Was partitioned table row estimation fixed in 9.0?
There were a flurry of patches around this from Stark and Aster Data, so
I'm checking if I should be testing on 9.0 or adding this to the TODO list.
The problem I'm grappling with is that OUTER JOINS against the master in
a partitioned table (joining to the append node) gives a row estimate
which does not take into account any CE applicable to the partitioned
table. The CE is properly executed when the join is actually executed,
but with the row estimate way off (orders of magnitude) you often get
bad plans for other joins in the query.
This seems fixable for 9.1 if it hasn't been done already.
--Josh Berkus
Josh Berkus <josh@agliodbs.com> writes:
There were a flurry of patches around this from Stark and Aster Data, so
I'm checking if I should be testing on 9.0 or adding this to the TODO list.
The problem I'm grappling with is that OUTER JOINS against the master in
a partitioned table (joining to the append node) gives a row estimate
which does not take into account any CE applicable to the partitioned
table. The CE is properly executed when the join is actually executed,
but with the row estimate way off (orders of magnitude) you often get
bad plans for other joins in the query.
Would you give a concrete example? There was some work done in this
area but it's impossible to tell whether it solves your problem with
only that much detail.
regards, tom lane