Gregory Stark <stark@enterprisedb.com> wrote:
> > 4. ANALYZE finishes in a short time.
> There was a paper with a nice algorithm posted a while back which required
> only constant memory but it depended on scanning the entire table. I think to
> do n_distinct estimates we'll need some statistics which are either gathered
> opportunistically whenever a seqscan happens or maintained by an index.
VACUUM would be another good timing for the alogrithm, because it does
a seqscan. If we do so, we need to separate an analyzing transaction
into sampling and updating-stats transactions to keep vacuums as
ignorable transactions.
However, VACUUM will not do a seqscan when we have Dead Space Map or
Segment Visibility Map. We will need incremental statistics updating
if reliable n_distinct estimation requires many samples.
Regards,
---
ITAGAKI Takahiro
NTT Open Source Software Center
