I am creating trigger in PostgresSQL. On update I would like to compare all of the values in a Hstore column and update changes in my mirror table. I managed to get names of my columns in variable k but I am not able to get values using it from NEW and OLD.
CREATE OR REPLACE FUNCTION function_replication() RETURNS TRIGGER AS
$BODY$
DECLARE
k text;
BEGIN
FOR k IN SELECT key FROM EACH(hstore(NEW)) LOOP
IF NEW.k != OLD.k THEN
EXECUTE 'UPDATE ' || TG_TABLE_NAME || '_2' || 'SET ' || k || '=' || new.k || ' WHERE ID=$1.ID;' USING OLD;
END IF;
END LOOP;
RETURN NEW;
END;
$BODY$
language plpgsql;
You should operate on hstore representations of the records new and old. Also, use the format() function for better control and readibility.
create or replace function function_replication()
returns trigger as
$body$
declare
newh hstore = hstore(new);
oldh hstore = hstore(old);
key text;
begin
foreach key in array akeys(newh) loop
if newh->key != oldh->key then
execute format(
'update %s_2 set %s = %L where id = %s',
tg_table_name, key, newh->key, oldh->'id');
end if;
end loop;
return new;
end;
$body$
language plpgsql;
Another version - with minimalistic numbers of updates - in partially functional design (where it is possible).
This trigger should be AFTER trigger, to be ensured correct behave.
CREATE OR REPLACE FUNCTION function_replication()
RETURNS trigger AS $$
DECLARE
newh hstore;
oldh hstore;
update_vec text[];
pair text[];
BEGIN
IF new IS DISTINCT FROM old THEN
IF new.id <> old.id THEN
RAISE EXCEPTION 'id should be immutable';
END IF;
newh := hstore(new); oldh := hstore(old); update_vec := '{}';
FOREACH pair SLICE 1 IN ARRAY hstore_to_matrix(newh - oldh)
LOOP
update_vec := update_vec || format('%I = %L', pair[1], pair[2]);
END LOOP;
EXECUTE
format('UPDATE %I SET %s WHERE id = $1',
tg_table_name || '_2',
array_to_string(update_vec, ', '))
USING old.id;
END IF;
RETURN NEW; -- the value is not important in AFTER trg
END;
$$ LANGUAGE plpgsql;
CREATE TABLE foo(id int PRIMARY KEY, a int, b int);
CREATE TABLE foo_2(LIKE foo INCLUDING ALL);
CREATE TRIGGER xxx AFTER UPDATE ON foo
FOR EACH ROW EXECUTE PROCEDURE function_replication();
INSERT INTO foo VALUES(1, NULL, NULL);
INSERT INTO foo VALUES(2, 1,1);
INSERT INTO foo_2 VALUES(1, NULL, NULL);
INSERT INTO foo_2 VALUES(2, 1,1);
UPDATE foo SET a = 20, b = 30 WHERE id = 1;
UPDATE foo SET a = NULL WHERE id = 1;
This code is little bit more complex, but all what should be escaped is escaped and reduce number of executed UPDATE commands. UPDATE is full SQL command and the overhead of full SQL commands should be significantly higher than code that reduce number of full SQL commands.
Related
I checked all related questions on SO but none helped in my case.
I have 2 loops(outside for the tables and inside for the columns). Tables are represented by 'r', and columns by 'm'. While being inside the 'm' loop which is supposed to send column values to the to-be-created trigger function. When I try to use 'NEW.m' (with trying many different formatting attempts) compiler always gives error.
Can you kindly advice on it please? Br
FOR r IN SELECT table_name FROM information_schema.tables LOOP
FOR m IN SELECT column_name FROM information_schema.columns WHERE (table_name = r.table_name ) LOOP
function_name := 'dictionary_functions_foreach_trigger';
EXECUTE format('CREATE OR REPLACE FUNCTION %s()
RETURNS trigger AS
$BODY$
DECLARE
BEGIN
IF NEW.m IS NOT NULL AND NEW.m IN (SELECT key FROM tableX.tableX_key)
THEN RETURN NEW;
END IF;
END;
$BODY$
LANGUAGE plpgsql VOLATILE
COST 100;
ALTER FUNCTION %s()
OWNER TO mydb;',function_name, function_name);
EXECUTE 'CREATE TRIGGER ' || function_name || ' BEFORE INSERT OR UPDATE ON ' || belonging_to_schema || '.' || r.table_name || ' FOR EACH ROW EXECUTE PROCEDURE ' || function_name || '();';
----Trigger Functions after edit-
EXECUTE format(
'CREATE OR REPLACE FUNCTION %s()
RETURNS trigger AS
$BODY$
DECLARE
insideIs text := %s ;
BEGIN
FOR %s IN 0..(TG_NARGS-1) LOOP
IF %I= TG_ARGV[%s]
THEN insideIs := %s ;
END IF;
END LOOP;
IF NEW.%I IS NOT NULL AND (insideIs =%s) AND NEW.%I IN (SELECT key FROM tableX.tableX_key)
THEN RETURN NEW;
ELSE RETURN OLD;
END IF;
END;
$BODY$
LANGUAGE plpgsql VOLATILE
COST 100;
ALTER FUNCTION %s()
OWNER TO mydb;' , function_name, 'notInside', 'i' , m.column_name, 'i' , 'ok', m.column_name, 'ok', m.column_name ,function_name);
You need to use another placeholder for the column name, they way you have written it, the column name "m" is hardcoded in the function.
You also don't really need the outer loop, as the table_name is also available in information_schema.columns.
Your trigger would also fail with a runtime error if the condition is not true as you don't have a return in that case. If you want to abort the statement, use return null;
You should also use format() for the create trigger statement.
FOR m IN SELECT table_schema, table_name, column_name
FROM information_schema.columns
WHERE table_name in (...)
LOOP
function_name := 'dictionary_functions_foreach_trigger';
EXECUTE format('CREATE OR REPLACE FUNCTION %I()
RETURNS trigger AS
$BODY$
DECLARE
BEGIN
IF NEW.%I IS NOT NULL AND NEW.%I IN (SELECT key FROM tableX.tableX_key) THEN
RETURN NEW;
END IF;
RETURN null; --<< you need some kind of return here!
END;
$BODY$
LANGUAGE plpgsql VOLATILE
COST 100;
ALTER FUNCTION %s()
OWNER TO mydb;', function_name, m.column_name, m.column_name, function_name, function_name);
EXECUTE format('CREATE TRIGGER %I BEFORE INSERT OR UPDATE ON %I.%I FOR EACH ROW EXECUTE PROCEDURE %I()',
function_name, m.table_schema, m.table_name, function_name);
END LOOP;
Online example
I have some function on PostgreSQL 9.6 returning a cursor (refcursor):
CREATE OR REPLACE FUNCTION public.test_returning_cursor()
RETURNS refcursor
IMMUTABLE
LANGUAGE plpgsql
AS $$
DECLARE
_ref refcursor = 'test_returning_cursor_ref1';
BEGIN
OPEN _ref FOR
SELECT 'a' :: text AS col1
UNION
SELECT 'b'
UNION
SELECT 'c';
RETURN _ref;
END
$$;
I need to write another function in which a temp table is created and all data from this refcursor are inserted to it. But INSERT INTO ... FETCH ALL FROM ... seems to be impossible. Such function can't be compiled:
CREATE OR REPLACE FUNCTION public.test_insert_from_cursor()
RETURNS table(col1 text)
IMMUTABLE
LANGUAGE plpgsql
AS $$
BEGIN
CREATE TEMP TABLE _temptable (
col1 text
) ON COMMIT DROP;
INSERT INTO _temptable (col1)
FETCH ALL FROM "test_returning_cursor_ref1";
RETURN QUERY
SELECT col1
FROM _temptable;
END
$$;
I know that I can use:
FOR _rec IN
FETCH ALL FROM "test_returning_cursor_ref1"
LOOP
INSERT INTO ...
END LOOP;
But is there better way?
Unfortunately, INSERT and SELECT don't have access to cursors as a whole.
To avoid expensive single-row INSERT, you could have intermediary functions with RETURNS TABLE and return the cursor as table with RETURN QUERY. See:
Return a query from a function?
CREATE OR REPLACE FUNCTION f_cursor1_to_tbl()
RETURNS TABLE (col1 text) AS
$func$
BEGIN
-- MOVE BACKWARD ALL FROM test_returning_cursor_ref1; -- optional, see below
RETURN QUERY
FETCH ALL FROM test_returning_cursor_ref1;
END
$func$ LANGUAGE plpgsql; -- not IMMUTABLE
Then create the temporary table(s) directly like:
CREATE TEMP TABLE t1 ON COMMIT DROP
AS SELECT * FROM f_cursor1_to_tbl();
See:
Creating temporary tables in SQL
Still not very elegant, but much faster than single-row INSERT.
Note: Since the source is a cursor only the first call succeeds. Executing the function a second time would return an empty set. You would need a cursor with the SCROLL option and move to the start for repeated calls.
This function does INSERT INTO from refcursor. It is universal for all the tables. The only requirement is that all columns of table corresponds to columns of refcursor by types and order (not necessary by names).
to_json() does the trick to convert any primitive data types to string with double-quotes "", which are later replaced with ''.
CREATE OR REPLACE FUNCTION public.insert_into_from_refcursor(_table_name text, _ref refcursor)
RETURNS void
LANGUAGE plpgsql
AS $$
DECLARE
_sql text;
_sql_val text = '';
_row record;
_hasvalues boolean = FALSE;
BEGIN
LOOP --for each row
FETCH _ref INTO _row;
EXIT WHEN NOT found; --there are no rows more
_hasvalues = TRUE;
SELECT _sql_val || '
(' ||
STRING_AGG(val.value :: text, ',') ||
'),'
INTO _sql_val
FROM JSON_EACH(TO_JSON(_row)) val;
END LOOP;
_sql_val = REPLACE(_sql_val, '"', '''');
_sql_val = TRIM(TRAILING ',' FROM _sql_val);
_sql = '
INSERT INTO ' || _table_name || '
VALUES ' || _sql_val;
--RAISE NOTICE 'insert_into_from_refcursor(): SQL is: %', _sql;
IF _hasvalues THEN --to avoid error when trying to insert 0 values
EXECUTE (_sql);
END IF;
END;
$$;
Usage:
CREATE TABLE public.table1 (...);
PERFORM my_func_opening_refcursor();
PERFORM public.insert_into_from_refcursor('public.table1', 'name_of_refcursor_portal'::refcursor);
where my_func_opening_refcursor() contains
DECLARE
_ref refcursor = 'name_of_refcursor_portal';
OPEN _ref FOR
SELECT ...;
I have a simple mission is inserting huge MD5 values into tables (partitioned table), and have created a trigger and also a trigger function to instead of INSERT operation. And in function I checked the first two characters of NEW.md5 to determine which table should be inserted.
DECLARE
tb text;
BEGIN
IF TG_OP = 'INSERT' THEN
tb = 'samples_' || left(NEW.md5, 2);
EXECUTE(format('INSERT INTO %s VALUES (%s);', tb, NEW.*)); <- WRONG
END IF;
RETURN NULL;
END;
The question is how to concat the NEW.* into the SQL statement?
Best with the USING clause of EXECUTE:
CREATE FUNCTION foo ()
RETURNS trigger AS
$func$
BEGIN
IF TG_OP = 'INSERT' THEN
EXECUTE format('INSERT INTO %s SELECT $1.*'
, 'samples_' || left(NEW.md5, 2);
USING NEW;
END IF;
RETURN NULL;
END
$func$ LANGUAGE plpgsql;
And EXECUTE does not require parentheses.
And you are aware that identifiers are folded to lower case unless quoted where necessary (%I instead of %s in format()).
More details:
INSERT with dynamic table name in trigger function
How to dynamically use TG_TABLE_NAME in PostgreSQL 8.2?
I wrote a PL/pgsql to batch create index on tables
CREATE OR REPLACE FUNCTION create_index() RETURNS void AS
$BODY$
DECLARE
r INTEGER;
BEGIN
FOR r IN 1..1000
LOOP
EXECUTE format(
' CREATE INDEX idx_abc_id_' || r::text ||
' ON abc_id_' || r::text ||
' USING btree
(key);');
END LOOP;
RETURN;
END
$BODY$
LANGUAGE plpgsql;
it has one problem, if partition abc_500 doesn't exist, then the how create index function will fail and do nothing.
How to make loop keep going through even if create_index made an error on one of the table in between?
I think a better approach would be to not hardcode the number for the loop, but iterate over the existing tables:
CREATE OR REPLACE FUNCTION create_index() RETURNS void AS
$BODY$
DECLARE
r record;
BEGIN
FOR r IN select tablename, regexp_replace(tablename, '[^0-9]+','') as idx_nr
from pg_tables
where tablename ~ 'abc_id_[0-9]+'
LOOP
EXECUTE format('CREATE INDEX %I ON %I USING btree (key)',
'idx_abc_id_'||r.idx_nr,
r.tablename);
END LOOP;
RETURN;
END
$BODY$
LANGUAGE plpgsql;
When you use the format() function is better to use the proper place holders for identifiers.
If you also want to ignore any error when creating the index on an existing table, you need to catch the exception and ignore it:
CREATE OR REPLACE FUNCTION create_index() RETURNS void AS
$BODY$
DECLARE
r record;
msg text;
BEGIN
FOR r IN select tablename, regexp_replace(tablename, '[^0-9]+','') as idx_nr
from pg_tables
where tablename ~ 'abc_id_[0-9]+'
LOOP
BEGIN
EXECUTE format('CREATE INDEX %I ON %I USING btree (key)',
'idx_abc_id_'||r.idx_nr,
r.tablename);
EXCEPTION
WHEN OTHERS THEN
GET STACKED DIAGNOSTICS msg = MESSAGE_TEXT;
RAISE NOTICE 'Could not create index for: %, %', r.idx_nr, msg;
END;
END LOOP;
RETURN;
END
$BODY$
LANGUAGE plpgsql;
I'm trying to implement a generic trigger procedure to enable a sort a versioning scheme on tables. Tables all have version and current fields. On updates, in some situations based on a condition, i want to create a new version of a row instead of updating the old one. I'm having trouble getting the default value for the primary key field (always id).
Here's what i've done:
CREATE FUNCTION version_trigger() RETURNS trigger AS $$
DECLARE
id_default text;
id_value text;
BEGIN
IF version_condition() THEN
old.current = false;
-- I can read the default value
SELECT column_default INTO id_default
FROM information_schema.columns
WHERE table_name = TG_TABLE_NAME AND column_name = 'id';
-- THIS DOESN'T WORK!
EXECUTE 'SELECT $1' INTO id_value USING id_default;
new.id = id_value;
EXECUTE 'INSERT INTO ' || quote_ident(TG_TABLE_NAME) || ' SELECT ($1).*' USING new;
RETURN old;
END IF;
-- regular UPDATE
RETURN new;
END
$$ LANGUAGE plpgsql;
I'm just missing the step where i read the default value for the id (it's just a nextval() call). Can anyone help out on this?
Thanks in advance!
You cannot use a placeholder for expressions.
If DEFAULT doesn't have any reference to record data, then you can use EXECUTE statement, but little bit different
postgres=# DO $$
DECLARE x text; y text;
BEGIN
x := (SELECT column_default
FROM information_schema.columns
WHERE table_name = 'omega' AND column_name = 'a');
EXECUTE 'SELECT ' || x INTO y;
RAISE NOTICE '%', y;
END;
$$;
NOTICE: 2
DO