[postgresql] CASCADE DELETE just once

I have a Postgresql database on which I want to do a few cascading deletes. However, the tables aren't set up with the ON DELETE CASCADE rule. Is there any way I can perform a delete and tell Postgresql to cascade it just this once? Something equivalent to

DELETE FROM some_table CASCADE;

The answers to this older question make it seem like no such solution exists, but I figured I'd ask this question explicitly just to be sure.

I cannot comment Palehorse's answer so I added my own answer. Palehorse's logic is ok but efficiency can be bad with big data sets.

DELETE FROM some_child_table sct 
 WHERE exists (SELECT FROM some_Table st 
                WHERE sct.some_fk_fiel=st.some_id);

DELETE FROM some_table;

It is faster if you have indexes on columns and data set is bigger than few records.

I wrote a (recursive) function to delete any row based on its primary key. I wrote this because I did not want to create my constraints as "on delete cascade". I wanted to be able to delete complex sets of data (as a DBA) but not allow my programmers to be able to cascade delete without thinking through all of the repercussions. I'm still testing out this function, so there may be bugs in it -- but please don't try it if your DB has multi column primary (and thus foreign) keys. Also, the keys all have to be able to be represented in string form, but it could be written in a way that doesn't have that restriction. I use this function VERY SPARINGLY anyway, I value my data too much to enable the cascading constraints on everything. Basically this function is passed in the schema, table name, and primary value (in string form), and it will start by finding any foreign keys on that table and makes sure data doesn't exist-- if it does, it recursively calls itsself on the found data. It uses an array of data already marked for deletion to prevent infinite loops. Please test it out and let me know how it works for you. Note: It's a little slow. I call it like so: select delete_cascade('public','my_table','1');

create or replace function delete_cascade(p_schema varchar, p_table varchar, p_key varchar, p_recursion varchar[] default null)
 returns integer as $$
declare
    rx record;
    rd record;
    v_sql varchar;
    v_recursion_key varchar;
    recnum integer;
    v_primary_key varchar;
    v_rows integer;
begin
    recnum := 0;
    select ccu.column_name into v_primary_key
        from
        information_schema.table_constraints  tc
        join information_schema.constraint_column_usage AS ccu ON ccu.constraint_name = tc.constraint_name and ccu.constraint_schema=tc.constraint_schema
        and tc.constraint_type='PRIMARY KEY'
        and tc.table_name=p_table
        and tc.table_schema=p_schema;

    for rx in (
        select kcu.table_name as foreign_table_name, 
        kcu.column_name as foreign_column_name, 
        kcu.table_schema foreign_table_schema,
        kcu2.column_name as foreign_table_primary_key
        from information_schema.constraint_column_usage ccu
        join information_schema.table_constraints tc on tc.constraint_name=ccu.constraint_name and tc.constraint_catalog=ccu.constraint_catalog and ccu.constraint_schema=ccu.constraint_schema 
        join information_schema.key_column_usage kcu on kcu.constraint_name=ccu.constraint_name and kcu.constraint_catalog=ccu.constraint_catalog and kcu.constraint_schema=ccu.constraint_schema
        join information_schema.table_constraints tc2 on tc2.table_name=kcu.table_name and tc2.table_schema=kcu.table_schema
        join information_schema.key_column_usage kcu2 on kcu2.constraint_name=tc2.constraint_name and kcu2.constraint_catalog=tc2.constraint_catalog and kcu2.constraint_schema=tc2.constraint_schema
        where ccu.table_name=p_table  and ccu.table_schema=p_schema
        and TC.CONSTRAINT_TYPE='FOREIGN KEY'
        and tc2.constraint_type='PRIMARY KEY'
)
    loop
        v_sql := 'select '||rx.foreign_table_primary_key||' as key from '||rx.foreign_table_schema||'.'||rx.foreign_table_name||'
            where '||rx.foreign_column_name||'='||quote_literal(p_key)||' for update';
        --raise notice '%',v_sql;
        --found a foreign key, now find the primary keys for any data that exists in any of those tables.
        for rd in execute v_sql
        loop
            v_recursion_key=rx.foreign_table_schema||'.'||rx.foreign_table_name||'.'||rx.foreign_column_name||'='||rd.key;
            if (v_recursion_key = any (p_recursion)) then
                --raise notice 'Avoiding infinite loop';
            else
                --raise notice 'Recursing to %,%',rx.foreign_table_name, rd.key;
                recnum:= recnum +delete_cascade(rx.foreign_table_schema::varchar, rx.foreign_table_name::varchar, rd.key::varchar, p_recursion||v_recursion_key);
            end if;
        end loop;
    end loop;
    begin
    --actually delete original record.
    v_sql := 'delete from '||p_schema||'.'||p_table||' where '||v_primary_key||'='||quote_literal(p_key);
    execute v_sql;
    get diagnostics v_rows= row_count;
    --raise notice 'Deleting %.% %=%',p_schema,p_table,v_primary_key,p_key;
    recnum:= recnum +v_rows;
    exception when others then recnum=0;
    end;

    return recnum;
end;
$$
language PLPGSQL;

If you really want DELETE FROM some_table CASCADE; which means "remove all rows from table some_table", you can use TRUNCATE instead of DELETE and CASCADE is always supported. However, if you want to use selective delete with a where clause, TRUNCATE is not good enough.

USE WITH CARE - This will drop all rows of all tables which have a foreign key constraint on some_table and all tables that have constraints on those tables, etc.

Postgres supports CASCADE with TRUNCATE command:

TRUNCATE some_table CASCADE;

Handily this is transactional (i.e. can be rolled back), although it is not fully isolated from other concurrent transactions, and has several other caveats. Read the docs for details.

You can use to automate this, you could define the foreign key constraint with ON DELETE CASCADE.
I quote the the manual of foreign key constraints:

CASCADE specifies that when a referenced row is deleted, row(s) referencing it should be automatically deleted as well.

The delete with the cascade option only applied to tables with foreign keys defined. If you do a delete, and it says you cannot because it would violate the foreign key constraint, the cascade will cause it to delete the offending rows.

If you want to delete associated rows in this way, you will need to define the foreign keys first. Also, remember that unless you explicitly instruct it to begin a transaction, or you change the defaults, it will do an auto-commit, which could be very time consuming to clean up.

If I understand correctly, you should be able to do what you want by dropping the foreign key constraint, adding a new one (which will cascade), doing your stuff, and recreating the restricting foreign key constraint.

For example:

testing=# create table a (id integer primary key);
NOTICE:  CREATE TABLE / PRIMARY KEY will create implicit index "a_pkey" for table "a"
CREATE TABLE
testing=# create table b (id integer references a);
CREATE TABLE

-- put some data in the table
testing=# insert into a values(1);
INSERT 0 1
testing=# insert into a values(2);
INSERT 0 1
testing=# insert into b values(2);
INSERT 0 1
testing=# insert into b values(1);
INSERT 0 1

-- restricting works
testing=# delete from a where id=1;
ERROR:  update or delete on table "a" violates foreign key constraint "b_id_fkey" on table "b"
DETAIL:  Key (id)=(1) is still referenced from table "b".

-- find the name of the constraint
testing=# \d b;
       Table "public.b"
 Column |  Type   | Modifiers 
--------+---------+-----------
 id     | integer | 
Foreign-key constraints:
    "b_id_fkey" FOREIGN KEY (id) REFERENCES a(id)

-- drop the constraint
testing=# alter table b drop constraint b_a_id_fkey;
ALTER TABLE

-- create a cascading one
testing=# alter table b add FOREIGN KEY (id) references a(id) on delete cascade; 
ALTER TABLE

testing=# delete from a where id=1;
DELETE 1
testing=# select * from a;
 id 
----
  2
(1 row)

testing=# select * from b;
 id 
----
  2
(1 row)

-- it works, do your stuff.
-- [stuff]

-- recreate the previous state
testing=# \d b;
       Table "public.b"
 Column |  Type   | Modifiers 
--------+---------+-----------
 id     | integer | 
Foreign-key constraints:
    "b_id_fkey" FOREIGN KEY (id) REFERENCES a(id) ON DELETE CASCADE

testing=# alter table b drop constraint b_id_fkey;
ALTER TABLE
testing=# alter table b add FOREIGN KEY (id) references a(id) on delete restrict; 
ALTER TABLE

Of course, you should abstract stuff like that into a procedure, for the sake of your mental health.

I took Joe Love's answer and rewrote it using the IN operator with sub-selects instead of = to make the function faster (according to Hubbitus's suggestion):

create or replace function delete_cascade(p_schema varchar, p_table varchar, p_keys varchar, p_subquery varchar default null, p_foreign_keys varchar[] default array[]::varchar[])
 returns integer as $$
declare

    rx record;
    rd record;
    v_sql varchar;
    v_subquery varchar;
    v_primary_key varchar;
    v_foreign_key varchar;
    v_rows integer;
    recnum integer;

begin

    recnum := 0;
    select ccu.column_name into v_primary_key
        from
        information_schema.table_constraints  tc
        join information_schema.constraint_column_usage AS ccu ON ccu.constraint_name = tc.constraint_name and ccu.constraint_schema=tc.constraint_schema
        and tc.constraint_type='PRIMARY KEY'
        and tc.table_name=p_table
        and tc.table_schema=p_schema;

    for rx in (
        select kcu.table_name as foreign_table_name, 
        kcu.column_name as foreign_column_name, 
        kcu.table_schema foreign_table_schema,
        kcu2.column_name as foreign_table_primary_key
        from information_schema.constraint_column_usage ccu
        join information_schema.table_constraints tc on tc.constraint_name=ccu.constraint_name and tc.constraint_catalog=ccu.constraint_catalog and ccu.constraint_schema=ccu.constraint_schema 
        join information_schema.key_column_usage kcu on kcu.constraint_name=ccu.constraint_name and kcu.constraint_catalog=ccu.constraint_catalog and kcu.constraint_schema=ccu.constraint_schema
        join information_schema.table_constraints tc2 on tc2.table_name=kcu.table_name and tc2.table_schema=kcu.table_schema
        join information_schema.key_column_usage kcu2 on kcu2.constraint_name=tc2.constraint_name and kcu2.constraint_catalog=tc2.constraint_catalog and kcu2.constraint_schema=tc2.constraint_schema
        where ccu.table_name=p_table  and ccu.table_schema=p_schema
        and TC.CONSTRAINT_TYPE='FOREIGN KEY'
        and tc2.constraint_type='PRIMARY KEY'
)
    loop
        v_foreign_key := rx.foreign_table_schema||'.'||rx.foreign_table_name||'.'||rx.foreign_column_name;
        v_subquery := 'select "'||rx.foreign_table_primary_key||'" as key from '||rx.foreign_table_schema||'."'||rx.foreign_table_name||'"
             where "'||rx.foreign_column_name||'"in('||coalesce(p_keys, p_subquery)||') for update';
        if p_foreign_keys @> ARRAY[v_foreign_key] then
            --raise notice 'circular recursion detected';
        else
            p_foreign_keys := array_append(p_foreign_keys, v_foreign_key);
            recnum:= recnum + delete_cascade(rx.foreign_table_schema, rx.foreign_table_name, null, v_subquery, p_foreign_keys);
            p_foreign_keys := array_remove(p_foreign_keys, v_foreign_key);
        end if;
    end loop;

    begin
        if (coalesce(p_keys, p_subquery) <> '') then
            v_sql := 'delete from '||p_schema||'."'||p_table||'" where "'||v_primary_key||'"in('||coalesce(p_keys, p_subquery)||')';
            --raise notice '%',v_sql;
            execute v_sql;
            get diagnostics v_rows = row_count;
            recnum := recnum + v_rows;
        end if;
        exception when others then recnum=0;
    end;

    return recnum;

end;
$$
language PLPGSQL;

Yeah, as others have said, there's no convenient 'DELETE FROM my_table ... CASCADE' (or equivalent). To delete non-cascading foreign key-protected child records and their referenced ancestors, your options include:

• Perform all the deletions explicitly, one query at a time, starting with child tables (though this won't fly if you've got circular references); or
• Perform all the deletions explicitly in a single (potentially massive) query; or
• Assuming your non-cascading foreign key constraints were created as 'ON DELETE NO ACTION DEFERRABLE', perform all the deletions explicitly in a single transaction; or
• Temporarily drop the 'no action' and 'restrict' foreign key constraints in the graph, recreate them as CASCADE, delete the offending ancestors, drop the foreign key constraints again, and finally recreate them as they were originally (thus temporarily weakening the integrity of your data); or
• Something probably equally fun.

It's on purpose that circumventing foreign key constraints isn't made convenient, I assume; but I do understand why in particular circumstances you'd want to do it. If it's something you'll be doing with some frequency, and if you're willing to flout the wisdom of DBAs everywhere, you may want to automate it with a procedure.

I came here a few months ago looking for an answer to the "CASCADE DELETE just once" question (originally asked over a decade ago!). I got some mileage out of Joe Love's clever solution (and Thomas C. G. de Vilhena's variant), but in the end my use case had particular requirements (handling of intra-table circular references, for one) that forced me to take a different approach. That approach ultimately became recursively_delete (PG 10.10).

I've been using recursively_delete in production for a while, now, and finally feel (warily) confident enough to make it available to others who might wind up here looking for ideas. As with Joe Love's solution, it allows you to delete entire graphs of data as if all foreign key constraints in your database were momentarily set to CASCADE, but offers a couple additional features:

• Provides an ASCII preview of the deletion target and its graph of dependents.
• Performs deletion in a single query using recursive CTEs.
• Handles circular dependencies, intra- and inter-table.
• Handles composite keys.
• Skips 'set default' and 'set null' constraints.