Here is something that seems anomalous to me: when I set a boolean field
to FALSE performance is much better than when I set it to TRUE. Any
reason for FALSE to be favored over TRUE?
Some details:
vacuum analyze my_table;
update my_table set is_foo=FALSE where some_id = 47;
--142 rows affected, 8047 ms execution time.
vacuum analyze my_table;
update my_table set is_foo=TRUE where some_id = 47;
--142 rows affected, 48609 ms execution time.
I have run these kinds of queries repeatedly and the timing above is
representative--the setting to FALSE case is about 6 times more
performant. The table my_table has about 105K rows and has many other
columns of various types. Thre is a trigger on the table, but it does
not do anything special based on this column's value. The some_id column
is indexed. This is on PG 8.1.3 on Linux.
George