MDEV-34713 Backup and restore mariadb_upgrade_info file

[FarihaIS]

Description

The mariadb_upgrade_info file was not being backed up during mariadb-backup --backup operations, causing MariaDB to incorrectly prompt for an upgrade even when restoring from a backup of the same version.

Without this file being preserved in the backup, mariadb-upgrade cannot determine the database version after restore and assumes an upgrade is needed, forcing users to run unnecessary upgrade operations during disaster recovery.

Add mariadb_upgrade_info to the list of files backed up from the datadir during backup operations, ensuring the upgrade state is preserved across backup and restore cycles.

Release Notes

N/A

How can this PR be tested?

Execute the mariabackup.full_backup test in mysql-test-run. This commit adds a test in this file that verifies the creation of mariadb_upgrade_info during a successful copy-back.

Before the fix

mysqltest: At line 91: command "file_exists" failed with error: 1  my_errno: 175  errno: 2

The result from queries just before the failure was:
< snip >
XA END 'zombie';
XA PREPARE 'zombie';
# restart: --innodb_undo_tablespaces=0
# xtrabackup backup
# xtrabackup prepare
# shutdown server
# remove datadir
# xtrabackup move back
# restart: --innodb_undo_tablespaces=0
# Display undo log files from target directory
undo001
undo002
XA COMMIT 'zombie';
DROP TABLE t;
#
# MDEV-34713: mariadb_upgrade_info should be backed up and restored
#
CREATE TABLE t2(i INT) ENGINE INNODB;
INSERT INTO t2 VALUES(100);
# xtrabackup backup

After the fix

mariabackup.full_backup '8k'             w4 [ pass ]  12850
mariabackup.full_backup '4k'             w3 [ pass ]  12990
mariabackup.full_backup '16k'            w5 [ pass ]  13000
mariabackup.full_backup '32k'            w2 [ pass ]  15527
mariabackup.full_backup '64k'            w1 [ pass ]  18216
--------------------------------------------------------------------------
The servers were restarted 0 times
Spent 72.583 of 25 seconds executing testcases

Completed: All 5 tests were successful.

Basing the PR against the correct MariaDB version

• This is a bug fix, and the PR is based against the main branch.

Copyright

All new code of the whole pull request, including one or several files that are either new files or modified ones, are contributed under the BSD-new license. I am contributing on behalf of my employer Amazon Web Services, Inc.

[CLAassistant]

All committers have signed the CLA.

[gkodinov]

Thank you for your contribution. This is a preliminary review.

Please also sort out the CLA bot! It must say "CLA signed".

[gkodinov]

Please see some answers to your questions below.

Also please clear the CLA bot!

[gkodinov]

Thank you for your work. I will ignore the CLA bot for now. Please let me know if you have any updates on that.

LGTM. Please stand by for the final review.

[vaintroub]

Nope, this idea is wrong. This file should not be created with mariabackup --copy-back with the version of mariabackup. This file should already be in the backup, original, from the MariaDB server, included during mariadb-backup --backup time. It should be correct version exactly how it was on the server, not the one taken from mariabackup on the destination machine.

Look into backup_files_from_datadir to see how it is done for AWS KMS keys (that's from key management plugin we do not distibute anymore), aria_log, and something from mroonga.

[vuvova]

changing the review status to reflect @vaintroub's comment

[vaintroub]

That looks good. My only comment is that the file mariadb_upgrade_info that was created in the datadir during the test, and now was proven to survive backup, restore and moveback, needs to be removed again, at the end of the test.

[vaintroub]

Looks good to me

[FarihaIS]

@vaintroub thank you for all the feedback!

[FarihaIS]

@gkodinov thank you for the approval. Looks like the only test that is failing now is main.vector, which seems unrelated to my changes - could you please help confirm?

[gkodinov]

@gkodinov thank you for the approval. Looks like the only test that is failing now is main.vector, which seems unrelated to my changes - could you please help confirm?

yes, vector is unrelated.