Re: Free space mapping (was Re: Multi-Versions and Vacuum)

Tom Lane wrote:

>It doesn't need a lot of examination in my mind: the cause is surely
>growth of the index on the toast table.
>  
>
You are indeed correct !

A quick check of my test case shows that the toast table growth tails 
off ( provided max_fsm_pages is suitably set), and the unbounded growth 
comes entirely from the toast index.
 
Applying REINDEX brings the space usage down to a level that compares 
with a non-toasted example.

With respect to the REINDEX workaround -  having to re-start your server 
single process for it is a bit fatal if you are  24/7 shop - I think 
altering those high hit tables to have attributes detoasted might be 
better (row length permitting).

Anyway that has *toasted* the need for my currrent little investigation 
(quite enjoyed my romp through the code, so I am not too worried).  I 
still have some free time (I am off work after knee operation) so I'm 
now looking for another project....

regards

Mark