Thread: Foreign Key work for 7.3+
Here are the things I'd like to get feedback on doing to the foreign key constraint triggers and support code. It's relatively high level since I'm at the beginning stages of looking at the code that'd be involved. This might therefore be a bit premature, but I figured I'd see if anyone saw anything obviously wrong and there were a few places where I'm uncertain what behavior is better or correct so I'm looking for advice. This isn't necessarily a list of things for 7.3 since I don't expect that I'd be able to finish all of them unless 7.3's devel goes on like 7.2's. * Rewrite foreign key triggers to remove dependencies on SPIUse the index and heap scanning functions to check the table.Thisshould let us remove the ReferentialIntegritySnapshotOverride globaland give us more control over what the triggersare doing by doingthe checks we need directly. My current test code duplicates a bunchof work (finding an index toscan if possible, etc) that other areasof the code have to be doing, but that's mostly a matter of findingwhat I can usefrom other sections. Hopefully this will also eventuallyallow us to move away from the current locking method (althoughI haven'tlooked at it yet) * Fix remaining visibility problemsThere are a couple of places where we have problems with this.For pk updates and deleteswith no action we need to make surethat there wasn't a row updated or inserted with the old key valuesthat would satisfythe constraint. For fk inserts/updates we needto make sure that we're not seeing an intermediate state of therow (ieit was changed again before the check was done). Deferredcascades have a few possible broken cases (I'm not sure if theseareactual errors or not, but they seem like it -- see questions below).Simply checking that the row we're checking isstill valid on the fkinserts/updates causes some similar behavior. * Switch over to oids and attnos for new constraintsMake the real functions for the constraints take oids and attnos.We'llneed to either hack around when we see the create constrainttrigger to use the new form or keep functions bythe old names/oids whichtake names, but make those effectively wrappers around the oid versions(look up the numbers andpass those in to the new ones). In any casenew constraints will use the oids and attnos version. * Make checking an entire table faster for Alter TableRight now alter table runs the insert/update trigger on the fk tableonceper existing fk row on alter table. If we don't care about thepossible disk space usage, we can make this fasterby going through bothtables in sorted key order (the disk space would be if we had to goto an explicit sort). Theother option is to only do this form if wehad an index on each table we could use to scan in order. * Change dumps to dump fk constraints as alter table statements by defaultMake dumps write out alter table statements atthe end rather thancreate constraint triggers. We may wish to provide an option to allowthe older style create constrainttrigger dumps for speed. * Extend fk constraints to work with inheritanceMake fk constraints inherit properly with both fk and pk base tables.Thiswill probably mean making the appropriate triggers on the childtables involved as well as scanning the additionaltables when checks andchanges are needed. For right now, I'd say we'd want to require that thechild tables at leastalso have unique constraints across the key. * Fixes on unique index checks for fk constraint checkingRight now I believe unique partial and functional indexes are consideredsufficientto be targets of an fk constraint. I think this is usuallywrong for partial indexes and that this maybe wrong for functionalindexes. Any thoughts? * Make a central place for storing fk constraint info?Keep track of fk constraint info in a centralized place keeping a moreuseableform of the tables and columns involved as well as the misc. data(match type, deferrability). We'd probably alsowant to keep referencesto the triggers involved in the constraint. This would make some of thethings (dumping fk constraintsas alter table, inheritance) easier Ibelieve since we wouldn't have to match up constraint triggers todeterminethe details of the constraint, but it would probably require ahack for checking create constraint trigger statementsto see if they'refk constraints. * Match partialMatch partial is rather complicated and could be even more so dependingon the answers to the behavioral questionsin the notes below, but it'dbe good to do it if we can. Behavioral questions: * Is this current behavior correct or should fk contain a row with thekey value 1 at the end of the sequence? create table pk(a int unique); NOTICE: CREATE TABLE / UNIQUE will create implicit index 'pk_a_key' for table 'pk' NOTICE: CREATE TABLE / UNIQUE will create implicit index 'pk_a_key' for table 'pk' CREATE create table fk(a int references pk(a) on delete cascade initially deferred); NOTICE: CREATE TABLE will create implicit trigger(s) for FOREIGN KEY check(s) NOTICE: CREATE TABLE will create implicit trigger(s) for FOREIGN KEY check(s) CREATE insert into pk values (1); INSERT 16645 1 begin; BEGIN delete from pk where a=1; DELETE 1 insert into pk values (1); INSERT 16646 1 insert into fk values (1); INSERT 16647 1 end; COMMIT select * from pk;a ---1 (1 row) select * from fk;a --- (0 rows) * Using the tables from above. Should the following have no rows in fkor a single row containing 2? delete from fk; DELETE 1 delete from pk; DELETE 1 insert into pk values (1); INSERT 16668 1 insert into fk values (1); INSERT 16669 1 begin; BEGIN delete from pk where a=1; DELETE 1 insert into pk values (2); INSERT 16670 1 update fk set a=2; UPDATE 1 end; COMMIT select * from pk;a ---2 (1 row) select * from fk;a ---2 (1 row)
On Wed, 2002-02-20 at 17:28, Stephan Szabo wrote:
> Here are the things I'd like to get feedback on doing to the foreign
> key constraint triggers and support code. ...
>
> * Extend fk constraints to work with inheritance
> Make fk constraints inherit properly with both fk and pk base tables.
> This will probably mean making the appropriate triggers on the child
> tables involved as well as scanning the additional tables when checks and
> changes are needed. For right now, I'd say we'd want to require that the
> child tables at least also have unique constraints across the key.
This means that we need a solution for unique indexes and primary keys
under inheritance. I wrote about this in a mail which is preserved in
doc/TODO.detail/inheritance.
3. Inheritance of a table implies inheriting all its constraints unless ONLY is used or the constraints are
subsequentlydropped; again, dropping operates through all descendant tables. A primary key, foreign key or unique
constraintcannot be dropped or modified for a descendant. A unique index on a column is shared by all tables below
thetable for which it is declared. It cannot be dropped for any descendant. In other words, only NOT NULL and
CHECKconstraints can be dropped in descendants.
[On reconsidering this, I'm not sure it is right to let these be
modified either.]
In multiple inheritance, a column may inherit multiple unique indices from its several ancestors. All inherited
constraintsmust be satisfied together (though check constraints may be dropped). 4. RI to a table implies the
inclusionof all its descendants in the check. Since a referenced column may be uniquely indexed further up the
hierarchythan in the table named, the check must ensure that the referenced value occurs in the right segment of the
hierarchy. RI to one particular level of the hierarchy, excluding descendants, requires the use of ONLY in the
constraint.
So an index must somehow be made to serve more than one table, and then
an index lookup must also discover whether the key is in the right
segment of the inheritance hierarchy:
table_a (id INTEGER PRIMARY KEY, ...) table_b (..., bfld CHAR UNIQUE) INHERITS (table_a)
table_c1(...) INHERITS (table_b) table_c2 (...) INHERITS (table_b) table_d (...) INHERITS (table_c1,
table_c2)
table_x (a INTEGER REFERENCES table_a(id)) -- looks at whole
-- hierarchy table_y (a INTEGER REFERENCES table_b(id), -- looks at whole
-- hierarchy from -- b down,
-- excluding a b CHAR REFERENCES table_b(bfld), -- looks at whole
-- hierarchy from -- b
downfor bfld c CHAR REFERENCES ONLY table_b(bfld) -- excludes b's )
-- descendants table_z (a INTEGER REFERENCES table_c2(id)) -- looks at c2 and
-- d
Perhaps the answer to this is to create a separate key table for each
unique field, which would act as an indirect index:
table_a_keys_id (id INTEGER PRIMARY KEY, table OID REFERENCES pg_class(oid)
ON UPDATE CASCADE ON DELETE CASCADE)
table_b_keys_bfld(id CHAR PRIMARY KEY, table OID REFERENCES pg_class(oid)
ON UPDATE CASCADE ON DELETE CASCADE)
So the key table would record which table contained each key, and it
would need triggers to update it whenever any rows in the hierarchy
changed. There would not be a unique index on any of the tables
directly, but instead there would be unique indexes on the keys tables.
On the creation of the first descendant table in a hierarchy, the key
tables would be created and filled with any existing keys from the
ancestor table; the existing unique indexes of the ancestor would be
dropped. On dropping the last descendant in a hierarchy, the reverse
would have to be done.
--
Oliver Elphick Oliver.Elphick@lfix.co.uk
Isle of Wight http://www.lfix.co.uk/oliver
GPG: 1024D/3E1D0C1C: CA12 09E0 E8D5 8870 5839 932A 614D 4C34 3E1D 0C1C
"For the Lord himself shall descend from heaven with a shout, with the voice of the archangel, and with the
trump of God; and the dead in Christ shall rise first; Then we which are alive and remain shall be caught up
togetherwith them in the clouds, to meet the Lord in the air; and so shall we ever be with the Lord."
I Thessalonians 4:16,17
On 20 Feb 2002, Oliver Elphick wrote: > On Wed, 2002-02-20 at 17:28, Stephan Szabo wrote: > > Here are the things I'd like to get feedback on doing to the foreign > > key constraint triggers and support code. ... > > > > * Extend fk constraints to work with inheritance > > Make fk constraints inherit properly with both fk and pk base tables. > > This will probably mean making the appropriate triggers on the child > > tables involved as well as scanning the additional tables when checks and > > changes are needed. For right now, I'd say we'd want to require that the > > child tables at least also have unique constraints across the key. > > This means that we need a solution for unique indexes and primary keys > under inheritance. I wrote about this in a mail which is preserved in Yeah, I'm sort of hoping that someone who works more with inheritance will jump on the primary keys thing so that I can utilize it from the foreign key constraints. This one's one of the parts likely to be put off since as much as I'd like to get it done it sounds like it's going to take a lot of work. > doc/TODO.detail/inheritance. > > 3. Inheritance of a table implies inheriting all its constraints > unless ONLY is used or the constraints are subsequently dropped; > again, dropping operates through all descendant tables. A primary > key, foreign key or unique constraint cannot be dropped or modified > for a descendant. A unique index on a column is shared by all > tables below the table for which it is declared. It cannot be > dropped for any descendant. > > In other words, only NOT NULL and CHECK constraints can be dropped > in descendants. > > [On reconsidering this, I'm not sure it is right to let these be > modified either.] I'm not sure either since a non-ONLY select on the parent will show the rows that wouldn't meet the parent's constraints. > 4. RI to a table implies the inclusion of all its descendants in the > check. Since a referenced column may be uniquely indexed further up > the hierarchy than in the table named, the check must ensure that > the referenced value occurs in the right segment of the hierarchy. > RI to one particular level of the hierarchy, excluding descendants, > requires the use of ONLY in the constraint. > > So an index must somehow be made to serve more than one table, and then > an index lookup must also discover whether the key is in the right > segment of the inheritance hierarchy: I'm sort of assuming that a cross table index will contain some reference to the table it came from if only to help provide a way to reference the associate heap tuple row.