* Qi Huang (huangqiyx@hotmail.com) wrote:
> Thanks, guys, for your hot discussion. I'll explore the ANALYZE command first and try make SYSTEM sampling work.
Otherparts, I'll look at them later.
That sounds like the right first steps to me. Reviewing this
discussion, it was my thought to create a new node, ala seqscan, which
implemented analyze's algorithm for scanning the table. The eventual
idea being that analyze would actually use it in the future. There was
mention up-thread about just calling the analyze code. Personally, I'd
rather we make analyze more SQL like (once we have this functionality
implemented) than make things which are supposed to be SQL call out into
analyze bits.
Thoughts? Other opinions?
Thanks,
Stephen