In my opinion many of the new features of MariaDB lacks some proper documentation/benchmarks and sometimes even testing, there are many features and new variables that could be used but almost nobody knows about these or the possible usecases/downsides are not clear and users will just keep it safe and not try these unless they are sure of what they are playing with. It might be better for the end-user to have less new features but more knowledge/clarity about what he can do with the already available ones. In that matter, creating tasks on Jira to follow up features/tools that has been added/modified and putting milestones on the task such as "Create KB page on this", "Do a benchmark for this on this case" or "Document the possible downsides and usecases of that feature" might be a good idea as you can follow up the advancement and other peoples might contribute as well. It might also avoid certain questions on the KB which is IMO poorly designed in that aspect (answers are put in comments). A very short list of examples of these features/options that would benefit from a better documentation/visibility : - The optimize_join_buffer_size optimizer switch is not documented at all and is disabled by default, what is its purpose and why isnt it activated by default (lack of testing of downsides?). - Its not clear on what these variables do and which engine they apply to (there is a bug about that https://mariadb.atlassian.net/browse/MDEV-4395) : deadlock_search_depth_long, deadlock_search_depth_short, deadlock_timeout_long, deadlock_timeout_short. - The default value of aria_repair_threads is 1, are there any downsides if an higher value is set, benchmarks might also come in interest if this value can make great performance differences. - What is the use/action (if any) of the thread_cache_size variable when the threadpool is activated. - More thorough benchmarks on Aria, especially with TRANSACTIONAL = 0 & 1 compaired to MyISAM and InnoDB. - Document how to define a correct value for the segmented key cache of MyISAM (knowing that it is not working like the innodb buffer pool instances) and is there any possible downside on its usage as its deactivated by default (except for some critical bug on it such as MDEV-4409). Le 14/06/2013 17:30, Roberto Spadim a écrit :
about quality assurance process, what should be done? unit tests? test team?