WHERE ... NOT NULL ... OR ... (SELECT...)

I have a query, like this:

valipile=# explain select * from account_analytic_line where move_id in
(SELECT id FROM account_move_line);
QUERY PLAN

---------------------------------------------------------------------------------------
Hash Semi Join (cost=60799.74..96694.82 rows=329568 width=162)
Hash Cond: (account_analytic_line.move_id = account_move_line.id)
-> Seq Scan on account_analytic_line (cost=0.00..9620.68
rows=329568 width=162)
-> Hash (cost=41292.66..41292.66 rows=1188966 width=4)
-> Seq Scan on account_move_line (cost=0.00..41292.66
rows=1188966 width=4)
(5 rows)

Which is all fine. However, as move_id in account_analytic_line is
NULLable I want to include that one into my query. But then:

valipile=# explain select * from account_analytic_line where move_id is
null or move_id in (SELECT id FROM account_move_line);
QUERY PLAN

-----------------------------------------------------------------------------------------
Seq Scan on account_analytic_line (cost=0.00..9039221110.12
rows=164784 width=162)
Filter: ((move_id IS NULL) OR (SubPlan 1))
SubPlan 1
-> Materialize (cost=0.00..51882.49 rows=1188966 width=4)
-> Seq Scan on account_move_line (cost=0.00..41292.66
rows=1188966 width=4)
(5 rows)

This, of course, takes forever.

(There are no indexes/constraints/whatever on the tables as I'm deleting
old data from the database)

Now, I did 'circumvent' the waiting with using UNION:

valipile=# explain select * from account_analytic_line where move_id in
(select id from account_move_line) union select * from
account_analytic_line where move_id is null;

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Unique (cost=193891.55..212017.84 rows=329569 width=162)
-> Sort (cost=193891.55..194715.47 rows=329569 width=162)
Sort Key: account_analytic_line.id,
account_analytic_line.create_uid, account_analytic_line.create_date,
account_analytic_line.write_date, account_analytic_line.write_uid,
account_analytic_line.amount, account_analytic_line.user_id, account_analy
-> Append (cost=60799.74..109611.18 rows=329569 width=162)
-> Hash Semi Join (cost=60799.74..96694.82 rows=329568
width=162)
Hash Cond: (account_analytic_line.move_id =
account_move_line.id)
-> Seq Scan on account_analytic_line
(cost=0.00..9620.68 rows=329568 width=162)
-> Hash (cost=41292.66..41292.66 rows=1188966
width=4)
-> Seq Scan on account_move_line
(cost=0.00..41292.66 rows=1188966 width=4)
-> Seq Scan on account_analytic_line
account_analytic_line_1 (cost=0.00..9620.68 rows=1 width=162)
Filter: (move_id IS NULL)
(11 rows)

but I'm curious why postgres chooses such poor query plan for the 'OR
column IS NULL' addition ?

Mario
--
Mario Splivalo
mario(at)splivalo(dot)hr

"I can do it quick, I can do it cheap, I can do it well. Pick any two."