Bug #113722 Test main.index_merge_innodb failing with a result difference

Description:
$ mtr main.index_merge_innodb
...
[ 50%] main.index_merge_innodb                   [ fail ]
        Test ended at 2024-01-23 13:32:12

CURRENT_TEST: main.index_merge_innodb
--- /Users/laurynas/vilniusdb/mysql-8.3.0/mysql-test/r/index_merge_innodb.result	2024-01-22 16:39:20
+++ /Users/laurynas/vilniusdb/mysql-8.3.0/_build-release/mysql-test/var/log/index_merge_innodb.reject	2024-01-23 14:32:11
@@ -647,24 +647,24 @@
     -> Inner hash join (no condition)  (rows=196)
         -> Filter: (((b.key7 = 1) and (b.key6 = 1) and (b.key5 = 1) and (b.key4 = 1) and (b.key3 = 1) and (b.key2 = 1) and (b.key1 = 1)) or (b.key8 = 1))  (rows=14)
             -> Deduplicate rows sorted by row ID  (rows=14)
-                -> Intersect rows sorted by row ID  (rows=13.4)
-                    -> Index range scan on b using i2 over (key2 = 1)  (rows=496)
-                    -> Index range scan on b using i3 over (key3 = 1)  (rows=496)
-                    -> Index range scan on b using i4 over (key4 = 1)  (rows=496)
-                    -> Index range scan on b using i5 over (key5 = 1)  (rows=496)
-                    -> Index range scan on b using i6,i7? over (key6 = 1)  (rows=496)
-                    -> Index range scan on b using i7 over (key7 = 1)  (rows=496)
+                -> Intersect rows sorted by row ID  (rows=13.6)
+                    -> Index range scan on b using i2 over (key2 = 1)  (rows=497)
+                    -> Index range scan on b using i3 over (key3 = 1)  (rows=497)
+                    -> Index range scan on b using i4 over (key4 = 1)  (rows=497)
+                    -> Index range scan on b using i5 over (key5 = 1)  (rows=497)
+                    -> Index range scan on b using i6,i7? over (key6 = 1)  (rows=497)
+                    -> Index range scan on b using i7 over (key7 = 1)  (rows=497)
                 -> Index range scan on b using i8 over (key8 = 1)  (rows=1)
         -> Hash
             -> Filter: (((a.key7 = 1) and (a.key6 = 1) and (a.key5 = 1) and (a.key4 = 1) and (a.key3 = 1) and (a.key2 = 1) and (a.key1 = 1)) or (a.key8 = 1))  (rows=14)
                 -> Deduplicate rows sorted by row ID  (rows=14)
-                    -> Intersect rows sorted by row ID  (rows=13.4)
-                        -> Index range scan on a using i2 over (key2 = 1)  (rows=496)
-                        -> Index range scan on a using i3 over (key3 = 1)  (rows=496)
-                        -> Index range scan on a using i4 over (key4 = 1)  (rows=496)
-                        -> Index range scan on a using i5 over (key5 = 1)  (rows=496)
-                        -> Index range scan on a using i6,i7? over (key6 = 1)  (rows=496)
-                        -> Index range scan on a using i7 over (key7 = 1)  (rows=496)
+                    -> Intersect rows sorted by row ID  (rows=13.6)
+                        -> Index range scan on a using i2 over (key2 = 1)  (rows=497)
+                        -> Index range scan on a using i3 over (key3 = 1)  (rows=497)
+                        -> Index range scan on a using i4 over (key4 = 1)  (rows=497)
+                        -> Index range scan on a using i5 over (key5 = 1)  (rows=497)
+                        -> Index range scan on a using i6,i7? over (key6 = 1)  (rows=497)
+                        -> Index range scan on a using i7 over (key7 = 1)  (rows=497)
                     -> Index range scan on a using i8 over (key8 = 1)  (rows=1)
 
 select max(a.key1 + b.key1 + a.key2 + b.key2 + a.key3 + b.key3 + a.key4 + b.key4 + a.key5 + b.key5)

mysqltest: Result content mismatch

How to repeat:
Server built with CMake options -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DBUILD_CONFIG=mysql_release -DCMAKE_BUILD_TYPE=Release -DMYSQL_MAINTAINER_MODE=ON -DWITH_SYSTEM_LIBS=ON -DFORCE_COLORED_OUTPUT=ON -DWITH_ZLIB=bundled

Hi,

Thank you for your bug report.

With our release binaries we get on both x86 and ARM:

./mtr main.index_merge_innodb
Logging: ./mtr  main.index_merge_innodb
MySQL Version 8.3.0
Checking supported features
Using 'all' suites
Collecting tests
Checking leftover processes
Removing old var directory
Creating var directory '/Users/sinisa/razno/work/repo/mysql-commercial-8.3.0-macos14-x86_64/mysql-test/var'
Installing system database
Using parallel: 1

==============================================================================
                  TEST NAME                       RESULT  TIME (ms) COMMENT
------------------------------------------------------------------------------
[ 50%] main.index_merge_innodb                   [ pass ]  12250
[100%] shutdown_report                           [ pass ]
------------------------------------------------------------------------------
The servers were restarted 0 times
The servers were reinitialized 0 times
Spent 12.250 of 22 seconds executing testcases

Completed: All 2 tests were successful.

Can you pinpoint which build option makes the problem ?????

HI Mr. Beveinis,

We have managed to build a release version of 8.3.0 with the following options:

-DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DBUILD_CONFIG=mysql_release -DCMAKE_BUILD_TYPE=Release -DMYSQL_MAINTAINER_MODE=ON -DWITH_SYSTEM_LIBS=ON -DFORCE_COLORED_OUTPUT=ON -DWITH_ZLIB=bundled

we added some more options. 

We managed to repeat the test failure on Sonoma with M2 CPU.

Verified as reported.

Same on 8.4.0.

Confirmed on 8.4.0.

Same on 8.4.1

Thank you Mr. Laurinas.

Same on 9.0.0

Same on 8.4.3

Thank you, Mr. Biveinis ......

Same on 8.4.4

Thank you, Mr. Laurynas.

Posted by developer:
 
The reported failure is fixed in MySQL 9.2.0 as part of bug#37111983.

Thank you, Knut.

Fixed in MySQL 9.2.0 by BUG#116255.

Testing only, no changelog entry needed.

Closed.

Thank you, Jon.

Still happens on 8.4.5 (as expected due to 9.2.0-only fix)

Still happens on 8.4.6 as expected

Still happens on 8.4.8

Still happens on 8.4.9