtr);	/*!< in: mtr */
-/**************************************************************//**
-Restores the stored position of a persistent cursor bufferfixing the page and
-obtaining the specified latches. If the cursor position was saved when the
-(1) cursor was positioned on a user record: this function restores the position
-to the last record LESS OR EQUAL to the stored record;
-(2) cursor was positioned on a page infimum record: restores the position to
-the last record LESS than the user record which was the successor of the page
-infimum;
-(3) cursor was positioned on the page supremum: restores to the first record
-GREATER than the user record which was the predecessor of the supremum.
-(4) cursor was positioned before the first or after the last in an empty tree:
-restores to before first or after the last in the tree.
-@return TRUE if the cursor position was stored when it was on a user
-record and it can be restored on a user record whose ordering fields
-are identical to the ones of the original user record */
-ibool
-btr_pcur_restore_position_func(
-/*===========================*/
-	ulint		latch_mode,	/*!< in: BTR_SEARCH_LEAF, ... */
-	btr_pcur_t*	cursor,		/*!< in: detached persistent cursor */
-	const char*	file,		/*!< in: file name */
-	unsigned	line,		/*!< in: line where called */
-	mtr_t*		mtr);		/*!< in: mtr */
+
 #define btr_pcur_restore_position(l,cur,mtr)				\
-	btr_pcur_restore_position_func(l,cur,__FILE__,__LINE__,mtr)
+	(cur)->restore_position(l,__FILE__,__LINE__,mtr)
 /*********************************************************//**
 Gets the rel_pos field for a cursor whose position has been stored.
 @return BTR_PCUR_ON, ... */
@@ -482,6 +460,18 @@ enum pcur_pos_t {
 selects, updates, and deletes. */
 
 struct btr_pcur_t{
+	/** Return value of restore_position() */
+	enum restore_status {
+		/** cursor position on user rec and points on the record with
+		the same field values as in the stored record */
+		SAME_ALL,
+		/** cursor position is on user rec and points on the record with
+		the same unique field values as in the stored record */
+		SAME_UNIQ,
+		/** cursor position is not on user rec or points on the record
+		with not the same uniq field values as in the stored record */
+		NOT_SAME
+	};
 	/** a B-tree cursor */
 	btr_cur_t	btr_cur;
 	/** see TODO note below!
@@ -539,6 +529,31 @@ struct btr_pcur_t{
 
 	/** Return the index of this persistent cursor */
 	dict_index_t*	index() const { return(btr_cur.index); }
+	/** Restores the stored position of a persistent cursor bufferfixing
+	the page and obtaining the specified latches. If the cursor position
+	was saved when the
+	(1) cursor was positioned on a user record: this function restores the
+	position to the last record LESS OR EQUAL to the stored record;
+	(2) cursor was positioned on a page infimum record: restores the
+	position to the last record LESS than the user record which was the
+	successor of the page infimum;
+	(3) cursor was positioned on the page supremum: restores to the first
+	record GREATER than the user record which was the predecessor of the
+	supremum.
+	(4) cursor was positioned before the first or after the last in an
+	empty tree: restores to before first or after the last in the tree.
+	@param latch_mode BTR_SEARCH_LEAF, ...
+	@param file file name
+	@param line line where called
+	@param mtr mtr
+	@return btr_pcur_t::SAME_ALL cursor position on user rec and points on
+	the record with the same field values as in the stored record,
+	btr_pcur_t::SAME_UNIQ cursor position is on user rec and points on the
+	record with the same unique field values as in the stored record,
+	btr_pcur_t::NOT_SAME cursor position is not on user rec or points on
+	the record with not the samebuniq field values as in the stored */
+	restore_status restore_position(ulint latch_mode, const char *file,
+	                                unsigned line, mtr_t *mtr);
 };
 
 #include "btr0pcur.inl"

storage/innobase/row/row0purge.cc

@@ -72,7 +72,9 @@ row_purge_reposition_pcur(
 	if (node->found_clust) {
 		ut_ad(node->validate_pcur());
 
-		node->found_clust = btr_pcur_restore_position(mode, &node->pcur, mtr);
+		node->found_clust =
+		  btr_pcur_restore_position(mode, &node->pcur, mtr) ==
+		    btr_pcur_t::SAME_ALL;
 
 	} else {
 		node->found_clust = row_search_on_row_ref(

storage/innobase/row/row0sel.cc

@@ -1422,8 +1422,9 @@ row_sel_restore_pcur_pos(
 
 	relative_position = btr_pcur_get_rel_pos(&(plan->pcur));
 
-	equal_position = btr_pcur_restore_position(BTR_SEARCH_LEAF,
-						   &(plan->pcur), mtr);
+	equal_position =
+	  btr_pcur_restore_position(BTR_SEARCH_LEAF, &plan->pcur, mtr) ==
+	  btr_pcur_t::SAME_ALL;
 
 	/* If the cursor is traveling upwards, and relative_position is
 
@@ -3579,36 +3580,28 @@ Row_sel_get_clust_rec_for_mysql::operator()(
 	return(err);
 }
 
-/********************************************************************//**
-Restores cursor position after it has been stored. We have to take into
+/** Restores cursor position after it has been stored. We have to take into
 account that the record cursor was positioned on may have been deleted.
 Then we may have to move the cursor one step up or down.
+@param[out] same_user_rec true if we were able to restore the cursor on a user
+record with the same ordering prefix in in the B-tree index
+@param[in] latch_mode latch mode wished in restoration
+@param[in] pcur cursor whose position has been stored
+@param[in] moves_up true if the cursor moves up in the index
+@param[in] mtr mtr; CAUTION: may commit mtr temporarily!
+@param[in] select_lock_type select lock type
 @return true if we may need to process the record the cursor is now
 positioned on (i.e. we should not go to the next record yet) */
-static
-bool
-sel_restore_position_for_mysql(
-/*===========================*/
-	ibool*		same_user_rec,	/*!< out: TRUE if we were able to restore
-					the cursor on a user record with the
-					same ordering prefix in in the
-					B-tree index */
-	ulint		latch_mode,	/*!< in: latch mode wished in
-					restoration */
-	btr_pcur_t*	pcur,		/*!< in: cursor whose position
-					has been stored */
-	ibool		moves_up,	/*!< in: TRUE if the cursor moves up
-					in the index */
-	mtr_t*		mtr)		/*!< in: mtr; CAUTION: may commit
-					mtr temporarily! */
+static bool sel_restore_position_for_mysql(bool *same_user_rec,
+                                           ulint latch_mode, btr_pcur_t *pcur,
+                                           bool moves_up, mtr_t *mtr,
+                                           lock_mode select_lock_type)
 {
-	ibool		success;
-
-	success = btr_pcur_restore_position(latch_mode, pcur, mtr);
+	auto status = btr_pcur_restore_position(latch_mode, pcur, mtr);
 
-	*same_user_rec = success;
+	*same_user_rec = status == btr_pcur_t::SAME_ALL;
 
-	ut_ad(!success || pcur->rel_pos == BTR_PCUR_ON);
+	ut_ad(!*same_user_rec || pcur->rel_pos == BTR_PCUR_ON);
 #ifdef UNIV_DEBUG
 	if (pcur->pos_state == BTR_PCUR_IS_POSITIONED_OPTIMISTIC) {
 		ut_ad(pcur->rel_pos == BTR_PCUR_BEFORE
@@ -3624,7 +3617,10 @@ sel_restore_position_for_mysql(
 
 	switch (pcur->rel_pos) {
 	case BTR_PCUR_ON:
-		if (!success && moves_up) {
+		if (!*same_user_rec && moves_up) {
+			if (status == btr_pcur_t::SAME_UNIQ
+			    && select_lock_type != LOCK_NONE)
+			  return true;
 next:
 			if (btr_pcur_move_to_next(pcur, mtr)
 			    && rec_is_metadata(btr_pcur_get_rec(pcur),
@@ -3634,7 +3630,7 @@ sel_restore_position_for_mysql(
 
 			return true;
 		}
-		return(!success);
+		return(!*same_user_rec);
 	case BTR_PCUR_AFTER_LAST_IN_TREE:
 	case BTR_PCUR_BEFORE_FIRST_IN_TREE:
 		return true;
@@ -4308,12 +4304,12 @@ row_search_mvcc(
 	Row_sel_get_clust_rec_for_mysql row_sel_get_clust_rec_for_mysql;
 	ibool		unique_search			= FALSE;
 	ibool		mtr_has_extra_clust_latch	= FALSE;
-	ibool		moves_up			= FALSE;
+	bool		moves_up			= false;
 	/* if the returned record was locked and we did a semi-consistent
 	read (fetch the newest committed version), then this is set to
 	TRUE */
 	ulint		next_offs;
-	ibool		same_user_rec;
+	bool		same_user_rec;
 	ibool		table_lock_waited		= FALSE;
 	byte*		next_buf			= 0;
 	bool		spatial_search			= false;
@@ -4622,10 +4618,10 @@ row_search_mvcc(
 	if (UNIV_UNLIKELY(direction == 0)) {
 		if (mode == PAGE_CUR_GE || mode == PAGE_CUR_G
 		    || mode >= PAGE_CUR_CONTAIN) {
-			moves_up = TRUE;
+			moves_up = true;
 		}
 	} else if (direction == ROW_SEL_NEXT) {
-		moves_up = TRUE;
+		moves_up = true;
 	}
 
 	thr = que_fork_get_first_thr(prebuilt->sel_graph);
@@ -4677,7 +4673,7 @@ row_search_mvcc(
 
 		bool	need_to_process = sel_restore_position_for_mysql(
 			&same_user_rec, BTR_SEARCH_LEAF,
-			pcur, moves_up, &mtr);
+			pcur, moves_up, &mtr, prebuilt->select_lock_type);
 
 		if (UNIV_UNLIKELY(need_to_process)) {
 			if (UNIV_UNLIKELY(prebuilt->row_read_type
@@ -4884,7 +4880,7 @@ row_search_mvcc(
 	if (UNIV_UNLIKELY(next_offs >= srv_page_size - PAGE_DIR)) {
 
 wrong_offs:
-		if (srv_force_recovery == 0 || moves_up == FALSE) {
+		if (srv_force_recovery == 0 || moves_up == false) {
 			ib::error() << "Rec address "
 				<< static_cast<const void*>(rec)
 				<< ", buf block fix count "
@@ -5658,12 +5654,14 @@ row_search_mvcc(
 		btr_pcur_store_position(pcur, &mtr);
 		mtr.commit();
 		mtr_has_extra_clust_latch = FALSE;
-
+		DEBUG_SYNC_C("row_search_clust_unlatched");
 		mtr.start();
 
 		if (sel_restore_position_for_mysql(&same_user_rec,
 						   BTR_SEARCH_LEAF,
-						   pcur, moves_up, &mtr)) {
+						   pcur, moves_up, &mtr,
+						   prebuilt->select_lock_type)
+		    ) {
 			goto rec_loop;
 		}
 	}
@@ -5754,7 +5752,7 @@ row_search_mvcc(
 		if (!dict_index_is_spatial(index)) {
 			sel_restore_position_for_mysql(
 				&same_user_rec, BTR_SEARCH_LEAF, pcur,
-				moves_up, &mtr);
+				moves_up, &mtr, prebuilt->select_lock_type);
 		}
 
 		if (trx->isolation_level <= TRX_ISO_READ_COMMITTED

storage/innobase/row/row0uins.cc

@@ -65,7 +65,6 @@ row_undo_ins_remove_clust_rec(
 /*==========================*/
 	undo_node_t*	node)	/*!< in: undo node */
 {
-	ibool		success;
 	dberr_t		err;
 	ulint		n_tries	= 0;
 	mtr_t		mtr;
@@ -101,13 +100,12 @@ row_undo_ins_remove_clust_rec(
 	We must log the removal, so that the row will be correctly
 	purged. However, we can log the removal out of sync with the
 	B-tree modification. */
-
-	success = btr_pcur_restore_position(
-		online
-		? BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED
-		: (node->rec_type == TRX_UNDO_INSERT_METADATA)
-		? BTR_MODIFY_TREE : BTR_MODIFY_LEAF, &node->pcur, &mtr);
-	ut_a(success);
+	ut_a(btr_pcur_restore_position(
+	      online ? BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED
+	      : (node->rec_type == TRX_UNDO_INSERT_METADATA)
+		  ? BTR_MODIFY_TREE
+		  : BTR_MODIFY_LEAF,
+	      &node->pcur, &mtr) == btr_pcur_t::SAME_ALL);
 
 	rec_t* rec = btr_pcur_get_rec(&node->pcur);
 
@@ -137,9 +135,8 @@ row_undo_ins_remove_clust_rec(
 			mtr.commit();
 
 			mtr.start();
-			success = btr_pcur_restore_position(
-				BTR_MODIFY_LEAF, &node->pcur, &mtr);
-			ut_a(success);
+			ut_a(btr_pcur_restore_position(BTR_MODIFY_LEAF,
+			      &node->pcur, &mtr)== btr_pcur_t::SAME_ALL);
 			break;
 		case DICT_COLUMNS_ID:
 			/* This is rolling back an INSERT into SYS_COLUMNS.
@@ -180,11 +177,8 @@ row_undo_ins_remove_clust_rec(
 	} else {
 		index->set_modified(mtr);
 	}
-
-	success = btr_pcur_restore_position(
-			BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
-			&node->pcur, &mtr);
-	ut_a(success);
+	ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
+	      &node->pcur, &mtr) == btr_pcur_t::SAME_ALL);
 
 	btr_cur_pessimistic_delete(&err, FALSE, &node->pcur.btr_cur, 0, true,
 				   &mtr);

storage/innobase/row/row0umod.cc

@@ -95,19 +95,14 @@ row_undo_mod_clust_low(
 	btr_pcur_t*	pcur;
 	btr_cur_t*	btr_cur;
 	dberr_t		err;
-#ifdef UNIV_DEBUG
-	ibool		success;
-#endif /* UNIV_DEBUG */
 
 	pcur = &node->pcur;
 	btr_cur = btr_pcur_get_btr_cur(pcur);
 
-#ifdef UNIV_DEBUG
-	success =
-#endif /* UNIV_DEBUG */
+	ut_d(auto pcur_restore_result =)
 	btr_pcur_restore_position(mode, pcur, mtr);
 
-	ut_ad(success);
+	ut_ad(pcur_restore_result == btr_pcur_t::SAME_ALL);
 	ut_ad(rec_get_trx_id(btr_cur_get_rec(btr_cur),
 			     btr_cur_get_index(btr_cur))
 	      == thr_get_trx(thr)->id
@@ -363,7 +358,8 @@ row_undo_mod_clust(
 		ut_ad(node->new_trx_id);
 
 		mtr.start();
-		if (!btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, &mtr)) {
+		if (btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, &mtr) !=
+		    btr_pcur_t::SAME_ALL) {
 			goto mtr_commit_exit;
 		}
 
@@ -385,9 +381,9 @@ row_undo_mod_clust(
 		btr_pcur_commit_specify_mtr(pcur, &mtr);
 
 		mtr.start();
-		if (!btr_pcur_restore_position(
+		if (btr_pcur_restore_position(
 			    BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
-			    pcur, &mtr)) {
+			    pcur, &mtr) != btr_pcur_t::SAME_ALL) {
 			goto mtr_commit_exit;
 		}
 
@@ -419,7 +415,8 @@ row_undo_mod_clust(
 		longer accessible by any active read view. */
 
 		mtr.start();
-		if (!btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, &mtr)) {
+		if (btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, &mtr)
+		    != btr_pcur_t::SAME_ALL) {
 			goto mtr_commit_exit;
 		}
 		rec_t* rec = btr_pcur_get_rec(pcur);
@@ -521,7 +518,6 @@ row_undo_mod_del_mark_or_remove_sec_low(
 {
 	btr_pcur_t		pcur;
 	btr_cur_t*		btr_cur;
-	ibool			success;
 	dberr_t			err	= DB_SUCCESS;
 	mtr_t			mtr;
 	mtr_t			mtr_vers;
@@ -593,9 +589,8 @@ row_undo_mod_del_mark_or_remove_sec_low(
 
 	mtr_vers.start();
 
-	success = btr_pcur_restore_position(BTR_SEARCH_LEAF, &(node->pcur),
-					    &mtr_vers);
-	ut_a(success);
+	ut_a(btr_pcur_restore_position(BTR_SEARCH_LEAF, &node->pcur, &mtr_vers)
+	    == btr_pcur_t::SAME_ALL);
 
 	/* For temporary table, we can skip to check older version of
 	clustered index entry, because there is no MVCC or purge. */

storage/innobase/row/row0upd.cc

@@ -2617,7 +2617,8 @@ row_upd_clust_rec(
 	the same transaction do not modify the record in the meantime.
 	Therefore we can assert that the restoration of the cursor succeeds. */
 
-	ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));
+	ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr) ==
+	    btr_pcur_t::SAME_ALL);
 
 	ut_ad(!rec_get_deleted_flag(btr_pcur_get_rec(pcur),
 				    dict_table_is_comp(index->table)));
@@ -2761,7 +2762,6 @@ row_upd_clust_step(
 {
 	dict_index_t*	index;
 	btr_pcur_t*	pcur;
-	ibool		success;
 	dberr_t		err;
 	mtr_t		mtr;
 	rec_t*		rec;
@@ -2821,9 +2821,8 @@ row_upd_clust_step(
 		mode = BTR_MODIFY_LEAF;
 	}
 
-	success = btr_pcur_restore_position(mode, pcur, &mtr);
-
-	if (!success) {
+	if (btr_pcur_restore_position(mode, pcur, &mtr) !=
+	    btr_pcur_t::SAME_ALL) {
 		err = DB_RECORD_NOT_FOUND;
 		goto exit_func;
 	}
@@ -2844,9 +2843,8 @@ row_upd_clust_step(
 		mtr.start();
 		index->set_modified(mtr);
 
-		success = btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur,
-						    &mtr);
-		if (!success) {
+		if (btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, &mtr) !=
+		    btr_pcur_t::SAME_ALL) {
 			err = DB_ERROR;
 
 			mtr.commit();