Re: questions about a table's row estimates

On Feb 10, 2010, at 10:28 PM, Greg Smith wrote:

> Ben Chobot wrote:
>> I'm looking at pg_stat_user_tables in 8.4.2, and I'm confused about n_live_tup. Shouldn't that be at least fairly close to (n_tup_ins - n_tup-del)? It doesn't seem to be, but I'm unclear why.
>>
> Insert 2000 tuples.
> Delete 1000 tuples.
> vacuum
> Insert 1000 tuples. These go into the free space the deleted tuples used to be in.
> analyze
>
> n_tup_ins=3000
> n_tup_del=1000
> n_live_tup=3000

Say what? It's not documented terrible well (http://www.postgresql.org/docs/8.4/interactive/monitoring-stats.html is the best I can find) but I thought n_live_tup was basically the number of tuples visible to a new transaction. If my assumption is wrong, that might explain things.

And unfortunately, Tom, we're not resetting stats counters. :(