Is there any current (October 2023) solution for MariaDB 11.1.2-GA on Windows 11 Pro crashing every couple of days or even sometimes every couple of hours with the following error in .err log:
InnoDB: Missing FILE_CREATE, FILE_DELETE or FILE_MODIFY before FILE_CHECKPOINT
I am asking for current report on this problem (or solution the best), because every report that I have found says that this is an old bug that was supposedly resolved long time ago. Which turns out to be not true.
Every DBA question, JIRA ticket or blog entry says that this was fixed in 10.7.5 and yet it happens in 11.1.2-GA again. And I can't find any report on any kind of regression. Of course, I cannot find any solution as well, hence this question.
Things I have tried so far:
Setting
innodb_force_recoveryon any value (here or here or here) doesn't work and still causesmysqldto quit with the above errorDeleting ib_logfile0 (here or here) doesn't work. MariaDB dies with error that this file is missing instead of recreating it, as said
Setting
innodb_log_file_size = 48Mdoesn't help. I already had it said to this value. I didn't tried to set it to1000M, but that sounds like an overkill.Setting
innodb_buffer_pool_size = 10Mdoesn't work as well. I already had it said to this value
Deleting my corrupted files is the only solution that I have found working:
- Rename c:\[path]\data into c:\[path]\data2
- Execute
mysqld(usually dies, but recreates files in c:\[path]\data) - Overwrite all files in c:\[path]\data2 with recreated copies from c:\[path]\data
- Delete c:\[path]\data
- Rename c:\[path]\data2 into c:\[path]\data
- Execute
mysqld
At this points MariaDB starts and stops normally, but of course the entire dataset is not correct, because such bluntly performed copy process isn't any backup-restore process at all. Everything must be recreated manually, i.e.:
- Access to phpMyAdmin, including all it's tables and settings
- All users, passwords, privileges using PMA
- All data tables by SQL import
Usually takes an hour or two. And then, after couple of hours or days at most, data gets corrupted again, and the whole story starts from the beginning.
This is my private laptop, my dev environment. I am the only person using it and PHP running my application is the only process writing to MariaDB. So by no mean we can say anything about some race conditions that are mentioned in MariaDB's JIRA tickets.