MariaDB · iMineLink · Mar 25, 2026 · Apr 7, 2026 · Apr 7, 2026
diff --git a/mysql-test/suite/innodb/r/index_lock_upgrade.result b/mysql-test/suite/innodb/r/index_lock_upgrade.result
@@ -0,0 +1,90 @@
+#
+# MDEV-38814: High rate of index_lock_upgrades due to
+# btr_cur_need_opposite_intention() mostly returning true
+#
+# Test with simple table structure:
+# Single INT primary key
+# DATETIME column dt that starts NULL and gets updated to a
+# distinct timestamp per row
+#
+# The NULL-to-DATETIME growth makes btr_cur_optimistic_update()
+# fall through to the pessimistic path (BTR_MODIFY_TREE).
+CREATE TABLE t1 (
+id INT NOT NULL,
+dt DATETIME NULL,
+PRIMARY KEY (id)
+) ENGINE=InnoDB;
+INSERT INTO t1 (id, dt) SELECT seq, NULL FROM seq_1_to_1000;
+# Rows inserted with NULL DATETIME
+SELECT COUNT(*) AS rows_inserted FROM t1;
+rows_inserted
+1000
+SELECT COUNT(DISTINCT dt) AS distinct_timestamps FROM t1;
+distinct_timestamps
+0
+# Index lock upgrades during inserts
+SELECT @lu01 AS index_lock_upgrades_insert;
+index_lock_upgrades_insert
+5
+SELECT @pu01 AS pessimistic_update_calls_insert;
+pessimistic_update_calls_insert
+0
+SELECT @ou01 AS optim_err_underflows_insert;
+optim_err_underflows_insert
+0
+SELECT @oo01 AS optim_err_overflows_insert;
+optim_err_overflows_insert
+0
+SELECT @ps01 AS index_page_splits_insert;
+index_page_splits_insert
+6
+SELECT @ma01 AS index_page_merge_attempts_insert;
+index_page_merge_attempts_insert
+0
+SELECT @ms01 AS index_page_merge_successful_insert;
+index_page_merge_successful_insert
+0
+SELECT @rs01 AS index_page_reorg_successful_insert;
+index_page_reorg_successful_insert
+0
+SELECT @pd01 AS index_page_discards_insert;
+index_page_discards_insert
+0
+# Now update all rows (batch update)
+# This changes DATETIME from NULL to a distinct timestamp per row,
+# triggering the pessimistic path due to record size change.
+UPDATE t1 JOIN seq_1_to_1000 s ON t1.id = s.seq
+SET t1.dt= '2025-01-01' + INTERVAL s.seq SECOND;
+# Rows updated
+SELECT COUNT(DISTINCT dt) AS distinct_timestamps FROM t1;
+distinct_timestamps
+1000
+# Index lock upgrades during updates
+SELECT @lu12 AS index_lock_upgrades_update;
+index_lock_upgrades_update
+6
+SELECT @pu12 AS pessimistic_update_calls_update;
+pessimistic_update_calls_update
+6
+SELECT @ou12 AS optim_err_underflows_update;
+optim_err_underflows_update
+0
+SELECT @oo12 AS optim_err_overflows_update;
+optim_err_overflows_update
+6
+SELECT @ps12 AS index_page_splits_update;
+index_page_splits_update
+6
+SELECT @ma12 AS index_page_merge_attempts_update;
+index_page_merge_attempts_update
+0
+SELECT @ms12 AS index_page_merge_successful_update;
+index_page_merge_successful_update
+0
+SELECT @rs12 AS index_page_reorg_successful_update;
+index_page_reorg_successful_update
+40
+SELECT @pd12 AS index_page_discards_update;
+index_page_discards_update
+0
+DROP TABLE t1;
diff --git a/mysql-test/suite/innodb/r/innodb_status_variables.result b/mysql-test/suite/innodb/r/innodb_status_variables.result
@@ -3,7 +3,11 @@ WHERE variable_name LIKE 'INNODB_%'
 AND variable_name NOT IN
 ('INNODB_ADAPTIVE_HASH_HASH_SEARCHES','INNODB_ADAPTIVE_HASH_NON_HASH_SEARCHES',
 'INNODB_MEM_ADAPTIVE_HASH',
-'INNODB_BUFFERED_AIO_SUBMITTED','INNODB_BUFFER_POOL_PAGES_LATCHED');
+'INNODB_BUFFERED_AIO_SUBMITTED','INNODB_BUFFER_POOL_PAGES_LATCHED',
+'INNODB_BTR_CUR_N_INDEX_LOCK_UPGRADES',
+'INNODB_BTR_CUR_PESSIMISTIC_UPDATE_CALLS',
+'INNODB_BTR_CUR_PESSIMISTIC_UPDATE_OPTIM_ERR_UNDERFLOWS',
+'INNODB_BTR_CUR_PESSIMISTIC_UPDATE_OPTIM_ERR_OVERFLOWS');
 variable_name
 INNODB_ASYNC_READS_PENDING
 INNODB_ASYNC_READS_TASKS_RUNNING

diff --git a/mysql-test/suite/innodb/t/index_lock_upgrade.test b/mysql-test/suite/innodb/t/index_lock_upgrade.test
@@ -0,0 +1,155 @@
+--source include/have_innodb.inc
+--source include/have_innodb_4k.inc
+--source include/have_sequence.inc
+--source include/have_debug.inc
+
+--echo #
+--echo # MDEV-38814: High rate of index_lock_upgrades due to
+--echo # btr_cur_need_opposite_intention() mostly returning true
+--echo #
+
+--echo # Test with simple table structure:
+--echo # Single INT primary key
+--echo # DATETIME column dt that starts NULL and gets updated to a
+--echo # distinct timestamp per row
+--echo #
+--echo # The NULL-to-DATETIME growth makes btr_cur_optimistic_update()
+--echo # fall through to the pessimistic path (BTR_MODIFY_TREE).
+CREATE TABLE t1 (
+  id INT NOT NULL,
+  dt DATETIME NULL,
+  PRIMARY KEY (id)
+) ENGINE=InnoDB;
+
+--disable_query_log
+SET GLOBAL innodb_monitor_enable= 'module_index';
+--enable_query_log
+
+let $lu0= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_n_index_lock_upgrades'`;
+let $pu0= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_calls'`;
+let $ou0= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_optim_err_underflows'`;
+let $oo0= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_optim_err_overflows'`;
+let $ps0= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_splits'`;
+let $ma0= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_merge_attempts'`;
+let $ms0= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_merge_successful'`;
+let $rs0= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_reorg_successful'`;
+let $pd0= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_discards'`;
+
+INSERT INTO t1 (id, dt) SELECT seq, NULL FROM seq_1_to_1000;
+
+--echo # Rows inserted with NULL DATETIME
+SELECT COUNT(*) AS rows_inserted FROM t1;
+SELECT COUNT(DISTINCT dt) AS distinct_timestamps FROM t1;
+
+let $lu1= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_n_index_lock_upgrades'`;
+let $pu1= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_calls'`;
+let $ou1= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_optim_err_underflows'`;
+let $oo1= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_optim_err_overflows'`;
+let $ps1= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_splits'`;
+let $ma1= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_merge_attempts'`;
+let $ms1= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_merge_successful'`;
+let $rs1= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_reorg_successful'`;
+let $pd1= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_discards'`;
+
+--disable_query_log
+eval SET @lu01= $lu1 - $lu0;
+eval SET @pu01= $pu1 - $pu0;
+eval SET @ou01= $ou1 - $ou0;
+eval SET @oo01= $oo1 - $oo0;
+eval SET @ps01= $ps1 - $ps0;
+eval SET @ma01= $ma1 - $ma0;
+eval SET @ms01= $ms1 - $ms0;
+eval SET @rs01= $rs1 - $rs0;
+eval SET @pd01= $pd1 - $pd0;
+--enable_query_log
+
+--echo # Index lock upgrades during inserts
+SELECT @lu01 AS index_lock_upgrades_insert;
+SELECT @pu01 AS pessimistic_update_calls_insert;
+SELECT @ou01 AS optim_err_underflows_insert;
+SELECT @oo01 AS optim_err_overflows_insert;
+SELECT @ps01 AS index_page_splits_insert;
+SELECT @ma01 AS index_page_merge_attempts_insert;
+SELECT @ms01 AS index_page_merge_successful_insert;
+SELECT @rs01 AS index_page_reorg_successful_insert;
+SELECT @pd01 AS index_page_discards_insert;
+
+--echo # Now update all rows (batch update)
+--echo # This changes DATETIME from NULL to a distinct timestamp per row,
+--echo # triggering the pessimistic path due to record size change.
+
+UPDATE t1 JOIN seq_1_to_1000 s ON t1.id = s.seq
+SET t1.dt= '2025-01-01' + INTERVAL s.seq SECOND;
+
+--echo # Rows updated
+SELECT COUNT(DISTINCT dt) AS distinct_timestamps FROM t1;
+
+let $lu2= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_n_index_lock_upgrades'`;
+let $pu2= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_calls'`;
+let $ou2= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_optim_err_underflows'`;
+let $oo2= `SELECT VARIABLE_VALUE FROM INFORMATION_SCHEMA.GLOBAL_STATUS
+               WHERE VARIABLE_NAME = 'Innodb_btr_cur_pessimistic_update_optim_err_overflows'`;
+let $ps2= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_splits'`;
+let $ma2= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_merge_attempts'`;
+let $ms2= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_merge_successful'`;
+let $rs2= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_reorg_successful'`;
+let $pd2= `SELECT COUNT_RESET FROM INFORMATION_SCHEMA.INNODB_METRICS
+               WHERE NAME = 'index_page_discards'`;
+
+--disable_query_log
+eval SET @lu12= $lu2 - $lu1;
+eval SET @pu12= $pu2 - $pu1;
+eval SET @ou12= $ou2 - $ou1;
+eval SET @oo12= $oo2 - $oo1;
+eval SET @ps12= $ps2 - $ps1;
+eval SET @ma12= $ma2 - $ma1;
+eval SET @ms12= $ms2 - $ms1;
+eval SET @rs12= $rs2 - $rs1;
+eval SET @pd12= $pd2 - $pd1;
+--enable_query_log
+
+--echo # Index lock upgrades during updates
+SELECT @lu12 AS index_lock_upgrades_update;
+SELECT @pu12 AS pessimistic_update_calls_update;
+SELECT @ou12 AS optim_err_underflows_update;
+SELECT @oo12 AS optim_err_overflows_update;
+SELECT @ps12 AS index_page_splits_update;
+SELECT @ma12 AS index_page_merge_attempts_update;
+SELECT @ms12 AS index_page_merge_successful_update;
+SELECT @rs12 AS index_page_reorg_successful_update;
+SELECT @pd12 AS index_page_discards_update;
+
+--disable_query_log
+SET GLOBAL innodb_monitor_disable= 'module_index';
+--disable_warnings
+SET GLOBAL innodb_monitor_disable= default;
+SET GLOBAL innodb_monitor_enable= default;
+--enable_warnings
+--enable_query_log
+
+DROP TABLE t1;
diff --git a/mysql-test/suite/innodb/t/innodb_status_variables.test b/mysql-test/suite/innodb/t/innodb_status_variables.test
@@ -4,4 +4,8 @@ WHERE variable_name LIKE 'INNODB_%'
 AND variable_name NOT IN
 ('INNODB_ADAPTIVE_HASH_HASH_SEARCHES','INNODB_ADAPTIVE_HASH_NON_HASH_SEARCHES',
  'INNODB_MEM_ADAPTIVE_HASH',
- 'INNODB_BUFFERED_AIO_SUBMITTED','INNODB_BUFFER_POOL_PAGES_LATCHED');
+ 'INNODB_BUFFERED_AIO_SUBMITTED','INNODB_BUFFER_POOL_PAGES_LATCHED',
+ 'INNODB_BTR_CUR_N_INDEX_LOCK_UPGRADES',
+ 'INNODB_BTR_CUR_PESSIMISTIC_UPDATE_CALLS',
+ 'INNODB_BTR_CUR_PESSIMISTIC_UPDATE_OPTIM_ERR_UNDERFLOWS',
+ 'INNODB_BTR_CUR_PESSIMISTIC_UPDATE_OPTIM_ERR_OVERFLOWS');
diff --git a/storage/innobase/btr/btr0cur.cc b/storage/innobase/btr/btr0cur.cc
@@ -105,6 +105,14 @@ ulint	btr_cur_n_sea_old;
 #ifdef UNIV_DEBUG
 /* Flag to limit optimistic insert records */
 uint	btr_cur_limit_optimistic_insert_debug;
+/** Number of times index lock was upgraded from SX to X */
+Atomic_counter<size_t> btr_cur_n_index_lock_upgrades;
+/** Number of times btr_cur_pessimistic_update() was called */
+Atomic_counter<size_t> btr_cur_pessimistic_update_calls;
+/** Number of times DB_UNDERFLOW was returned as optimistic update error in btr_cur_pessimistic_update() */
+Atomic_counter<size_t> btr_cur_pessimistic_update_optim_err_underflows;
+/** Number of times DB_OVERFLOW was returned as optimistic update error in btr_cur_pessimistic_update() */
+Atomic_counter<size_t> btr_cur_pessimistic_update_optim_err_overflows;
 #endif /* UNIV_DEBUG */
 
 /** In the optimistic insert, if the insert does not fit, but this much space
@@ -1599,6 +1607,7 @@ ATTRIBUTE_COLD void mtr_t::index_lock_upgrade()
   index_lock *lock= static_cast<index_lock*>(slot.object);
   lock->u_x_upgrade(SRW_LOCK_CALL);
   slot.type= MTR_MEMO_X_LOCK;
+  ut_d(++btr_cur_n_index_lock_upgrades);
 }
 
 /** Mark a non-leaf page "least recently used", but avoid invoking
@@ -3602,10 +3611,10 @@ btr_cur_optimistic_update(
 		goto func_exit;
 	}
 
-	if (UNIV_UNLIKELY(page_get_data_size(page)
+	if (UNIV_UNLIKELY((new_rec_size < old_rec_size) && page_get_data_size(page)
 			  - old_rec_size + new_rec_size
 			  < BTR_CUR_PAGE_COMPRESS_LIMIT(index))) {
-		/* The page would become too empty */
+		/* The page would become too empty due to record shrinkage */
 		err = DB_UNDERFLOW;
 		goto func_exit;
 	}
@@ -3797,6 +3806,8 @@ btr_cur_pessimistic_update(
 	block = btr_cur_get_block(cursor);
 	index = cursor->index();
 
+	ut_d(++btr_cur_pessimistic_update_calls);
+
 	ut_ad(mtr->memo_contains_flagged(&index->lock, MTR_MEMO_X_LOCK |
 					 MTR_MEMO_SX_LOCK));
 	ut_ad(mtr->memo_contains_flagged(block, MTR_MEMO_PAGE_X_FIX));
@@ -3822,6 +3833,9 @@ btr_cur_pessimistic_update(
 		cursor, offsets, offsets_heap, update,
 		cmpl_info, thr, trx_id, mtr);
 
+	ut_d(btr_cur_pessimistic_update_optim_err_underflows += (err == DB_UNDERFLOW));
+	ut_d(btr_cur_pessimistic_update_optim_err_overflows += (err == DB_OVERFLOW));
+
 	switch (err) {
 	case DB_ZIP_OVERFLOW:
 	case DB_UNDERFLOW:
@@ -3992,8 +4006,17 @@ btr_cur_pessimistic_update(
 		goto return_after_reservations;
 	}
 
-	rec = btr_cur_insert_if_possible(cursor, new_entry,
-					 offsets, offsets_heap, n_ext, mtr);
+	if (optim_err == DB_OVERFLOW
+	    && !buf_block_get_page_zip(block)
+	    && page_get_max_insert_size_after_reorganize(
+	        block->page.frame, 1) < BTR_CUR_PAGE_REORGANIZE_LIMIT) {
+		/* The page is too full: force a split instead of
+		reinserting on the same page. */
+		rec = NULL;
+	} else {
+		rec = btr_cur_insert_if_possible(cursor, new_entry,
+						 offsets, offsets_heap, n_ext, mtr);
+	}
 
 	if (rec) {
 		page_cursor->rec = rec;

diff --git a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc
@@ -1071,6 +1071,17 @@ static SHOW_VAR innodb_status_variables[]= {
   /* InnoDB bulk operations */
   {"bulk_operations", &export_vars.innodb_bulk_operations, SHOW_SIZE_T},
 
+#ifdef UNIV_DEBUG
+  {"btr_cur_n_index_lock_upgrades",
+   &btr_cur_n_index_lock_upgrades, SHOW_SIZE_T},
+  {"btr_cur_pessimistic_update_calls",
+   &btr_cur_pessimistic_update_calls, SHOW_SIZE_T},
+  {"btr_cur_pessimistic_update_optim_err_underflows",
+   &btr_cur_pessimistic_update_optim_err_underflows, SHOW_SIZE_T},
+  {"btr_cur_pessimistic_update_optim_err_overflows",
+   &btr_cur_pessimistic_update_optim_err_overflows, SHOW_SIZE_T},
+#endif /* UNIV_DEBUG */
+
   {NullS, NullS, SHOW_LONG}
 };
 

diff --git a/storage/innobase/include/btr0cur.h b/storage/innobase/include/btr0cur.h
@@ -833,6 +833,14 @@ extern ulint	btr_cur_n_sea_old;
 #ifdef UNIV_DEBUG
 /* Flag to limit optimistic insert records */
 extern uint	btr_cur_limit_optimistic_insert_debug;
+/** Number of times index lock was upgraded from SX to X */
+extern Atomic_counter<size_t> btr_cur_n_index_lock_upgrades;
+/** Number of times btr_cur_pessimistic_update() was called */
+extern Atomic_counter<size_t> btr_cur_pessimistic_update_calls;
+/** Number of times DB_UNDERFLOW was returned as optimistic update error in btr_cur_pessimistic_update() */
+extern Atomic_counter<size_t> btr_cur_pessimistic_update_optim_err_underflows;
+/** Number of times DB_OVERFLOW was returned as optimistic update error in btr_cur_pessimistic_update() */
+extern Atomic_counter<size_t> btr_cur_pessimistic_update_optim_err_overflows;
 #endif /* UNIV_DEBUG */
 
 #include "btr0cur.inl"