Working on large upgrade of MySQL 4.1 running Innodb to MySQL 5.0 and doing oprofile analyzes we found very interesting issue of buf_get_latched_pages_number being responsible for most CPU usage. It did not look right.
The close look revealed this is the function which is used to compute number of latched pages in Innodb Buffer Pool, which is called when you run SHOW INNODB STATUS and SHOW STATUS. In this case for same of monitoring SHOW GLOBAL STATUS was run every couple of seconds being considered fast operation, as it is in MySQL 4.1
But why buf_get_latched_pages_number ? Because this function traverses whole buffer pool to compute number of latched pages, which is quite slow with buffer pools containing millions of pages. But even worse this is done while holding global buffer pool mutex.
Another …
[Read more]