Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.5.8, 10.5
-
None
Description
It is reported by one of our customers the my_print_defaults -e option is not working with MariaDB server 10.5
Here are the test results:
At MariaDB server 10.4.17, my_print_defaults -e giving the required output. Whereas at MariaDB server 10.5.x, -e option is not working instead has to specify --defaults-extra-file (alias).
/root/my.cnf
[mariadb-10.4]
|
key_buffer_size=20M
|
max_allowed_packet=250M
|
table_open_cache=1000
|
table_definition_cache=2000
|
read_buffer_size=1M
|
thread_cache_size=8
|
max_connections=1024
|
long_query_time=60
|
slow_query_log=1
|
[root@test001 ~]# mysql --version
|
mysql Ver 15.1 Distrib 10.4.17-MariaDB, for Linux (x86_64) using readline 5.1
|
|
|
[root@test001 ~]# my_print_defaults -e /root/my.cnf --mysqld mysql.server
|
--key_buffer_size=20M
|
--max_allowed_packet=250M
|
--table_open_cache=1000
|
--table_definition_cache=2000
|
--read_buffer_size=1M
|
--thread_cache_size=8
|
--max_connections=1024
|
--long_query_time=60
|
--slow_query_log=1
|
/root/my.cnf
[mariadb-10.5]
|
key_buffer_size=20M
|
max_allowed_packet=250M
|
table_open_cache=1000
|
table_definition_cache=2000
|
read_buffer_size=1M
|
thread_cache_size=8
|
max_connections=1024
|
long_query_time=60
|
slow_query_log=1
|
[root@test001 ~]# mysql --version
|
mysql Ver 15.1 Distrib 10.5.8-MariaDB, for Linux (x86_64) using readline 5.1
|
[root@test001 ~]#
|
[root@test001 ~]# my_print_defaults -e /root/my.cnf --mysqld mysql.server
|
[root@test001 ~]# my_print_defaults --defaults-extra-file=/root/my.cnf --mysqld mysql.server
|
--key_buffer_size=20M
|
--max_allowed_packet=250M
|
--table_open_cache=1000
|
--table_definition_cache=2000
|
--read_buffer_size=1M
|
--thread_cache_size=8
|
--max_connections=1024
|
--long_query_time=60
|
--slow_query_log=1
|
Attachments
Issue Links
- relates to
-
-
- Closed
-
Activity
rucha174, no, as we discussed there should be no changes in my_default.c, instead my_print_defaults should generate a new command line specifically for calling load_default(), as doesn't make sense to reuse it in this case.
There are two approaches, basically. One is to let handle_options() handle --defaults* options too. This will make -e to work, --defaults-extra-file⎵ to work, in any position on the command line, very flexible, anything works. But this is very inconsistent with how --default* options work in any other executable (only at the beginning on the command line, only with =).
The other option is to let load_defaults() (that is, get_defaults_options()) handle --default* options. This will provide exactly the same behavior as elsewhere, so one won't, for example, wonder why --defaults-extra-file my.cnf works in my_print_defaults, but not in mysqldump. But -e and -g won't work.
I personally think that the second option would be less confusing long term.
Created a patch to address this specific issue by changing mysql.server script to use --defaults-extra-file instead of -e so that this issue can be closed faster.
Created another issue about -e not working correctly in 10.5+ : MDEV-25908
How is MDEV-24248 fixed? mysql_print_default -e still doesn't work and mysql.server is still using -e, isn't it?
Hi serg It is only MDEV-24248. As discussed I had changed mysql.server to use --defaults-extra-file instead o f -e. mysql.server is not using -e anymore. See https://github.com/MariaDB/server/commit/534553897f7178eb1456b11aa6c70f6025364407
I had reopened this issue because my_print_defaults.test was failing in old buildbot. The reason was it was reading default config file before reading the extra default file. And so there was diff in the result file on buildbot. Fixed it by adding adding a default config file for the test along with defaults extra file.
MDEV-25908 is still open where I have fixed -e option 10.5 and removed short options for 10.7.
Yes, you fixed it in https://github.com/MariaDB/server/commit/f88d130e710
but in https://github.com/MariaDB/server/commit/826eab3f9b2 you reverted that change.
As of now, mysql.server in 10.5 is using -e again. That's why I asked, why this issue is closed.
closing the issue because fix is already pushed: https://github.com/MariaDB/server/commit/534553897f7178eb1456b11aa6c70f6025364407
(And small fixup to remove config file made during test, which was missed in previous commit: https://github.com/MariaDB/server/commit/675b72d0da1839a3988ec29bc3ea06372890be54)
The difference comes from this commit:
commit 3e56972712395d371f82cda564b039dcbda3100aAuthor: Sergei GolubchikDate: Sat Oct 12 21:21:50 2019 +0200cleanup: unify --defaults* option handlingprocess all --defaults* options uniformly,get rid of special case for --no-defaults and --print-defaultsuse realpath instead of blindly concatenating pwd and relative path.