MDEV-39440: Failed to match the stats from replay context with the optimizer stats

bsrikanth-mariadb · bsrikanth-mariadb · commit 17dece4a0204 · 2026-05-22T13:56:47.000+05:30
The problem is that handler-&gt;multi_range_read_info_const() call in
check_quick_select() returned rows equal to HA_POS_ERROR, both during
context capture and replay. However, we never stored the ranges info
into the context when rows=HA_POS_ERROR. However, during replay, we try
to infuse stats for the given range, and since no match was found in the
context, we produced a warning.

Solution is to store range_info for all the ranges even when they get
HA_POS_ERROR number of rows.
diff --git a/mysql-test/main/opt_context_replay_basic.result b/mysql-test/main/opt_context_replay_basic.result
@@ -315,4 +315,26 @@ id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	SIMPLE	t1	system	NULL	NULL	NULL	NULL	1	
 set optimizer_replay_context='';
 drop table t1;
+#
+# MDEV-39440: Failed to match the stats from replay context with the optimizer stats
+#
+create table t1 (btn char(10) not null, key using HASH (btn)) engine=heap;
+insert into t1 values ("a"),("b"),("c"),("d");
+alter table t1 add column new_col char(1) not null, add key using HASH (btn,new_col), drop key btn;
+update t1 set new_col=left(btn,1);
+set optimizer_record_context=1;
+explain select * from t1 where btn="a";
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	btn	NULL	NULL	NULL	4	Using where
+select context into dumpfile "../../tmp/dump1.sql"
+from information_schema.optimizer_context;
+set optimizer_record_context=0;
+drop table t1;
+set optimizer_replay_context='opt_context';
+# Same query as above, must have same explain:
+explain select * from t1 where btn="a";
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	btn	NULL	NULL	NULL	4	Using where
+set optimizer_replay_context='';
+drop table t1;
 drop database db1;
diff --git a/mysql-test/main/opt_context_replay_basic.test b/mysql-test/main/opt_context_replay_basic.test
@@ -173,4 +173,31 @@ set optimizer_replay_context='';
 --remove_file "$MYSQLTEST_VARDIR/tmp/dump1.sql"
 drop table t1;
 
+--echo #
+--echo # MDEV-39440: Failed to match the stats from replay context with the optimizer stats
+--echo #
+create table t1 (btn char(10) not null, key using HASH (btn)) engine=heap;
+insert into t1 values ("a"),("b"),("c"),("d");
+alter table t1 add column new_col char(1) not null, add key using HASH (btn,new_col), drop key btn;
+update t1 set new_col=left(btn,1);
+
+set optimizer_record_context=1;
+explain select * from t1 where btn="a";
+select context into dumpfile "../../tmp/dump1.sql"
+from information_schema.optimizer_context;
+set optimizer_record_context=0;
+drop table t1;
+--disable_query_log
+--disable_result_log
+--source "$MYSQLTEST_VARDIR/tmp/dump1.sql"
+--enable_query_log
+--enable_result_log
+set optimizer_replay_context='opt_context';
+--echo # Same query as above, must have same explain:
+explain select * from t1 where btn="a";
+
+set optimizer_replay_context='';
+--remove_file "$MYSQLTEST_VARDIR/tmp/dump1.sql"
+drop table t1;
+
 drop database db1;
diff --git a/sql/opt_context_store_replay.cc b/sql/opt_context_store_replay.cc
@@ -775,13 +775,9 @@ Optimizer_context_recorder::search(uchar *tbl_name, size_t tbl_name_len)
 }
 
 void Optimizer_context_recorder::record_multi_range_read_info_const(
-    const TABLE_LIST *tbl,
-    uint keynr,
-    Range_print_enumerator *ranges,
-    ha_rows rows,
-    const Cost_estimate *cost,
-    ha_rows max_index_blocks,
-    ha_rows max_row_blocks)
+    const TABLE_LIST *tbl, uint keynr, Range_print_enumerator *ranges,
+    ha_rows rows, const Cost_estimate *cost, const ha_rows *max_index_blocks,
+    const ha_rows *max_row_blocks)
 {
   /*
     Do not record calls that are made at execution phase by "Range checked
@@ -801,8 +797,17 @@ void Optimizer_context_recorder::record_multi_range_read_info_const(
 
   range_ctx->rows= rows;
   range_ctx->cost= *cost;
-  range_ctx->max_index_blocks= max_index_blocks;
-  range_ctx->max_row_blocks= max_row_blocks;
+  if (rows != HA_POS_ERROR)
+  {
+    range_ctx->max_index_blocks= *max_index_blocks;
+    range_ctx->max_row_blocks= *max_row_blocks;
+  }
+  else
+  {
+    // Not provided. Write 0.
+    range_ctx->max_index_blocks= 0;
+    range_ctx->max_row_blocks= 0;
+  }
 
   while (!ranges->next())
   {
diff --git a/sql/opt_context_store_replay.h b/sql/opt_context_store_replay.h
@@ -47,13 +47,12 @@ class Optimizer_context_recorder
 
   ~Optimizer_context_recorder();
 
-  void record_multi_range_read_info_const(const TABLE_LIST *tbl,
-                                          uint keynr,
+  void record_multi_range_read_info_const(const TABLE_LIST *tbl, uint keynr,
                                           Range_print_enumerator *ranges,
                                           ha_rows rows,
                                           const Cost_estimate *cost,
-                                          ha_rows max_index_blocks,
-                                          ha_rows max_row_blocks);
+                                          const ha_rows *max_index_blocks,
+                                          const ha_rows *max_row_blocks);
 
   void record_cost_index_read(const TABLE_LIST *tbl,
                               uint key, ha_rows records, bool eq_ref,
diff --git a/sql/opt_range.cc b/sql/opt_range.cc
@@ -12471,6 +12471,7 @@ ha_rows check_quick_select(PARAM *param, uint idx, ha_rows limit,
   ha_rows replay_ctx_max_index_blocks;
   ha_rows replay_ctx_max_row_blocks;
   bool replay_ctx_rc;
+  TABLE::OPT_RANGE *range= param->table->opt_range + keynr;
   DBUG_ENTER("check_quick_select");
 
   /* Range not calculated yet */
@@ -12534,7 +12535,6 @@ ha_rows check_quick_select(PARAM *param, uint idx, ha_rows limit,
   param->quick_rows[keynr]= rows;
   if (rows != HA_POS_ERROR)
   {
-    TABLE::OPT_RANGE *range= param->table->opt_range + keynr;
     ha_rows table_records= param->table->stat_records();
     if (rows > table_records)
     {
@@ -12560,15 +12560,6 @@ ha_rows check_quick_select(PARAM *param, uint idx, ha_rows limit,
     range->max_row_blocks=
         MY_MIN(file->row_blocks(), rows * file->stats.block_size / IO_SIZE);
 
-    if (Optimizer_context_recorder *rec= param->thd->opt_ctx_recorder)
-    {
-      Range_print_enumerator_impl range_iter(param, idx, tree);
-      rec->record_multi_range_read_info_const(param->table->pos_in_table_list,
-                                              keynr, &range_iter, rows, cost,
-                                              range->max_index_blocks,
-                                              range->max_row_blocks);
-    }
-
     if (update_tbl_stats)
     {
       param->table->opt_range_keys.set_bit(keynr);
@@ -12597,6 +12588,19 @@ ha_rows check_quick_select(PARAM *param, uint idx, ha_rows limit,
     }
   }
 
+  if (Optimizer_context_recorder *rec= param->thd->opt_ctx_recorder)
+  {
+    Range_print_enumerator_impl range_iter(param, idx, tree);
+    /*
+      We pass range->max_index_blocks, and range->max_row_blocks by address,
+      as they might not have been initialized except when rows != HA_POS_ERROR.
+      This way, ASAN and valgrind also wouldn't complain.
+    */
+    rec->record_multi_range_read_info_const(
+        param->table->pos_in_table_list, keynr, &range_iter, rows, cost,
+        &range->max_index_blocks, &range->max_row_blocks);
+  }
+
   /* Figure out if the key scan is ROR (returns rows in ROWID order) or not */
   enum ha_key_alg key_alg= param->table->key_info[seq.real_keyno].algorithm;
   if ((key_alg != HA_KEY_ALG_BTREE) && (key_alg!= HA_KEY_ALG_UNDEF))
