So, I have some observations. Is this what you are seeing as well?
So when we CLUSTER a table heavily-updated table:
CLUSTER does appear to reset *n_dead_tup*, *n_tup_ins*, *n_tup_del*,
*n_tup_hot_upd*, but NOT *n_live_tup*
pg_stat_reset() truly clears out all the statistics counters. I tried this
because *n_live_tup* is not correct.
A subsequent ANALYZE will update *n_dead_tup* and *n_live_tup* to some
values that could not possibly be based on the newly CLUSTERed table
So, how to get correct statistics for a heavily updated table? In my
experience, we only need to get the exponent correct, but we're not even
getting that.
BTW, I've upgraded to 8.4.15.
--
View this message in context:
http://postgresql.1045698.n5.nabble.com/BUG-7853-Incorrect-statistics-in-table-with-many-dead-rows-tp5743845p5746602.html
Sent from the PostgreSQL - bugs mailing list archive at Nabble.com.