After upgrading my testing machine from 10.2.32-r2 -> 10.2.33 a huge performance drop was discovered during our production to testing restore procedure on Innodb. Normally we restore our 700G db in 2 days, but now it's only in 1/3 of the restore. It's probably caused by this: https://jira.mariadb.org/browse/MDEV-23475 I'll try to apply the patch (https://github.com/MariaDB/server/commit/309302a3dad5f06cb62b0846dcb8a3671d91ff29) locally and report back.
The restoring is much faster with the patch. In 9 hours we have restored 240G of data while it took 2 days to process that amount of data without the patch.
I can confirm that our weekly restore took the normal 2 day time using the patch (already included in 10.2.34).
Just checking the 10.2.34 changelog, but the fix is not included: https://mariadb.com/kb/en/mariadb-10234-changelog/ Instead, only 1 security fix is included and 1 bugfix, while the 10.2 branch is full of other fixes (along with the fix for this bug).
Yes, that's why we decided to rev bump current stable version to add the patch because we cannot stabilize new version.
Agreed, thanks.
The fix is included in https://mariadb.com/kb/en/mariadb-10235-changelog/, but failed to find the fix in later major versions.
Should be good, https://github.com/mariadb/server/commit/309302a3dad5f06cb62b0846dcb8a3671d91ff29
10.2: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=4a652a4365ac05c008f472f89e2dccd26b7a7411 10.3: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=ea72808af1e5a2b50828e64a4dbdb12cc8281a06 10.4: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=4b0ba59f737be3a7be0ced51dad072bef67342cf 10.5: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=a0f1f125fcb728c3ef1d5ac254a7f793bc6b931b
Huge thanks Thomas!
