Question

Given a structure like this:

CREATE TABLE reference_table (
  reference_table_key numeric NOT NULL,
  reference_value numeric,
  CONSTRAINT reference_table_pk PRIMARY KEY (reference_table_key)
);

CREATE TABLE other_table (
  other_table_key numeric NOT NULL,
  reference_table_key numeric,
  CONSTRAINT other_table_pk PRIMARY KEY (other_table_key),
  ONSTRAINT other_table_reference_fk FOREIGN KEY (reference_table_key)
      REFERENCES reference_table (reference_table_key) MATCH SIMPLE
      ON UPDATE NO ACTION ON DELETE SET NULL
);

CREATE TABLE another_table (
  another_table_key numeric NOT NULL,
  do_stuff_key numeric,
  CONSTRAINT another_table_pk PRIMARY KEY (another_table_key),
  ONSTRAINT another_table_reference_fk FOREIGN KEY (do_stuff_key)
      REFERENCES reference_table (reference_table_key) MATCH SIMPLE
      ON UPDATE NO ACTION ON DELETE SET NULL
);

--there are 50-60 tables which have similar foreign key references to reference_table

I want to write a query that tells me the primary keys in other_table and another_table and potentially more tables where reference_value is NULL.

In psuedo-code:

SELECT table_name, table_primary_key, table_fk_column_name
FROM ?????? some PG table ???????, reference_table
WHERE reference_table.reference_value IS NULL;

The result would look something like:

table_name | table_primary_key | table_fk_column_name | reference_table_pk
---------------------------------------------------------------------------
other_table   | 2                |  reference_table_key | 7
other_table   | 4                |  reference_table_key | 56
other_table   | 45               |  reference_table_key | 454
other_table   | 65765            |  reference_table_key | 987987
other_table   | 11               |  reference_table_key | 3213
another_table | 3                |  do_stuff_key        | 4645
another_table | 5                |  do_stuff_key        | 43546
another_table | 7                |  do_stuff_key        | 464356
unknown_table | 1                |  unkown_column_key   | 435435
unknown_table | 1                |  some_other_column_key | 34543
unknown_table | 3                |  unkown_column_key   | 124
unknown_table | 3                |  some_other_column_key | 123

This is similar to, but not a duplicate of Postgres: SQL to list table foreign keys . That question shows the table structure. I want to find specific instances.

Essentially if I were to DELETE FROM reference_table WHERE reference_value IS NULL;, postgres has to do something internally to figure out that it needs to set reference_table_key in row 2 in other_table to NULL. I want to see what those rows would be.

Is there a query that can do this? Is there a modifier that I can pass to a DELETE call that would tell me what tables/rows/columns would be affected by that DELETE?

Answer 1

NULL values in referencing columns

This query produces the DML statement to find all rows in all tables, where a column has a foreign-key constraint referencing another table but hold a NULL value in that column:

WITH x AS (
 SELECT c.conrelid::regclass    AS tbl
      , c.confrelid::regclass   AS ftbl
      , quote_ident(k.attname)  AS fk
      , quote_ident(pf.attname) AS pk
 FROM   pg_constraint c
 JOIN   pg_attribute  k ON (k.attrelid, k.attnum) = (c.conrelid, c.conkey[1])
 JOIN   pg_attribute  f ON (f.attrelid, f.attnum) = (c.confrelid, c.confkey[1])
 LEFT   JOIN pg_constraint p  ON p.conrelid = c.conrelid AND p.contype = 'p'
 LEFT   JOIN pg_attribute  pf ON (pf.attrelid, pf.attnum)
                               = (p.conrelid, p.conkey[1])
 WHERE  c.contype   = 'f'
 AND    c.confrelid = 'fk_tbl'::regclass  -- references to this tbl
 AND    f.attname   = 'fk_tbl_id'         -- and only to this column
)
SELECT string_agg(format(
'SELECT %L AS tbl
     , %L AS pk
     , %s::text AS pk_val
     , %L AS fk
     , %L AS ftbl
FROM   %1$s WHERE %4$s IS NULL'
                  , tbl
                  , COALESCE(pk 'NONE')
                  , COALESCE(pk 'NULL')
                  , fk
                  , ftbl), '
UNION ALL
') || ';'
FROM   x;

Produces a query like this:

SELECT 'some_tbl' AS tbl
     , 'some_tbl_id' AS pk
     , some_tbl_id::text AS pk_val
     , 'fk_tbl_id' AS fk
     , 'fk_tbl' AS ftbl
FROM   some_tbl WHERE fk_tbl_id IS NULL
UNION ALL
SELECT 'other_tbl' AS tbl
     , 'other_tbl_id' AS pk
     , other_tbl_id::text AS pk_val
     , 'some_name_id' AS fk
     , 'fk_tbl' AS ftbl
FROM   other_tbl WHERE some_name_id IS NULL;

Produces output like this:

    tbl    |     pk       | pk_val |    fk        |  ftbl
-----------+--------------+--------+--------------+--------
 some_tbl  | some_tbl_id  | 49     | fk_tbl_id    | fk_tbl
 some_tbl  | some_tbl_id  | 58     | fk_tbl_id    | fk_tbl
 other_tbl | other_tbl_id | 66     | some_name_id | fk_tbl
 other_tbl | other_tbl_id | 67     | some_name_id | fk_tbl

• Does not cover multi-column foreign or primary keys reliably. You have to make the query more complex for this.

• I cast all primary key values to text to cover all types.

• Adapt or remove these lines to find foreign key pointing to an other or any column / table:

  AND    c.confrelid = 'fk_tbl'::regclass
  AND    f.attname = 'fk_tbl_id' -- and only this column
  

• Tested with PostgreSQL 9.1.4. I use the pg_catalog tables. Realistically nothing of what I use here is going to change, but that is not guaranteed across major releases. Rewrite it with tables from information_schema if you need it to work reliably across updates. That is slower, but sure.

• I did not sanitize table names in the generated DML script, because quote_ident() would fail with schema-qualified names. It is your responsibility to avoid harmful table names like "users; DELETE * FROM users;". With some more effort, you can retrieve schema-name and table name separately and use quote_ident().

---

NULL values in referenced columns

My first solution does something subtly different from what you ask, because what you describe (as I understand it) is non-existent. The value NULL is “unknown” and cannot be referenced. If you actually want to find rows with a NULL value in a column that has FK constraints pointing to it (not to the particular row with the NULL value, of course), then the query can be much simplified:

WITH x AS (
 SELECT c.confrelid::regclass   AS ftbl
       ,quote_ident(f.attname)  AS fk
       ,quote_ident(pf.attname) AS pk
       ,string_agg(c.conrelid::regclass::text, ', ') AS referencing_tbls
 FROM   pg_constraint c
 JOIN   pg_attribute  f ON (f.attrelid, f.attnum) = (c.confrelid, c.confkey[1])
 LEFT   JOIN pg_constraint p  ON p.conrelid = c.confrelid AND p.contype = 'p'
 LEFT   JOIN pg_attribute  pf ON (pf.attrelid, pf.attnum)
                               = (p.conrelid, p.conkey[1])
 WHERE  c.contype = 'f'
 -- AND    c.confrelid = 'fk_tbl'::regclass  -- only referring this tbl
 GROUP  BY 1, 2, 3
)
SELECT string_agg(format(
'SELECT %L AS ftbl
     , %L AS pk
     , %s::text AS pk_val
     , %L AS fk
     , %L AS referencing_tbls
FROM   %1$s WHERE %4$s IS NULL'
                  , ftbl
                  , COALESCE(pk, 'NONE')
                  , COALESCE(pk, 'NULL')
                  , fk
                  , referencing_tbls), '
UNION ALL
') || ';'
FROM   x;

Finds all such rows in the entire database (commented out the restriction to one table). Tested with Postgres 9.1.4 and works for me.

I group multiple tables referencing the same foreign column into one query and add a list of referencing tables to give an overview.