From 554587b8f7a6a7e743de65e429791d2959470585 Mon Sep 17 00:00:00 2001
From: Sergei Petrunia <sergey@mariadb.com>
Date: Thu, 5 Feb 2026 16:51:31 +0200
Subject: [PATCH] MDEV-38273: Optimizer trace should have selectivities
 collected via sampling

Add to optimizer trace:
 "sampled_selectivity": [
    { "cond":"condition", "selectivity": n.nnnn }
    ...
  ]
---
 mysql-test/main/opt_trace_selectivity.result | 56 ++++++++++++++++++++
 mysql-test/main/opt_trace_selectivity.test   | 23 ++++++++
 sql/opt_range.cc                             |  8 +++
 3 files changed, 87 insertions(+)

diff --git a/mysql-test/main/opt_trace_selectivity.result b/mysql-test/main/opt_trace_selectivity.result
index 26d5671f0645f..0d247049f97ba 100644
--- a/mysql-test/main/opt_trace_selectivity.result
+++ b/mysql-test/main/opt_trace_selectivity.result
@@ -367,4 +367,60 @@ JS
     ]
 ]
 drop table t1;
+#
+# MDEV-38273: Optimizer trace should have selectivities collected via sampling
+#
+create table t1 (a int, b int, c varchar(32), d varchar(32));
+insert into t1
+select
+seq, seq,
+if(mod(seq, 10) < 4,'c-ccc-c', 'no-match'),
+if(mod(seq, 10) < 3,'d-ddd-d', 'no-match')
+from seq_1_to_1000;
+analyze table t1 persistent for all;
+Table	Op	Msg_type	Msg_text
+test.t1	analyze	status	Engine-independent statistics collected
+test.t1	analyze	status	OK
+set statement optimizer_use_condition_selectivity=5 for
+explain select * from t1
+where a < 700 and b < 500 and c like '%ccc%' and d like '%ddd%';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	1000	Using where
+select JSON_DETAILED(JSON_EXTRACT(trace, '$**.rows_estimation[0]')) as JS
+from INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+JS
+[
+    {
+        "selectivity_for_indexes": 
+        [],
+        "selectivity_for_columns": 
+        [
+            {
+                "column_name": "a",
+                "ranges": 
+                ["NULL < a < 700"],
+                "selectivity_from_histogram": 0.699
+            },
+            {
+                "column_name": "b",
+                "ranges": 
+                ["NULL < b < 500"],
+                "selectivity_from_histogram": 0.499
+            }
+        ],
+        "sampled_selectivity": 
+        [
+            {
+                "cond": "t1.c like '%ccc%'",
+                "selectivity": 0.4
+            },
+            {
+                "cond": "t1.d like '%ddd%'",
+                "selectivity": 0.3
+            }
+        ],
+        "cond_selectivity": 0.04185612
+    }
+]
+drop table t1;
 set optimizer_trace='enabled=off';
diff --git a/mysql-test/main/opt_trace_selectivity.test b/mysql-test/main/opt_trace_selectivity.test
index 40f78d91db788..ef23bdf6299cb 100644
--- a/mysql-test/main/opt_trace_selectivity.test
+++ b/mysql-test/main/opt_trace_selectivity.test
@@ -83,4 +83,27 @@ select JSON_DETAILED(JSON_EXTRACT(trace, '$**.selectivity_for_indexes')) as JS
 from INFORMATION_SCHEMA.OPTIMIZER_TRACE;
 
 drop table t1;
+
+--echo #
+--echo # MDEV-38273: Optimizer trace should have selectivities collected via sampling
+--echo #
+create table t1 (a int, b int, c varchar(32), d varchar(32));
+insert into t1
+select
+  seq, seq,
+  if(mod(seq, 10) < 4,'c-ccc-c', 'no-match'),
+  if(mod(seq, 10) < 3,'d-ddd-d', 'no-match')
+from seq_1_to_1000;
+
+analyze table t1 persistent for all;
+
+set statement optimizer_use_condition_selectivity=5 for
+explain select * from t1
+where a < 700 and b < 500 and c like '%ccc%' and d like '%ddd%';
+
+select JSON_DETAILED(JSON_EXTRACT(trace, '$**.rows_estimation[0]')) as JS
+from INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+
+drop table t1;
+
 set optimizer_trace='enabled=off';
diff --git a/sql/opt_range.cc b/sql/opt_range.cc
index bb00556e92b07..6e0def3618ff9 100644
--- a/sql/opt_range.cc
+++ b/sql/opt_range.cc
@@ -3873,6 +3873,7 @@ bool calculate_cond_selectivity_for_table(THD *thd, TABLE *table, Item **cond)
   if (*cond && check_rows > SELECTIVITY_SAMPLING_THRESHOLD &&
       thd->variables.optimizer_use_condition_selectivity > 4)
   {
+    Json_writer_array trace_sampled_sel(thd, "sampled_selectivity");
     find_selective_predicates_list_processor_data *dt=
       (find_selective_predicates_list_processor_data *)
       alloc_root(thd->mem_root,
@@ -3902,6 +3903,13 @@ bool calculate_cond_selectivity_for_table(THD *thd, TABLE *table, Item **cond)
                               (double)stat->positive / examined_rows));
           double selectivity= ((double)stat->positive) / examined_rows;
           table->multiply_cond_selectivity(selectivity);
+
+          if (unlikely(trace_sampled_sel.trace_started()))
+          {
+            Json_writer_object selectivity_for_cond(thd);
+            selectivity_for_cond.add("cond", stat->cond);
+            selectivity_for_cond.add("selectivity", selectivity);
+          }
           /*
             If a field is involved then we register its selectivity in case
             there in an equality with the field.