Comments on: The confusion over global and session status http://ronaldbradford.com/blog/the-confusion-over-global-and-session-status-2009-07-03/ Expert times and information on MySQL Fri, 12 Mar 2010 19:35:19 -0400 http://wordpress.org/?v=2.8.4 hourly 1 By: Arjen Lentz http://ronaldbradford.com/blog/the-confusion-over-global-and-session-status-2009-07-03/comment-page-1/#comment-592 Arjen Lentz Fri, 03 Jul 2009 22:05:51 +0000 http://ronaldbradford.com/blog/?p=1714#comment-592 What Mark says. The microslow patch is more effective in finding tmp table use, and it can distinguish between mem and disk also. Diddums for sorts. What Mark says.
The microslow patch is more effective in finding tmp table use, and it can distinguish between mem and disk also.
Diddums for sorts.

]]> By: Mark Leith http://ronaldbradford.com/blog/the-confusion-over-global-and-session-status-2009-07-03/comment-page-1/#comment-593 Mark Leith Fri, 03 Jul 2009 18:43:13 +0000 http://ronaldbradford.com/blog/?p=1714#comment-593 Hey Ronald, Yes, previously they incremented both GLOBAL and SESSION outputs - they were fixed as part of http://bugs.mysql.com/bug.php?id=10210 and http://bugs.mysql.com/bug.php?id=29153. It's not just the tmp tables either, it's also things like Handler_write, and Handler_read_rnd_next. The reason? SHOW (and INFORMATION_SCHEMA) create temporary tables in the background, fill them with rows, and then select * from them to show the output to the user. For debugging as you describe, you do not want SESSION based counters to increment on top of what you are trying to debug - whilst at the GLOBAL level you want to know all actions that the server was doing. The fix actually makes it slightly worse to discount any monitoring action statistics as well (you can not take a monitoring session's SESSION STATUS and subtract it from a GLOBAL STATUS for the thread local variables). Cheers, Mark Hey Ronald,

Yes, previously they incremented both GLOBAL and SESSION outputs – they were fixed as part of http://bugs.mysql.com/bug.php?id=10210 and http://bugs.mysql.com/bug.php?id=29153.

It’s not just the tmp tables either, it’s also things like Handler_write, and Handler_read_rnd_next. The reason?

SHOW (and INFORMATION_SCHEMA) create temporary tables in the background, fill them with rows, and then select * from them to show the output to the user.

For debugging as you describe, you do not want SESSION based counters to increment on top of what you are trying to debug – whilst at the GLOBAL level you want to know all actions that the server was doing.

The fix actually makes it slightly worse to discount any monitoring action statistics as well (you can not take a monitoring session’s SESSION STATUS and subtract it from a GLOBAL STATUS for the thread local variables).

Cheers,

Mark

]]>