Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.2(EOL), 10.3(EOL)
-
None
Description
https://travis-ci.org/elenst/travis-tests/jobs/409014461
# 2018-07-28T04:59:50 [24239] Running mysql_upgrade:
|
# 2018-07-28T04:59:50 [24239] "/home/travis/server/bin/mysql_upgrade" --host=127.0.0.1 --port=19300 -uroot
|
ERROR 1071 (42000) at line 596: Specified key was too long; max key length is 768 bytes
|
ERROR 1071 (42000) at line 600: Specified key was too long; max key length is 768 bytes
|
ERROR 1071 (42000) at line 603: Specified key was too long; max key length is 768 bytes
|
FATAL ERROR: Upgrade failed
|
# 2018-07-28T04:59:51 [24239][ERROR] mysql_upgrade failed
|
Nothing special is needed on MySQL side, apart from bootstrap with page_size 4K. The pre-created datadir is attached. The exact steps to create it were these:
- start MySQL 5.7.23 in initialize mode:
/data/bld/mysql-5.7/bin/mysqld --no-defaults --basedir=/data/bld/mysql-5.7 --datadir=/data/bld/mysql-5.7/data --log-error=/data/bld/mysql-5.7/data/log.err --loose-lc-messages-dir= --loose-language= --port=3306 --socket=/tmp/mysql.sock --tmpdir=/tmp --loose-core-file --initialize --innodb-page-size=4K - start the same server in normal mode, with skip-grant-tables:
/data/bld/mysql-5.7/bin/mysqld --no-defaults --basedir=/data/bld/mysql-5.7 --datadir=/data/bld/mysql-5.7/data --log-error=/data/bld/mysql-5.7/data/log.err --loose-lc-messages-dir= --loose-language= --port=3306 --socket=/tmp/mysql.sock --tmpdir=/tmp --loose-core-file --innodb-page-size=4K --skip-grant-tables - execute
MySQL [(none)]> flush privileges;
Query OK, 0 rows affected (0.01 sec)
MySQL [(none)]> set password for root@localhost = '';
Query OK, 0 rows affected (0.00 sec)
- shut down the server normally.
Note: Upon initialization, MySQL sets one-time password which it writes to the error log. MySQL one-time passwords don't work for me reliably on some reason, that's why I reset it using skip-grant-tables. It's not important for the upgrade issue.
To reproduce the upgrade problem, start recent MariaDB 10.2 on the created datadir (which you either created as above, or unpacked from the attachment). Don't forget innodb_page_size=4K.
The server should complain in the error log but start all right. Then run mysql_upgrade with root user without parameters, like
bin/mysql_upgrade -uroot --protocol=tcp
|
It complains a lot while running, but the real (fatal) problem happens on phase 4/7:
Phase 4/7: Running 'mysql_fix_privilege_tables'
|
ERROR 1071 (42000) at line 596: Specified key was too long; max key length is 768 bytes
|
ERROR 1071 (42000) at line 600: Specified key was too long; max key length is 768 bytes
|
ERROR 1071 (42000) at line 603: Specified key was too long; max key length is 768 bytes
|
FATAL ERROR: Upgrade failed
|
Attachments
Issue Links
- relates to
-
-
- Closed
-
Activity
alter table mysql.innodb_index_stats modify last_update timestamp not null default current_timestamp on update current_timestamp; |
alter table mysql.innodb_table_stats modify last_update timestamp not null default current_timestamp on update current_timestamp; |
mysql_upgrade runs the statement for innodb_table_stats twice, hence 3 errors.
The SQL error can be reproduced by manually executing the same statements, mysql_upgrade doesn't do anything special to achieve it.
The dataset works with bb-10.2-marko, with the second commit of MDEV-14637 which (among other things) makes ha_innobase::max_supported_key_length() return a longer limit for innodb_page_size=4k.
I think that we must fix this as part of
MDEV-14637. What are the failing SQL statements?