I have this problem. Given a users
table that consists of users\' username in a social network and friends
table that contain a user\'s name and a
SELECT *
FROM (
SELECT username
FROM friends
START WITH
username = 'myname'
CONNECT BY
friendname = PRIOR username
AND level <= 3
)
WHERE username = 'friendname'
AND rownum = 1
Update the level as necessary: you may search for the third layer friends etc.
If the friendship relationship is symmetric, you should make the following query:
WITH q AS
(
SELECT username, friendname
FROM friends
UNION ALL
SELECT friendname, username
FROM friends
),
f AS
(
SELECT friendname, level
FROM q
START WITH
username = 'Thomas'
CONNECT BY NOCYCLE
username = PRIOR friendname
)
SELECT *
FROM f
WHERE friendname = 'Jo'
AND rownum = 1
This query can be made much faster if you denormalize your table: store two records per friendship, like this:
CREATE TABLE dual_friends (orestes NOT NULL, pylades NOT NULL, CONSTRAINT pk_dualfriends_op PRIMARY KEY (orestes, pylades)) ORGANIZATION INDEX
AS
SELECT username, friendname
FROM friends
UNION ALL
SELECT friendname, username
FROM friends
Then you can just replace the CTE
above with the dual_friends
:
WITH f AS
(
SELECT pylades, level
FROM dual_friends
START WITH
orestes = 'Thomas'
CONNECT BY NOCYCLE
orestes = PRIOR pylades
AND level <= 3
)
SELECT *
FROM f
WHERE pylades = 'Jo'
AND rownum = 1
, which will use the index and be much more efficient, especially if you limit the level to some reasonable value.
You can use connect by