Delete all records NOT referenced by Foreign Keys

[D. Dante Lorenso]

I'd like to run a clean up command on my tables to
eliminate rows that I'm no longer using in the database.

I want to do something like this:

DELETE FROM tablename
WHERE IS_REFERENCED_B Y_FOREIGN_KEY IS FALSE;

Does anyone know how something like this could be done
in PostgreSQL? I know I can search all the tables that
I know refer to this table and see if my primary key
exists, but I want a solution that does not require me to
rewrite my code every time a new foreign key constraint
is added to the database.

There must be a way to ask PostgreSQL for a reference count
on a given row or something.

Thanks,

Dante

---------
D. Dante Lorenso
da***@lorenso.c om


---------------------------(end of broadcast)---------------------------
TIP 8: explain analyze is your friend

[Bruno Wolff III]

On Sat, Dec 13, 2003 at 02:20:15 -0600,
"D. Dante Lorenso" <da***@lorenso. com> wrote:

I'd like to run a clean up command on my tables to
eliminate rows that I'm no longer using in the database.

I want to do something like this:

DELETE FROM tablename
WHERE IS_REFERENCED_B Y_FOREIGN_KEY IS FALSE;

Does anyone know how something like this could be done
in PostgreSQL? I know I can search all the tables that
I know refer to this table and see if my primary key
exists, but I want a solution that does not require me to
rewrite my code every time a new foreign key constraint
is added to the database.

There must be a way to ask PostgreSQL for a reference count
on a given row or something.

If you are more concerned about flexibility than speed you can do something
like the following:

Set all of your foreign key references to the desired table to use an
on delete restrict clause.

Have your application read all of the key values from the desired table
and for each key issue a delete of that key in its own transaction.
This will fail for keys that are referenced (because of the restrict clause).

A more complicated, less future proof, but more efficient approach would
be to have your application find out which tables have references to the
table of interest by looking at the system catalog and then write a
delete query using appropiate where not exist clauses.

---------------------------(end of broadcast)---------------------------
TIP 7: don't forget to increase your free space map settings

[D. Dante Lorenso]

Bruno Wolff III wrote:

On Sat, Dec 13, 2003 at 02:20:15 -0600,
"D. Dante Lorenso" <da***@lorenso. com> wrote:

I'd like to run a clean up command on my tables to
eliminate rows that I'm no longer using in the database.

I want to do something like this:

DELETE FROM tablename
WHERE IS_REFERENCED_B Y_FOREIGN_KEY IS FALSE;

Does anyone know how something like this could be done
in PostgreSQL? I know I can search all the tables that
I know refer to this table and see if my primary key
exists, but I want a solution that does not require me to
rewrite my code every time a new foreign key constraint
is added to the database.

There must be a way to ask PostgreSQL for a reference count
on a given row or something.
If you are more concerned about flexibility than speed you can do something
like the following:

Set all of your foreign key references to the desired table to use an
on delete restrict clause.

Have your application read all of the key values from the desired table
and for each key issue a delete of that key in its own transaction.
This will fail for keys that are referenced (because of the restrict clause).

This is something very ugly indeed and is what I'll have to resort to unless
I can find something cleaner. Ideally, I would be able to run this cleanup
on a subset of the table data after an insert into the table. I would like
the query to be fast, though.

Does anyone know if there is any way to say something like:

DELETE FROM tablename
IGNORE ERRORS;

Where a delete that is possible is performed but ones that throw referencial
integrity voilations would silently fail without abandoning the entire
transaction?

I have the 'on delete restrict' clause on my foreign keys already.
A more complicated, less future proof, but more efficient approach would
be to have your application find out which tables have references to the
table of interest by looking at the system catalog and then write a
delete query using appropiate where not exist clauses.

---------------------------(end of broadcast)---------------------------
TIP 7: don't forget to increase your free space map settings

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

[Martijn van Oosterhout]

On Sat, Dec 13, 2003 at 09:48:16PM -0600, D. Dante Lorenso wrote:

This is something very ugly indeed and is what I'll have to resort to unless
I can find something cleaner. Ideally, I would be able to run this cleanup
on a subset of the table data after an insert into the table. I would like
the query to be fast, though.
What about just:

delete from a where a.id not in (select id from b);

or the equivalent exists query.
--
Martijn van Oosterhout <kl*****@svana. org> http://svana.org/kleptog/ (... have gone from d-i being barely usable even by its developers
anywhere, to being about 20% done. Sweet. And the last 80% usually takes
20% of the time, too, right?) -- Anthony Towns, debian-devel-announce

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE/3AB0Y5Twig3Ge+Y RAtAlAJ9u8mCreN Ipr/Tk69PINKTbHdE+a wCfVuoC
r9kWIkzslXKFGiI kBeuj0iA=
=7P25
-----END PGP SIGNATURE-----

[D. Dante Lorenso]

Martijn van Oosterhout wrote:

On Sat, Dec 13, 2003 at 09:48:16PM -0600, D. Dante Lorenso wrote:

This is something very ugly indeed and is what I'll have to resort to unless
I can find something cleaner. Ideally, I would be able to run this cleanup
on a subset of the table data after an insert into the table. I would like
the query to be fast, though.

What about just:

delete from a where a.id not in (select id from b);

or the equivalent exists query.

You missed the previous part of the thread. I have N tables that
have a foreign key to the table in question. Tomorrow there may be
more or fewer foreign key references. Without having to know which
tables have foreign keys on my table, I want to delete all rows
that are not used by any any other table.

PG already can block a delete when it knows that foreign key exists, so
why can't I perform a query that says...

DELETE FROM tablename
WHERE FOREIGN_KEY_EXI STS(oid) IS FALSE;

You see? Something like what I seek never requires ME the developer or
DBA to know about foreign key relationships because I know that PostgreSQL
already does.

To NOT have this functionality does not cause problems, but it does cause
me to waste disk space on rows that are no longer in use. I just want to
do some automated cleanup on tables and just leave that process running
in a crontab nightly or something. I don't want to have to re-write the
cleanup process every time a new dependency is introduced or removed.

I think Bruno had a good idea about using the system tables to determine
relationships, but how to do that is beyond my PostgreSQL expertise at
the moment. I just think there's gotta be an easier way, though...someth ing
like what I describe above.

Dante

----------
D. Dante Lorenso
da***@lorenso.c om

---------------------------(end of broadcast)---------------------------
TIP 1: subscribe and unsubscribe commands go to ma*******@postg resql.org

[Stephan Szabo]

On Sun, 14 Dec 2003, D. Dante Lorenso wrote:

PG already can block a delete when it knows that foreign key exists, so
why can't I perform a query that says...

DELETE FROM tablename
WHERE FOREIGN_KEY_EXI STS(oid) IS FALSE;

That's fairly different from the checks that are performed for the foreign
keys which happen after the action has happened and errors to prevent the
action from being visible. The where clause happens long before that. If
the above has to check each referencing table for matching rows for each
row in tablename, I'd also expect it to perform poorly.

---------------------------(end of broadcast)---------------------------
TIP 1: subscribe and unsubscribe commands go to ma*******@postg resql.org

[Greg Stark]

"D. Dante Lorenso" <da***@lorenso. com> writes:

To NOT have this functionality does not cause problems, but it does cause
me to waste disk space on rows that are no longer in use. I just want to
do some automated cleanup on tables and just leave that process running
in a crontab nightly or something. I don't want to have to re-write the
cleanup process every time a new dependency is introduced or removed.

You could just try to delete every record in the desired tables and see if you
get a foreign key violation. You would have to do each delete in a separate
transaction and just ignore any errors.

so you would have to do something like

foreach table in <list of "auto-cleanup" tables>
select id from table
delete from table where id = xx

This would mean your auto-cleanup crontab script doesn't even have to look in
the system catalog to find out the dependencies. It just depends on postgres
knowing all the dependencies and checking them all.

Inevitably though there will be some tables that have some implicit
depenencies that cannot be represented as foreign key references. Or are just
the master records and don't need anything else in the database to depend on
them. So you'll need a list somewhere of tables that are purely subservient to
other tables and can be summarily cleaned up this way.
--
greg
---------------------------(end of broadcast)---------------------------
TIP 3: if posting/reading through Usenet, please send an appropriate
subscribe-nomail command to ma*******@postg resql.org so that your
message can get through to the mailing list cleanly

[D. Dante Lorenso]

Stephan Szabo wrote:

On Sun, 14 Dec 2003, D. Dante Lorenso wrote:

PG already can block a delete when it knows that foreign key exists, so
why can't I perform a query that says...

DELETE FROM tablename
WHERE FOREIGN_KEY_EXI STS(oid) IS FALSE;

That's fairly different from the checks that are performed for the foreign
keys which happen after the action has happened and errors to prevent the
action from being visible. The where clause happens long before that. If
the above has to check each referencing table for matching rows for each
row in tablename, I'd also expect it to perform poorly.

My hope was that there was some sort of (semaphore? / counter?) associated
with each row that indicated whether a dependency existed at all. Although
that would most likely not be an indexed column, I could apply additional
WHERE constraints to avoid a full table scan.

If such a counter existed, it would certainly speed up deletes when no
foreign
key checks were necessary. But I suppose it would also slow down
deletes when
it becomes necessary to decrement a counter for the dependencies created
by the
row being deleted.

So...internally , how does a row KNOW that it can not be deleted because of
a foreign key constraint? Whatever that mechanism is, isn't there a way for
me to make use of that as I try my query?

And... if not...how about something like this in PL/SQL:

-- Find all the address records which might need to be removed...
FOR my_rec IN
SELECT addr_id
FROM address
WHERE acct_id = in_acct_id
AND addr_is_active IS FALSE
LOOP

-- try to delete this record...
DELETE FROM address
WHERE addr_id = my_rec.addr_id;

END LOOP;

This will loop through the records one at a time and try to delete them.
However, I don't want to have any exceptions thrown if the DELETE action
can not be performed. Is there a TRY/CATCH type of code that I can
surround the DELETE with to prevent the entire operation from being
aborted on the first error found?

Dante

----------
D. Dante Lorenso
da***@lorenso.c om

---------------------------(end of broadcast)---------------------------
TIP 3: if posting/reading through Usenet, please send an appropriate
subscribe-nomail command to ma*******@postg resql.org so that your
message can get through to the mailing list cleanly

[Martijn van Oosterhout]

On Sun, Dec 14, 2003 at 03:02:49AM -0600, D. Dante Lorenso wrote:

My hope was that there was some sort of (semaphore? / counter?) associated
with each row that indicated whether a dependency existed at all. Although
that would most likely not be an indexed column, I could apply additional
WHERE constraints to avoid a full table scan.
Ah, I see. There is no counter. When you delete a row, it does a check on
the referencing table to see if it would break any foreign keys. The system
has a defined trigger for that purpose. If you don't want to have the
trigger error out, you do the same test. That's what my query did.

In fact, you'll be able to do it more efficiently, since you know you'll be
deleting many rows, you can arrange to only scan the referencing table once.

The table with the foreign keys in it would be pg_constraint I think. You
can use -E on psql to see you \d gets the info.
This will loop through the records one at a time and try to delete them.
However, I don't want to have any exceptions thrown if the DELETE action
can not be performed. Is there a TRY/CATCH type of code that I can
surround the DELETE with to prevent the entire operation from being
aborted on the first error found?
That would be subtransactions , and they're not done yet.
--
Martijn van Oosterhout <kl*****@svana. org> http://svana.org/kleptog/ (... have gone from d-i being barely usable even by its developers
anywhere, to being about 20% done. Sweet. And the last 80% usually takes
20% of the time, too, right?) -- Anthony Towns, debian-devel-announce

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE/3C6aY5Twig3Ge+Y RAq1LAKCGEBx300 VIoBJ5vLNxBMGj/HYBWACgiiJo
hrxdaBPPLNwM9Kp ZnEs3P1o=
=K5PU
-----END PGP SIGNATURE-----

[greg]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Does anyone know how something like this could be done
in PostgreSQL? I know I can search all the tables that
I know refer to this table and see if my primary key
exists, but I want a solution that does not require me to
rewrite my code every time a new foreign key constraint
is added to the database.

Here is a function that removes all non-referenced rows from
a table. Make a backup before using of course. :)

CREATE OR REPLACE FUNCTION delete_nonrefs( TEXT) RETURNS TEXT AS '
DECLARE
mytable ALIAS FOR $1;
mytext TEXT;
myrec RECORD;
deltext TEXT;
myrows INTEGER;
BEGIN

mytext := ''
SELECT
c2.relname AS tname,
SUBSTRING((SELE CT pg_get_constrai ntdef(r.oid)) FROM \'\'\\\\\\\\)[^(]+.([^)]+)\'\') AS fkcol,
SUBSTRING((SELE CT pg_get_constrai ntdef(r.oid)) FROM \'\'\\\\\\\\(([^)]+)\'\') AS mycol
FROM pg_class c, pg_class c2, pg_constraint r
WHERE c.relname = \' || quote_literal(m ytable) || \'
AND r.confrelid = c.oid
AND r.contype = \'\'f\'\'
AND c2.oid = r.conrelid'';

FOR myrec IN EXECUTE mytext LOOP
IF deltext IS NULL THEN deltext := \' \\nWHERE \';
ELSE deltext := deltext || \'\\nAND \';
END IF;
deltext := deltext || \'NOT EXISTS \' ||
\'(SELECT 1 FROM \' || quote_ident(myr ec.tname) || \' t WHERE \' ||
quote_ident(myt able) || \'.\' || myrec.fkcol || \' = t.\' || myrec.mycol || \')\';
END LOOP;

IF deltext IS NULL THEN
RETURN \'Table \' || quote_ident(myt able) || \' is not referenced by any other tables\';
END IF;

deltext := \'DELETE FROM \' || quote_ident(myt able) || deltext;
EXECUTE deltext;
GET DIAGNOSTICS myrows := ROW_COUNT;
RETURN \'Rows deleted from table \' || quote_ident(myt able) || \': \' || myrows;

END;
' LANGUAGE plpgsql STABLE STRICT;

- --
Greg Sabino Mullane gr**@turnstep.c om
PGP Key: 0x14964AC8 200312141306

-----BEGIN PGP SIGNATURE-----

iD8DBQE/3KeCvJuQZxSWSsg RAnNwAJ4v0bh/ATZtTaPqqid43qZ uaFB/0ACdG+GL
m6AtGBa3tNKsoZm y1ir6/KY=
=S39B
-----END PGP SIGNATURE-----

---------------------------(end of broadcast)---------------------------
TIP 7: don't forget to increase your free space map settings