Обсуждение: Re: 7.0 FK trigger question
> Hi, > > due to the limitations in alter table, I generate some SQL to > implement changes to tables. This works along the lines of > > 1. drop fk triggers on old table > 2. rename serial sequences on old table > 3. drop indexes on old table > 4. rename old table > 5. create new table > 6. insert into new table select ... from old table > 7. drop new sequences/rename old sequences > 8. recreate fk triggers 9. drop old table If you do 9. you can skip 1. because that's done automatically. > [...] > > This looks kind-of hairy to drop and recreate correctly. > > I thought an alternative may be to change the oid's in pg_trigger. But I > saw that the oid's of the tables are part of the trigger name. I could > probably recreate the trigger names with different oid's but this looks > like asking for trouble. Not exactly. The OIDs in the trigger names are just ones that CREATE CONSTRAINT TRIGGER allocates itself to giveany of them a unique name. They aren't used anywhere else, so don't care. And BTW: specifying a constraint reallyinvokes these commands internally. > So what is the best solution? It would be great if there would be some > way to drop foreign key triggers and re-instate them. This would also > help with loading data where there are circular dependencies of foreign > keys, as one could drop a trigger to break the loop, load the data, and > re-instate the triggers. Ideally you would use correct ALTER TABLE ... ADD CONSTRAINT commands, which are implemented in 7.0. pg_dump actually does sort of this "disable RI triggers" for data only dumps. You might want to setup a simple test database and take a data only dump to see the mechanism. > > So I guess my question really boils down to: is it possible to write a > function that drops a foreign key trigger or re-instates it? This should > really be ALTER TABLE table ALTER COLUMN column (DROP|CREATE) > CONSTRAINT.... or something along those lines. There's still something missing in ALTER TABLE. DROP CONSTRAINT is one of them, but since your sequencs with renaming the old etc. is the safest possibility anyway, it's not that high priority. Jan -- #======================================================================# # It's easier to get forgiveness for being wrong than for being right. # # Let's break this rule - forgive me. # #========================================= wieck@debis.com (Jan Wieck) #
> Thanks for the reply. For the time begin I've solved this by copying every table in the database to a backup table without any constraints, recreating the tables and copying the data back in. I have to be a bit careful with doing it all in the right order, although I think I can solve this by doing everything in a transaction as the constraints are only checked at the end of transaction? > > > > So I guess my question really boils down to: is it possible to write a > > function that drops a foreign key trigger or re-instates it? This should > > really be ALTER TABLE table ALTER COLUMN column (DROP|CREATE) > > CONSTRAINT.... or something along those lines. > > There's still something missing in ALTER TABLE. DROP > CONSTRAINT is one of them, but since your sequencs with > renaming the old etc. is the safest possibility anyway, it's > not that high priority. OK, I'm definitely not being very bright here, but i cannot get my system to accept the alter column commands. An example on the man pages ,ay help a lot here! I tried test=# create table t (i int4); CREATE test=# create table t1 (k int4); CREATE test=# alter table t1 alter column k add constraint references t(i); ERROR: parser: parse error at or near "add" test=# alter table t1 alter column k constraint references t(i); ERROR: parser: parse error at or near "constraint" test=# alter table t1 alter k constraint references t(i); ERROR: parser: parse error at or near "constraint" test=# alter table t1 alter column k create constraint references t(i); ERROR: parser: parse error at or near "create" So what am I doing wrong? Thanks, Adriaan
> Thanks for the reply. For the time begin I've solved this by copying every > table in the database to a backup table without any constraints, recreating > the tables and copying the data back in. I have to be a bit careful with > doing it all in the right order, although I think I can solve this by doing > everything in a transaction as the constraints are only checked at the end of > transaction? By default, constraints are checked at end of statement. Constraints can be specified DEFERRABLE, then you cando SET CONSTRAINTS ... DEFERRED which will delay them until COMMIT. > OK, I'm definitely not being very bright here, but i cannot get my system to > accept the alter column commands. An example on the man pages ,ay help a lot > here! I tried > > test=# create table t (i int4); > CREATE > test=# create table t1 (k int4); > CREATE > test=# alter table t1 alter column k add constraint references t(i); > ERROR: parser: parse error at or near "add" > test=# alter table t1 alter column k constraint references t(i); > ERROR: parser: parse error at or near "constraint" > test=# alter table t1 alter k constraint references t(i); > ERROR: parser: parse error at or near "constraint" > test=# alter table t1 alter column k create constraint references t(i); > ERROR: parser: parse error at or near "create" > > So what am I doing wrong? alter table t1 add constraint chk_k foreign key (k) references t (i); The referenced column(s) (t.i in your case above) must not be a primary key - any combination is accepted. SQL standard requires that there is a unique index defined for the referenced columns so it is guaranteed that FKsreference to exactly ONE row. Actually Postgres doesn't check or force it, so you have to take care yourself. Forexample: create table t (i integer, j integer); create unique index t_pk_idx_1 on t (i, j); -- DON'T FORGET THIS! createtable t1 (k integer, l integer, foreign key (k, l) references t (i, j)); BTW: all existing data is checked at ALTER TABLE time. And our implementation of FK is based on SQL3. So you can specify match type FULL (PARTIAL will be in 7.1), and referential actions (ON DELETE CASCADE etc.) too. It is nice to define ON UPDATE CASCADE, because if you UPDATEa PK, all referencing FKs will silently follow then. Jan -- #======================================================================# # It's easier to get forgiveness for being wrong than for being right. # # Let's break this rule - forgive me. # #========================================= wieck@debis.com (Jan Wieck) #