max_fsm_pages

Hi everybody,

I'm trying to understand how vacuumdb and free space map work together for retrieving free spaces after delete or update statements.
max_fsm_pages and max_fsm_relations are set to default values (respectively 10000 and 100).
I do my tests on a table containing at the beginning 1931968 tuples stocked in 11944 pages :
INFO: Pages 11944: Changed 11944, Empty 0; Tup 1931968: Vac 0, Keep 0, UnUsed 0

I delete 1931936 tuples (letting 32 tuples left).
I thought the number of pages affected by this delete will overflow fsm storage capacity, but when I run a vacuumdb verbose analyze on this table,
I have the following log information :
INFO: Removed 1397914 tuples in 8641 pages.

CPU 0.76s/0.89u sec elapsed 25.72 sec.

INFO: Removed 534022 tuples in 3303 pages.

CPU 0.37s/0.29u sec elapsed 7.62 sec.

INFO: Pages 11944: Changed 11944, Empty 0; Tup 32: Vac 1931936, Keep 0, UnUsed 0.

Total CPU 1.57s/1.90u sec elapsed 42.01 sec.

It seems that FSM traced all of my deletes since it mentions removes in 8641+3303=11944 pages (>10000 which was set for max_fsm_page).

Can someone explain me, where I'm wrong ?

My first aim was to know if max_fsm_pages was set for each table or for a whole database. Per table I suppose, but Is it a good supposition ?
Where is stored (on disk) this fsm ?

Thanks for your answers.

Luc