70

Background

When dropping (or replacing) objects in PostgreSQL, if there are dependencies, the drop will fail (without specifying CASCADE).

Problem

The error message returned by the database does not list the dependent objects.

Example Solution

The query might look something like:

SELECT * FROM information_schema i, pg_depend pd WHERE
  i.object_id = pd.object_id AND
  i.object_type = 'TABLE' AND
  i.object_schema = 'public' AND
  i.object_name = 'table_with_dependents';

The objid is missing.

Related

Question

How do you generate a list of dependent objects by name and type?

tshepang
  • 12,111
  • 21
  • 91
  • 136
Dave Jarvis
  • 30,436
  • 41
  • 178
  • 315
  • As [jpmc26 noted in a comment](https://stackoverflow.com/questions/4462908/find-dependent-objects-for-a-table-or-view#comment51564026_4464894), in Postgres 9.3+ the error message *does* list the dependent objects. – Wildcard Dec 22 '17 at 00:01
  • 3
    @Wildcard: Error messages are too brittle for machine parsing. – Dave Jarvis Dec 22 '17 at 05:30

5 Answers5

112

The suggested solution didn't work for me with postgresql 9.1.4

this worked:

SELECT dependent_ns.nspname as dependent_schema
, dependent_view.relname as dependent_view 
, source_ns.nspname as source_schema
, source_table.relname as source_table
, pg_attribute.attname as column_name
FROM pg_depend 
JOIN pg_rewrite ON pg_depend.objid = pg_rewrite.oid 
JOIN pg_class as dependent_view ON pg_rewrite.ev_class = dependent_view.oid 
JOIN pg_class as source_table ON pg_depend.refobjid = source_table.oid 
JOIN pg_attribute ON pg_depend.refobjid = pg_attribute.attrelid 
    AND pg_depend.refobjsubid = pg_attribute.attnum 
JOIN pg_namespace dependent_ns ON dependent_ns.oid = dependent_view.relnamespace
JOIN pg_namespace source_ns ON source_ns.oid = source_table.relnamespace
WHERE 
source_ns.nspname = 'my_schema'
AND source_table.relname = 'my_table'
AND pg_attribute.attnum > 0 
AND pg_attribute.attname = 'my_column'
ORDER BY 1,2;
Brandan
  • 14,735
  • 3
  • 56
  • 71
encc
  • 1,525
  • 1
  • 12
  • 14
  • It would be nice if you could show the schema of the dependee.relname because like this you know the object name but not the schema in which it is found. And because of the distinct, if there are two objects with the same relname in different schemeas, you will think there is only one dependancy – prince Feb 05 '14 at 10:21
  • 2
    @prince to get the schema, add `ns.nspname` to the SELECT list and `JOIN pg_namespace ns ON ns.oid = dependee.relnamespace` to the FROM list. – Aryeh Leib Taurog Jul 09 '14 at 10:16
33

The easy way is:

BEGIN;
DROP TABLE tablename CASCADE;
DROP VIEW viewname CASCADE;
ROLLBACK;
Stew
  • 4,495
  • 6
  • 31
  • 41
Scott Marlowe
  • 8,490
  • 3
  • 23
  • 21
  • I'm sure there's a better way involving hitting the pg_depend and pg_class catalogs. running psql -E and then using \d to look at tables should show you the queries psql uses to determine what's dependent on what. – Scott Marlowe Dec 17 '10 at 00:18
  • 8
    I don't think the `CASCADE` is needed in 9.3+. The error message detail includes the dependent objects now when it fails to drop. – jpmc26 Aug 04 '15 at 21:18
  • 3
    if you are dropping a table you need to include "table": begin; drop *table* fld_ovrd cascade; rollback; – mountainclimber11 Oct 20 '16 at 15:08
  • For the column it would be: `ALTER TABLE table_name DROP COLUMN column_name;` – chill appreciator Sep 21 '21 at 10:22
  • 1
    Regarding `CASCADE`. In case you specify `CASCADE` and object has too many dependent objects logs will be compressed to message like: `[2021-09-21 13:37:38] [00000] drop cascades to 22 other objects`. But if you don't specify `CASCADE` and there are dependent objects it will get even worse since transaction will be never closed(ROLLBACK never executed). So you would need to either manually close session or ran `ROLLBACK;`. Otherwise you may wonder why any command like `SELECT 1;` repeats failed transaction ^ – chill appreciator Sep 21 '21 at 10:44
  • 1
    How do you see the results? the return message just says "deleted 2 things" – pixelpax Oct 29 '22 at 19:50
12

Include nested views in the query as follows:

WITH RECURSIVE view_deps AS (
SELECT DISTINCT dependent_ns.nspname as dependent_schema
, dependent_view.relname as dependent_view
, source_ns.nspname as source_schema
, source_table.relname as source_table
FROM pg_depend
JOIN pg_rewrite ON pg_depend.objid = pg_rewrite.oid
JOIN pg_class as dependent_view ON pg_rewrite.ev_class = dependent_view.oid
JOIN pg_class as source_table ON pg_depend.refobjid = source_table.oid
JOIN pg_namespace dependent_ns ON dependent_ns.oid = dependent_view.relnamespace
JOIN pg_namespace source_ns ON source_ns.oid = source_table.relnamespace
WHERE NOT (dependent_ns.nspname = source_ns.nspname AND dependent_view.relname = source_table.relname)
UNION
SELECT DISTINCT dependent_ns.nspname as dependent_schema
, dependent_view.relname as dependent_view
, source_ns.nspname as source_schema
, source_table.relname as source_table
FROM pg_depend
JOIN pg_rewrite ON pg_depend.objid = pg_rewrite.oid
JOIN pg_class as dependent_view ON pg_rewrite.ev_class = dependent_view.oid
JOIN pg_class as source_table ON pg_depend.refobjid = source_table.oid
JOIN pg_namespace dependent_ns ON dependent_ns.oid = dependent_view.relnamespace
JOIN pg_namespace source_ns ON source_ns.oid = source_table.relnamespace
INNER JOIN view_deps vd
    ON vd.dependent_schema = source_ns.nspname
    AND vd.dependent_view = source_table.relname
    AND NOT (dependent_ns.nspname = vd.dependent_schema AND dependent_view.relname = vd.dependent_view)
)

SELECT *
FROM view_deps
ORDER BY source_schema, source_table;

If you care about specific table attributes add this to the top portion of the recursive CTE:

JOIN pg_attribute ON pg_depend.refobjid = pg_attribute.attrelid 
    AND pg_depend.refobjsubid = pg_attribute.attnum 
...
WHERE 
source_ns.nspname = 'my_schema'
AND source_table.relname = 'my_table'
AND pg_attribute.attnum > 0 
AND pg_attribute.attname = 'my_column'
Dave Jarvis
  • 30,436
  • 41
  • 178
  • 315
DB140141
  • 411
  • 4
  • 8
  • I get an error from the top statement "[Amazon](500310) Invalid operation: Recursive CTE must not omit column names.;" – MattG Jun 09 '22 at 01:05
10

For PostgreSQL 9.3 onward use the following view and functions to show any user object dependency. I also updated https://wiki.postgresql.org/wiki/Pg_depend_display.

/**** Usage Examples ****
-- Examine the entire object hierarchy
SELECT report.dependency_tree('');

-- Dependencies for any relations with names containing match (in regular expression)
SELECT report.dependency_tree('match');

-- Dependencies for relations person & address
SELECT report.dependency_tree('{person,address}'::text[]);

-- Dependencies for function slice
SELECT report.dependency_tree(ARRAY['slice'::regproc]);

-- Dependencies for type hstore
SELECT report.dependency_tree(ARRAY['hstore'::regtype]);

-- Dependencies for triggers by the name updated
SELECT report.dependency_tree(ARRAY(
  SELECT oid FROM pg_trigger WHERE tgname ~ 'updated'
  ));

-- Dependencies for foreign key constraint names starting with product
SELECT report.dependency_tree(ARRAY(
  SELECT oid FROM pg_constraint
  WHERE conname ~ '^product.*_fk'
  ));
*/

DROP VIEW IF EXISTS report.dependency;
CREATE OR REPLACE VIEW report.dependency AS
WITH RECURSIVE preference AS (
  SELECT 10 AS max_depth
    , 16384 AS min_oid -- user objects only
    , '^(londiste|pgq|pg_toast)'::text AS schema_exclusion
    , '^pg_(conversion|language|ts_(dict|template))'::text AS class_exclusion
    , '{"SCHEMA":"00", "TABLE":"01", "TABLE CONSTRAINT":"02", "DEFAULT VALUE":"03",
        "INDEX":"05", "SEQUENCE":"06", "TRIGGER":"07", "FUNCTION":"08",
        "VIEW":"10", "MATERIALIZED VIEW":"11", "FOREIGN TABLE":"12"}'::json AS type_sort_orders
)
, dependency_pair AS (
    SELECT objid
      , array_agg(objsubid ORDER BY objsubid) AS objsubids
      , upper(obj.type) AS object_type
      , coalesce(obj.schema, substring(obj.identity, E'(\\w+?)\\.'), '') AS object_schema
      , obj.name AS object_name
      , obj.identity AS object_identity
      , refobjid
      , array_agg(refobjsubid ORDER BY refobjsubid) AS refobjsubids
      , upper(refobj.type) AS refobj_type
      , coalesce(CASE WHEN refobj.type='schema' THEN refobj.identity
                                                ELSE refobj.schema END
          , substring(refobj.identity, E'(\\w+?)\\.'), '') AS refobj_schema
      , refobj.name AS refobj_name
      , refobj.identity AS refobj_identity
      , CASE deptype
            WHEN 'n' THEN 'normal'
            WHEN 'a' THEN 'automatic'
            WHEN 'i' THEN 'internal'
            WHEN 'e' THEN 'extension'
            WHEN 'p' THEN 'pinned'
        END AS dependency_type
    FROM pg_depend dep
      , LATERAL pg_identify_object(classid, objid, 0) AS obj
      , LATERAL pg_identify_object(refclassid, refobjid, 0) AS refobj
      , preference
    WHERE deptype = ANY('{n,a}')
    AND objid >= preference.min_oid
    AND (refobjid >= preference.min_oid OR refobjid = 2200) -- need public schema as root node
    AND coalesce(obj.schema, substring(obj.identity, E'(\\w+?)\\.'), '') !~ preference.schema_exclusion
    AND coalesce(CASE WHEN refobj.type='schema' THEN refobj.identity
                                                ELSE refobj.schema END
          , substring(refobj.identity, E'(\\w+?)\\.'), '') !~ preference.schema_exclusion
    GROUP BY objid, obj.type, obj.schema, obj.name, obj.identity
      , refobjid, refobj.type, refobj.schema, refobj.name, refobj.identity, deptype
)
, dependency_hierarchy AS (
    SELECT DISTINCT
        0 AS level,
        refobjid AS objid,
        refobj_type AS object_type,
        refobj_identity AS object_identity,
        --refobjsubids AS objsubids,
        NULL::text AS dependency_type,
        ARRAY[refobjid] AS dependency_chain,
        ARRAY[concat(preference.type_sort_orders->>refobj_type,refobj_type,':',refobj_identity)] AS dependency_sort_chain
    FROM dependency_pair root
    , preference
    WHERE NOT EXISTS
       (SELECT 'x' FROM dependency_pair branch WHERE branch.objid = root.refobjid)
    AND refobj_schema !~ preference.schema_exclusion
    UNION ALL
    SELECT
        level + 1 AS level,
        child.objid,
        child.object_type,
        child.object_identity,
        --child.objsubids,
        child.dependency_type,
        parent.dependency_chain || child.objid,
        parent.dependency_sort_chain || concat(preference.type_sort_orders->>child.object_type,child.object_type,':',child.object_identity)
    FROM dependency_pair child
    JOIN dependency_hierarchy parent ON (parent.objid = child.refobjid)
    , preference
    WHERE level < preference.max_depth
    AND child.object_schema !~ preference.schema_exclusion
    AND child.refobj_schema !~ preference.schema_exclusion
    AND NOT (child.objid = ANY(parent.dependency_chain)) -- prevent circular referencing
)
SELECT * FROM dependency_hierarchy
ORDER BY dependency_chain ;

-- Procedure to report depedency tree using regexp search pattern (relation-only)
CREATE OR REPLACE FUNCTION report.dependency_tree(search_pattern text)
  RETURNS TABLE(dependency_tree text)
  SECURITY DEFINER LANGUAGE SQL
  AS $function$
WITH target AS (
  SELECT objid, dependency_chain
  FROM report.dependency
  WHERE object_identity ~ search_pattern
)
, list AS (
  SELECT
    format('%*s%s %s', -4*level
          , CASE WHEN object_identity ~ search_pattern THEN '*' END
          , object_type, object_identity
    ) AS dependency_tree
  , dependency_sort_chain
  FROM target
  JOIN report.dependency report
    ON report.objid = ANY(target.dependency_chain) -- root-bound chain
    OR target.objid = ANY(report.dependency_chain) -- leaf-bound chain
  WHERE length(search_pattern) > 0
  -- Do NOT waste search time on blank/null search_pattern.
  UNION
  -- Query the entire dependencies instead.
  SELECT
    format('%*s%s %s', 4*level, '', object_type, object_identity) AS depedency_tree
  , dependency_sort_chain
  FROM report.dependency
  WHERE length(coalesce(search_pattern,'')) = 0
)
SELECT dependency_tree FROM list
ORDER BY dependency_sort_chain;
$function$ ;

-- Procedure to report depedency tree by specific relation name(s) (in text array)
CREATE OR REPLACE FUNCTION report.dependency_tree(object_names text[])
  RETURNS TABLE(dependency_tree text)
  SECURITY DEFINER LANGUAGE SQL
  AS $function$
WITH target AS (
  SELECT objid, dependency_chain
  FROM report.dependency
  JOIN unnest(object_names) AS target(objname) ON objid = objname::regclass
)
, list AS (
  SELECT DISTINCT
    format('%*s%s %s', -4*level
          , CASE WHEN object_identity = ANY(object_names) THEN '*' END
          , object_type, object_identity
    ) AS dependency_tree
  , dependency_sort_chain
  FROM target
  JOIN report.dependency report
    ON report.objid = ANY(target.dependency_chain) -- root-bound chain
    OR target.objid = ANY(report.dependency_chain) -- leaf-bound chain
)
SELECT dependency_tree FROM list
ORDER BY dependency_sort_chain;
$function$ ;

-- Procedure to report depedency tree by oid
CREATE OR REPLACE FUNCTION report.dependency_tree(object_ids oid[])
  RETURNS TABLE(dependency_tree text)
  SECURITY DEFINER LANGUAGE SQL
  AS $function$
WITH target AS (
  SELECT objid, dependency_chain
  FROM report.dependency
  JOIN unnest(object_ids) AS target(objid) USING (objid)
)
, list AS (
  SELECT DISTINCT
    format('%*s%s %s', -4*level
          , CASE WHEN report.objid = ANY(object_ids) THEN '*' END
          , object_type, object_identity
    ) AS dependency_tree
  , dependency_sort_chain
  FROM target
  JOIN report.dependency report
    ON report.objid = ANY(target.dependency_chain) -- root-bound chain
    OR target.objid = ANY(report.dependency_chain) -- leaf-bound chain
)
SELECT dependency_tree FROM list
ORDER BY dependency_sort_chain;
$function$ ;
Kong Man
  • 111
  • 1
  • 7
0

pg_constraint contains all constrains in the database you can list the oid of dependent tables using confrelid and conrelid from all all the foreign key constraints

query looks like this

select confrelid,conrelid from pg_constraint where contype='f';

hkbharath
  • 317
  • 7
  • 15