On Thu, Jan 26, 2006 at 10:13:52AM -0500, John D. Burger wrote:
> I suspect that if a transaction rolls back, any inserts done in the
> interim turn into dead tuples.
Yep:
test=> CREATE TABLE foo (x integer);
CREATE TABLE
test=> BEGIN; INSERT INTO foo SELECT 1 FROM generate_series(1, 10000); ROLLBACK;
BEGIN
INSERT 0 10000
ROLLBACK
test=> INSERT INTO foo SELECT 1 FROM generate_series(1, 20000);
INSERT 0 20000
test=> \x
Expanded display is on.
test=> SELECT * FROM pgstattuple('foo');
-[ RECORD 1 ]------+--------
table_len | 1089536
tuple_count | 20000
tuple_len | 640000
tuple_percent | 58.74
dead_tuple_count | 10000
dead_tuple_len | 320000
dead_tuple_percent | 29.37
free_space | 6872
free_percent | 0.63
test=> VACUUM foo;
VACUUM
test=> SELECT * FROM pgstattuple('foo');
-[ RECORD 1 ]------+--------
table_len | 1089536
tuple_count | 20000
tuple_len | 640000
tuple_percent | 58.74
dead_tuple_count | 0
dead_tuple_len | 0
dead_tuple_percent | 0
free_space | 326692
free_percent | 29.98
--
Michael Fuhr
---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
choose an index scan if your joining column's datatypes do not
match
