As usual, I mostly care about optimizer, InnoDB and replication related bugs. Here is the list:
- Bug #97476 - "Range optimizer skips rows". This bug reported by Ilya Raudsepp looks like a clear regression in MySQL 8.0.x comparing to MySQL 5.7.x at least. I get the following correct results with 5.7.29:
mysql> SELECT t.id
-> FROM Test t
-> JOIN (
-> SELECT item_id, MAX(created_at) AS created_at
-> FROM Test t
-> WHERE (platform_id = 2) AND (item_id IN (3,2,111)) AND (type = 'Default')
-> GROUP BY item_id
-> ) t2 ON t.item_id = t2.item_id
-> t.item_id = t2.item_id
-> AND t.created_at = t2.created_at
-> AND t.type = 'Default'
-> WHERE t.platform_id = 2;
+----+
| id |
+----+
| 6 |
| 3 |
| 5 |
+----+
3 rows in set (0,03 sec)
mysql> select version();
+-----------+
| version() |
+-----------+
| 5.7.29 |
+-----------+
1 row in set (0,02 sec) - Bug #97531 - "5.7 replication breakage with syntax error with GRANT management". This tricky bug reported by Simon Mudd applies also to MySQL 8.0.x. It is closed as fixed, but the fix had not made it to recent 5.7.29 and 8.0.19 releases, so you'll have to wait for few more months.
- Bug #97552 - "Regression: LEFT JOIN with Impossible ON condition performs slowly". Yet another optimizer regression in MySQL 8 (comparing to 5.7.x) that is fixed only in MySQL 8.0.20+. The bug was reported by Fredric Johansson.
- Bug #97648 - "Bug in order by clause in union clause". Yet another regression (at least from user's point of view) in recent MySQL 5.7.x and 8.0.x comparing to 5.6.x. This time without a "regression" tag. The bug was reported by Andrei Mart.
- Bug #97662 - "MySQL v8.0.18 FIPS mode is no longer supported". According to Ryan L, MySQL 8.0.18+ is no longer supporting ssl_fips_mode=STRICT, as OpenSSL 1.1.1 is not FIPS-compatible and MySQL Server must be compiled using OpenSSL 1.1.1 or higher. That's interesting. Check also this link.
- Bug #97682 - "Handler fails to trigger on Error 1049 or SQLSTATE 42000 or plain sqlexception". This regression (comparing to MySQL 5.7) was reported by Jericho Rivera. It is fixed in MySQL 8.0.20. The patch was provided by Kamil Holubicki.
- Bug #97692 - "Querying information_schema.TABLES issue". I do not see any documented attempt to check on MySQL 8.0, so I had to add a comment to the bug report. From what I see, in MySQL 8.0.19 we still get different (empty) result from the second query, but at least now we have a warning:
mysql> SELECT ts.TABLE_SCHEMA
The bug was reported by Vinicius Malvestio Grippa.
-> FROM information_schema.TABLES ts
-> WHERE ts.TABLE_TYPE ='VIEW'
-> AND ts.TABLE_SCHEMA NOT IN ('sys')
-> AND ts.TABLE_COMMENT LIKE '%invalid%';
+--------------+
| TABLE_SCHEMA |
+--------------+
| test |
+--------------+
1 row in set, 1 warning (0,00 sec)
mysql> show warnings\G
*************************** 1. row ***************************
Level: Warning
Code: 1356
Message: View 'test.v' references invalid table(s) or column(s) or function(s) or definer/invoker of view lack rights to use them
1 row in set (0,00 sec)
mysql> select version();
+-----------+
| version() |
+-----------+
| 8.0.19 |
+-----------+
1 row in set (0,00 sec)
- Bug #97693 - "ALTER USER user IDENTIFIED BY 'password' broken by invalid authentication_string". The bug was reported by Nikolai Ikhalainen. MySQL 8.0.19 is still affected.
- Bug #97694 - "MySQL 8.0.18 fails on STOP SLAVE/START SLAVE stress test". For some reason I do not see any documented attempt to verify this on MySQL 5.7 also. The bug was reported by Przemysław Skibiński, who also suggested a fix.
- Bug #97734 - "Document the correct method to stop slaving with MTS without a warning or error". I can only agree with this request from Buchan Milne. Please. do :)
- Bug #97735 - "ALTER USER IF EXISTS ... WITH_MAX_USER_CONNECTIONS 9999 not applied correctly". yet another bug report by Simon Mudd in this list. For some reason, again, I do not see any documented attempt to verify the bug on MySQL 8.0.x, while there is no clear reason to think it is not affected.
- Bug #97742 - "bad item ref from correlated subquery to outer distinct table". This bug was reported by Song Zhibai, who also had contributed a patch. Based on further comments from Øystein Grøvlen and these results:
mysql> EXPLAIN SELECT f3 FROM t1 HAVING (SELECT 1 FROM t2 HAVING f2 LIMIT 1);
I'd say that MySQL 5.7.x is also affected, but for some reason nobody documented any attempt to verify it there. So, I've added a comment.
+----+--------------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+--------------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
| 1 | PRIMARY | t1 | NULL | ALL | NULL | NULL | NULL | NULL | 3 | 100.00 | NULL |
| 2 | DEPENDENT SUBQUERY | t2 | NULL | index | NULL | PRIMARY | 4 | NULL | 1 | 100.00 | Using index |
+----+--------------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
2 rows in set, 2 warnings (0,00 sec)
mysql> show warnings\G
*************************** 1. row ***************************
Level: Note
Code: 1276
Message: Field or reference 'f2' of SELECT #2 was resolved in SELECT #1
*************************** 2. row ***************************
Level: Note
Code: 1003
Message: /* select#1 */ select `test`.`t1`.`f3` AS `f3` from `test`.`t1` having (/* select#2 */ select 1 from `test`.`t2` having `test`.`t1`.`f2` limit 1)
2 rows in set (0,00 sec)
mysql> select version();
+-----------+
| version() |
+-----------+
| 5.7.29 |
+-----------+
1 row in set (0,00 sec) - Bug #97777 - "separate global variables (from hot variables) using linker script (ELF)". Beautiful bug report from Daniel Black.With a lot of details, perf and readelf outputs and patch contributed. See also his Bug #97822 - "buf_page_get_gen buf_pool->stat.n_page_gets++ is a cpu waste", with perf analysis up to a single assembler instruction level and fix suggested.
- Bug #97825 - "dd_mdl_acquire in dd_table_open with dict_sys->mutex hold may cause deadlock". Here I am really puzzled by no visible attempt to check the arguments of bug reporter, Dave Do, who tried to perform lock order analysis by code review. All we see as a result is this:
"Lock order could be different, but it is irrelevant, since these are locks on totally different levels and can't, in themselves, cause any deadlock."
What a great argument! Not a bug, surely... We trust you.
 |
| "What bugs are you talking about? I have no bugs, neither does MySQL 8!" |
- MySQL 8 introduces some optimizer (and some other) regressions. They seem to be fixed fast enough, but I wonder why only Community users were able to find them not Oracle's QA...
- MySQL 8.0.19 is surely great, but I see many serious bugs fixed o0nly in 8.0.20+.
- Percona, Booking and Facebook engineers still continue contributing high quality bug reports, comments/verification details and patches. Oracle is lucky to have such nice partners in making MySQL better.
- I still see problems with following proper verification procedures and documenting the results. Too often the bug reported for 8.0.x is NOT checked on 5.7.x as well, regression tag is not set, and so on. Sometimes reports are closed as "Not a bug" without any attempt to follow the analysis provided or prove the point. This is sad and wrong.
No comments:
Post a Comment