rows is heuristic estimate of how many rows will need to be read. r_rows is the number of rows actually read. Since the ANALYZE outputs are close enough that they don't explain the difference, this is probably going to require some deeper debugging. Consider filing a performance bug on MariaDB MDEV JIRA board. On Tue, 25 Mar 2025 at 10:48, Olivier Miquel <olivier.miquel@mri.cnrs.fr> wrote:
Hello Gordan,
Yes this is same disk, same filesystem. Only the mariadb engine version differs.
Here are the outputs required.
What does mean this rows value difference in the output ?
Many thanks,
O.
Mariadb 10.3 :
MariaDB [cfms_accounting]> ANALYZE select count(recipient) from transactions; +------+-------------+--------------+------+---------------+------+---------+------+---------+------------+----------+------------+-------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | r_rows | filtered | r_filtered | Extra | +------+-------------+--------------+------+---------------+------+---------+------+---------+------------+----------+------------+-------+ | 1 | SIMPLE | transactions | ALL | NULL | NULL | NULL | NULL | 1779448 | 1912563.00 | 100.00 | 100.00 | | +------+-------------+--------------+------+---------------+------+---------+------+---------+------------+----------+------------+-------+
Mariadb 10.11 :
+------+-------------+--------------+------+---------------+------+---------+------+---------+------------+----------+------------+-------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | r_rows | filtered | r_filtered | Extra | +------+-------------+--------------+------+---------------+------+---------+------+---------+------------+----------+------------+-------+ | 1 | SIMPLE | transactions | ALL | NULL | NULL | NULL | NULL | 1942730 | 1912563.00 | 100.00 | 100.00 | | +------+-------------+--------------+------+---------------+------+---------+------+---------+------------+----------+------------+-------+
Le 24/03/2025 à 18:30, Gordan Bobic a écrit :
On Mon, 24 Mar 2025 at 18:57, Olivier Miquel <olivier.miquel@mri.cnrs.fr> wrote:
Yes you're right! Sorry, i had swapped the two results. It is way slower with 10.11.
I could add an index on recipient, but why this is so faster with 10.3 without any indexes ?
What do you get from: A
I'm guessing that there's something i missed here. Same disks, same file system?
-- Olivier Miquel Administrateur Système et Réseau Plate-forme technologique MRI Tel : 06.50.19.27.43 Mail:olivier.miquel@mri.cnrs.fr
-- Gordan Bobic Database Specialist, Shattered Silicon Ltd. https://shatteredsilicon.net