Zoltan Boszormenyi <zb@cybertec.at> writes:
> Tom Lane =EDrta:
>> Zoltan Boszormenyi <zb@cybertec.at> writes:
>>> At the end of the day, the behaviour is the same, isn't it?
>>
>> No, there's a difference in terms of the priority for pushing this
>> column out to toast storage, versus pushing other columns of the row
>> out to toast.
> Thanks very much for clarifying.
> I was thinking of a binary data that wouldn't fit
> into the maximum inline tuple size. In this case
> both MAIN and EXTENDED end up compressed
> and out-of-line. I didn't consider having multiple
> bytea or text columns filled with small amount of data.
It'd be pretty unwise to mark a column MAIN if it's likely to contain
wide values ("wide" meaning more than 1K or so). As you say, it'll
still get toasted --- but not until after everything else in the row has
been toasted, even quite narrow values that happen to be of toastable
types.
regards, tom lane