Here is a patch discussed in
http://archives.postgresql.org/pgsql-hackers/2007-02/msg00010.php
Concurrent vacuum will save n_dead_tuples value at the beginning.
Stats collector will be subtract the value from n_dead_tuples
instead of setting it to zero. The statistics accuracy of n_dead_tuples
will be better, especially just after finish of a vacuum.
The behavior in VACUUM FULL is not changed because concurrent updates
are not allowed during VACUUM FULL.
Comments welcome.
Regards,
---
ITAGAKI Takahiro
NTT Open Source Software Center