# mysql -A
Welcome to the MySQL monitor. Commands end with ; or \g.
Your MySQL connection id is 2817 to server version: 5.0.51b-log

Type ‘help;’ or ‘\h’ for help. Type ‘\c’ to clear the buffer.

mysql> create database foobar;
Query OK, 1 row affected (0.00 sec)

mysql> use foobar
Database changed
mysql> create table bbb SELECT * FROM aaa;
ERROR 2013 (HY000): Lost connection to MySQL server during query

Here is the patch information:
http://www.mysqlperformanceblog.com/2007/10/31/new-patch-for-mysql-performance/

We only track IO for Innodb – for MyISAM it is kind of impossible to split logical IO from OS cache from physical IO from the media unless you know a way to check if logical pread required physical IO inside of OS

Obviously as every extra code this will have some impact on performance, however these are mostly simple counters, so I would call it marginal. Though I have no chance to test it in the live environment such as you mention. As for the memory requirements these can be more significant with the numbers you give, looking briefly at the code the cost would probably not exceed 100 bytes per entry (for any of user, table or index statistics). However you can have the statis aggregated somewhere else once in a while and then flushed which erases all the information from memory.

I may be biased, but I could not do without this feature. It makes it very easy to determine which users create the most load on the server and how that load changes over time. Once a load problem is isolated to one user, statement sampling can be used to determine the statements that may be causing the problem. I have been able to significantly reduce load on important servers with this by determining the statements that should be changed.

I think Percona published a patch that tracks IO by statement (can you reference that patch again). I hope to add a column for IO per user to SHOW USER_STATISTICS. I think that the patch left out FLUSH TABLE_STATISTICS, but that will be in a future patch.

What about servers with 64k+ users (with one database per user), about 8k+ users active (sending at least one query per day). I’m asking about performance penalty and extra memory usage.