Forcing query plans

[Robert Wille]

I have a number of very common queries that the optimizer plans a very inefficient plan for. I am using postgres 7.2.3. I vacuum hourly. I'm wonderingwhat I can do to make the queries faster.

Here are the relevant tables:

create table image(
imageid integer not null, /* The image's ID */
containerid integer not null, /* The container that owns it */
name varchar(120) not null, /* Its name */
state bigint not null default 0, /* Its state */
primary key (imageid),
unique (containerid, name) /* All images in a container must be uniquely named */
);

create table ancestry(
containerid integer not null, /* The container that has an ancestor*/
ancestorid integer not null, /* The ancestor of the container */
unique (containerid, ancestorid),
unique (ancestorid, containerid)
);

I have somewhere around 3M rows in the image table, and 37K rows in the ancestry table. The following is representative of some of the common queries I issue:

select * from image natural join ancestry where ancestorid=1000 000 and (state & 7::bigint) = 0::bigint;

When I ask postgres to EXPLAIN it, I get the following:

Merge Join (cost=81858.22. .81900.60 rows=124 width=49)
-> Sort (cost=81693.15. .81693.15 rows=16288 width=41)
-> Seq Scan on image (cost=0.00..802 79.17 rows=16288 width=41)
-> Sort (cost=165.06..1 65.06 rows=45 width=8)
-> Index Scan using ancestry_ancest orid_key on ancestry (cost=0..00..16 3.83 rows=45 width=8)

It appears to me that the query executes as follows:

1. Scan every row in the image table to find those where (state & 7::bigint) = 0::bigint
2. Sort the results
3. Use an index on ancestry to find rows where ancestorid=1000 000
4. Sort the results
5. Join the two

It seems to me that if this query is going to return a small percentage of the rows (which is the common case), it could be done much faster by first joining (all columns involved in the join are indexed), and then by applying the (state & 7::bigint) = 0::bigint constraint to the results. I realize that the query planner is going to have a difficult time estimating the number of rows returned by the bit operator. However, I'd be happy forcing it to always perform the join first, and then apply the state constraint to the results.

Similarly, when I update, I get the following:

explain update image set state=0 from ancestry where ancestorid=1000 000and ancestry.contai nerid=image.con tainerid and (state & 7::bigint) = 0::bigint;

NOTICE: QUERY PLAN:

Merge Join (cost=81841.92. .81884.30 rows=124 width=43)
-> Sort (cost=81676.74. .81676.74 rows=16288 width=39)
-> Seq Scan on image (cost=0.00..802 79.17 rows=16288 width=39)
-> Sort (cost=165.19..1 65.19 rows=45 width=4)
-> Index Scan using ancestry_ancest orid_key on ancestry (cost=0..00..16 3.95 rows=45 width=4)

Is there any way to give the planner a hint, or reword the query and updateso that it executes the way I want?

Thanks in advance.

Robert Wille