Details
-
Bug
-
Status: Open (View Workflow)
-
Critical
-
Resolution: Unresolved
-
10.6, 10.11, 11.4, 11.8, 12.2, 12.3
-
Can result in unexpected behaviour
Description
In recent testing, a lot of system table .frm corruption errors started showing. A few examples:
MARIADBD_ERROR|mariadbd: Information erronnée dans le fichier: './mysql/proc.frm'
|
MARIADBD_ERROR|mariadbd: Nesprávná informace v souboru './mysql/innodb_index_stats.frm'
|
MARIADBD_ERROR|mariadbd: Nesprávná informace v souboru './mysql/innodb_table_stats.frm'
|
MARIADBD_ERROR|mariadbd: Некорректная информация в файле './mysql/general_log.frm'
|
MARIADBD_ERROR|mariadbd: Некорректная информация в файле './mysql/global_priv.frm'
|
MARIADBD_ERROR|mariadbd: Некорректная информация в файле './mysql/proc.frm'
|
MARIADBD_ERROR|mariadbd: Некорректная информация в файле './mysql/tables_priv.frm'
|
MARIADBD_ERROR|mariadbd: Некорректная информация в файле './sys/sys_config.frm'
|
MARIADBD_ERROR|mariadbd: არასწორი ინფორმაცია ფაილში: './mysql/proc.frm'
|
MARIADBD_ERROR|mariadbd: ファイル './mysql/general_log.frm' 内の情報が不正です。
|
MARIADBD_ERROR|mariadbd: 不能从加载mysql.host。表可能损坏了
|
MARIADBD_ERROR|mariadbd: 文件中的信息不正确:'./mysql/columns_priv.frm'
|
(Ignore the 'different languages' part; this is due to general test variability)
Looking at a few specific ones (in English):
SET max_session_mem_used=8192; |
CALL foo();
|
SHUTDOWN;
|
Leads to:
|
CS 10.6.25 cd02709a315c9f08965d6b8fb7e75baaae17a4f4 (Debug, Clang 21.1.3-20250923) Build 23/01/2026 |
2026-02-17 17:42:19 4 [ERROR] mariadbd: Incorrect information in file: './mysql/proc.frm'
|
And:
SET max_session_mem_used=8192; |
SELECT * FROM mysql.innodb_table_stats; |
SHUTDOWN;
|
Leads to:
|
CS 10.6.25 cd02709a315c9f08965d6b8fb7e75baaae17a4f4 (Debug, Clang 21.1.3-20250923) Build 23/01/2026 |
2026-02-17 17:42:58 4 [ERROR] mariadbd: Incorrect information in file: './mysql/innodb_table_stats.frm'
|
There are many other .frm tables mentioned (as can be seen from the various frm error examples at the top).
git-bisect showed the issue started happening after:
commit e5994025bec0d34ddef25df2e252b8f1d2e34e51
|
Author: Sergei Golubchik <serg@mariadb.org>
|
Date: Fri Oct 24 12:05:30 2025 +0200
|
|
|
MDEV-37938 very long query cannot be killed quickly
|
|
check for thd->killed in the lexer
|
|
|
sql/sql_lex.cc | 6 ++++++
|
1 file changed, 6 insertions(+)
|
After a restart, it does not look like the .frm corruptions persist so it may be that the memory-limit terminated query does not cause actual .frm file corruption. Still, the errors make it to the error log leading system DBA's to likely believe otherwise.
Attachments
Issue Links
- is caused by
-
MDEV-37938 very long query cannot be killed quickly
-
- Closed
-