The cache statistics currently do not differentiate between dirty leaf and dirty internal pages in the cache.

The current relevant statistics are:

cache_pages_dirty - tracked dirty pages in the cache
cache_bytes_leaf - tracked bytes belonging to internal pages in the cache
cache_bytes_internal - tracked bytes belonging to leaf pages in the cache

t2 then derives the dirty fill ratio metric using:

tracked dirty bytes in the cache/maximum bytes configured

This is often used to figure out whether application threads are evicting or if the dirty bytes in the cache is the cause of it. However, this is incorrect as the dirty trigger is based on the dirty leaf pages alone, not the total of dirty pages.

We should then add a new derivation for the dirty leaf pages:

tracked dirty leaf bytes in the cache/maximum bytes configured