Tom Lane wrote:
>
> Further info:
>
> I think this behavior may be state-dependent. The first update after
> loading the table rewrites all the blocks, but subsequent ones do not.
> Is it just a matter of marking recently-written tuples as confirmed
> good? If so, then there is not a problem (or at least, not as big
> a problem as I thought).
Exactly!
One have to run vacuum after loading a big amount of data.
Vadim
