Martyn Dale a écrit :
Hi,
Personally I do not think it is advisable to move forward until certain issues have been addressed. I am talking about https://sourceforge.net/p/phpmyadmin/bugs/3944 and related bug reports.
The alleged solution for 4.1 is requiring MySQL 5.5 as a minimum version. This is not going to solve the problems that people have reported when using the 4.x series with large systems, since while 5.5 is better than 5.1 and below, a full directory scan can still be triggered by any number of combinations of variables as described here: http://dev.mysql.com/doc/refman/5.6/en/information-schema-optimization.html
I explained this in that bug report where i showed PMAs default usage of IS for the navigation on a *5.6* system (with innodb_stats_on_metadata switched off). While that was just a 2 second example, I have seen that get up to 60 seconds before query killers kick in. Reports of users on shared hosting accounts, where this kind of behavior can get users accounts terminated has also already been reported ( https://sourceforge.net/p/phpmyadmin/bugs/3945, second page "Here is what one provider sent as a warning before *deleting my account*")
If you remove the use of disableIS and related functionality, it will not solve any issues. Take the SCHEMATA queries for example for getting database lists, this still effectively does a full directory scan, as documented: http://dev.mysql.com/doc/refman/5.6/en/schemata-table.html In this case for example, switching to IS to avoid using the slow performing show databases is redundant since they basically do the same thing. As a basic example, take a directory, with tens of thousands of files, and attempt to do an ls in it, even unsorted (-U) it will take a long time, simply because traversal of the filesystem takes time. Requiring higher versions isnt going to miraculously speed up the iteration of directories.
On a similar note, when I referenced *innodb_stats_on_metadata *in earlier mentioned bug report, I was using it as an example of why relying on IS can be as bad, or worse than directly doing a show tables, since it can potentially end up doing additional work. I am pretty worried that modifying the variable was considered as a potential solution. The description of that very setting describes how it affects how the system handles IS queries. An administrator should be able to trust that installing something like PMA is not going to change any system variables that could otherwise affect the operation of the system
Thank you for your consideration.
Regards, Martyn
Thanks Martyn for your report. We'll discuss this issue before taking a decision about 4.1.0-rc1.
Sadly, 4.1.0 will be released with some bugs in it, but we should avoid regressions. So, in your opinion, is the current state of 4.1.0-beta worse than 4.0.9, regarding this IS issue?