On Sat, Aug 16, 2008 at 1:28 PM, Decibel! <decibel@decibel.org> wrote:
> On Aug 12, 2008, at 3:15 PM, Jeff Gentry wrote:
>>
>> So I've seen the header file where the 1600 column limit is defined
>
>
> IIRC, that limit is directly related to block size in the header, so one
> possible fix is to increase block size. AFAIK anything up to 64K blocks
> should be safe.
Unless something's changed, I'm pretty sure things start breaking
after 32k blocks.
> BTW, keep in mind that if you're storing anything that's a varlena (anything
> that's variable length, including NUMBER) where you have that many columns,
> every single varlena is going to end up toasted. That's bound to have a
> *serious* performance impact.
Yeah, usually you're better off using arrayed types than 1600+ columns.