Details
-
Bug
-
Status: In Progress (View Workflow)
-
Critical
-
Resolution: Unresolved
-
13.0
-
Can result in unexpected behaviour
-
Q2/2026 Server Development
Description
If all InnoDB write-ahead logs are available from the very beginning (the initial LSN=12288), crash recovery should be possible even if all data files are missing. The following will simulate that scenario when executed in an out-of-source tree build directory:
mysql-test/mtr --mysqld=--innodb-log-archive main.1st
|
rm mysql-test/var/mysqld.1/data/ibdata1 mysql-test/var/mysqld.1/data/undo00*
|
sql/mariadbd --datadir "$PWD"/mysql-test/var/mysqld.1/data
|
|
main 7f0ed40ad551e7db7506b5576f00d1435abb4b72 |
2026-05-21 10:08:13 0 [ERROR] InnoDB: redo log file './/ib_0000000000003000.log' exists. Creating system tablespace with existing redo log file is not recommended. Please delete redo log file before creating new system tablespace.
|
If only the system tablespace is missing, then the message will be different:
|
main 7f0ed40ad551e7db7506b5576f00d1435abb4b72 |
2026-05-21 10:07:58 0 [ERROR] InnoDB: undo tablespace './/undo001' exists. Creating system tablespace with existing undo tablespaces is not supported. Please delete all undo tablespaces before creating new system tablespace.
|
A recovery attempt with a dummy empty tablespace file
touch mysql-test/var/mysqld.1/data/ibdata1
|
sql/mariadbd --datadir "$PWD"/mysql-test/var/mysqld.1/data
|
fails in a different way:
|
main 7f0ed40ad551e7db7506b5576f00d1435abb4b72 |
2026-05-21 10:15:41 0 [ERROR] InnoDB: The Auto-extending data file './ibdata1' is of a different size 0 pages than specified by innodb_data_file_path
|
Let us try a bit harder:
fallocate -l 12m mysql-test/var/mysqld.1/data/ibdata1
|
sql/mariadbd --datadir "$PWD"/mysql-test/var/mysqld.1/data
|
|
main 7f0ed40ad551e7db7506b5576f00d1435abb4b72 |
2026-05-21 10:17:21 0 [Note] InnoDB: Header page consists of zero bytes in datafile: ./ibdata1, Space ID: 0, Flags: 0
|
2026-05-21 10:17:21 0 [ERROR] InnoDB: The data file './ibdata1' has the wrong space ID. It should be 0, but 0 was found
|
echo one|dd bs=4 conv=notrunc seek=7 of=mysql-test/var/mysqld.1/data/ibdata1
|
sql/mariadbd --datadir "$PWD"/mysql-test/var/mysqld.1/data
|
|
main 7f0ed40ad551e7db7506b5576f00d1435abb4b72 |
2026-05-21 10:21:06 0 [Note] InnoDB: End of log at LSN=54104
|
2026-05-21 10:21:06 0 [ERROR] InnoDB: The change buffer is corrupted
|
It would be better if recovery worked in all these scenarios. We would want MDEV-14992 BACKUP SERVER to avoid copying any InnoDB data files that have been created after the start of the archived history.
Attachments
Issue Links
- blocks
-
MDEV-14992 BACKUP SERVER to mounted file system
-
- In Progress
-
- is caused by
-
-
- Closed
-