PostgreSQL: Trigger Function bypasses trigger of another table - sql

Question

There are two tables: gc_res and anchor. Anchor table has a unique key - anchor_id.
When Data arrives in gc_res the res_eval trigger is fired and the data is partially evaluated by gc_res_eval().
Some of the entries remain, others are pushed with an insert into anchor table. The anchor table has a trigger before insert, which also pre-evaluates the data and either accepts the insert or updates an old entry. In latter case it might additionally write to a third (log) table.
Both triggers look the same:
CREATE TRIGGER res_eval BEFORE INSERT ON gc_res
FOR EACH ROW EXECUTE PROCEDURE gc_res_eval();
CREATE TRIGGER anchor_smart_insert BEFORE INSERT ON anchor
FOR EACH ROW EXECUTE PROCEDURE anchor_smart_insert();
When I directly try to write to anchor table, the anchor_smart_insert trigger fires properly and works fine.
The gc_res_eval part which tries to insert the data works with a CTE:
WITH anchor_insert AS (
INSERT INTO cbm.anchor (anchor_id, pin, geo, rwo, rwo_value, addr) (
SELECT osm_id, pin, geo, rwo, rwo_value, addr FROM gc_eval
WHERE similarity > accept_polygon AND geo_type = 'Polygon'
) RETURNING anchor_id
) SELECT array_agg(anchor_id) FROM anchor_insert INTO anchors
;
When gc_res_eval tries to insert into the anchor table and there are no conflicts on anchor_id - everything works fine. If there is a conflict, the insert fails on the unique constraint. Thus I assume - that the second trigger does not fire.
CREATE OR REPLACE FUNCTION anchor_smart_insert()
RETURNS trigger
LANGUAGE plpgsql
AS $function$
DECLARE
geog geography;
acceptable_intersect int := 10;
acceptable_overlap_index NUMERIC := 0.56;
BEGIN
SELECT INTO geog geo FROM cbm.anchor a
WHERE a.anchor_id=NEW.anchor_id;
RAISE NOTICE 'anchor_smart_insert is running';
IF ST_Area(ST_Intersection(geog, NEW.geo))>acceptable_intersect THEN
UPDATE cbm.anchor a SET
pin=COALESCE(NEW.pin, ST_Centroid(NEW.geo)),
geo=NEW.geo,
rwo=NEW.rwo,
rwo_value=NEW.rwo_value,
addr=NEW.addr
WHERE a.anchor_id=NEW.anchor_id AND (
a.geo IS DISTINCT FROM NEW.geo
OR a.rwo IS DISTINCT FROM NEW.rwo
OR a.rwo_value IS DISTINCT FROM NEW.rwo_value
OR a.addr IS DISTINCT FROM NEW.addr);
INSERT INTO cbm.anchor_history (anchor_id, geo) VALUES (NEW.anchor_id, geog);
IF ST_Area(ST_Intersection(geog, NEW.geo))^2/(ST_Area(geog)*ST_Area(NEW.geo))<acceptable_overlap_index THEN
INSERT INTO cbm.anchor_review(anchor_id, geo, rwo, rwo_value, addr) VALUES
(NEW.anchor_id, NEW.geo, NEW.rwo, NEW.rwo_value, NEW.addr);
END IF;
RETURN NULL;
END IF;
RETURN NEW;
END;
$function$
;
The DBMS is a pg11 AWS RDS.

Answer 1

Blaming myself:
The anchor_smart_insert() function was working fine as long as the polygons which have been compared were greater than 10sqm. There was the assumption, that there should be only two options: greater than 10sqm, or null.
With such logical reasoning the part, which captures all exemptions was deleted since it seemed unnecessary.
This is also why the tests via direct insert worked well: the polygons were >10sqm and resulted in an update.
Thanks #Adrian Klaver - your suggestions helped!
The restored/improved function properly handles polygons <10sqm in a way, that it pre-compares it to initial and final sizes - if they are smaller, it takes a safer value.
Additionally it takes explicitly care of possible inserts (existing anchor_id ISNULL).
CREATE OR REPLACE FUNCTION cbm.anchor_smart_insert()
RETURNS trigger
LANGUAGE plpgsql
AS $function$
DECLARE
geog geography;
acceptable_intersect NUMERIC:= 10;
acceptable_overlap_index NUMERIC := 0.56;
BEGIN
SELECT INTO geog geo FROM cbm.anchor a
WHERE a.anchor_id=NEW.anchor_id;
IF geog ISNULL THEN RETURN NEW; END IF;
IF ST_Area(ST_Intersection(geog, NEW.geo))>LEAST(acceptable_intersect, GREATEST(ST_Area(geog), ST_Area(NEW.geo))) THEN
UPDATE cbm.anchor a SET
pin=COALESCE(NEW.pin, ST_Centroid(NEW.geo)),
geo=NEW.geo,
rwo=NEW.rwo,
rwo_value=NEW.rwo_value,
addr=NEW.addr
WHERE a.anchor_id=NEW.anchor_id AND (
a.geo IS DISTINCT FROM NEW.geo
OR a.rwo IS DISTINCT FROM NEW.rwo
OR a.rwo_value IS DISTINCT FROM NEW.rwo_value
OR a.addr IS DISTINCT FROM NEW.addr);
IF ST_Area(ST_Intersection(geog, NEW.geo))^2/(ST_Area(geog)*ST_Area(NEW.geo))<acceptable_overlap_index THEN
INSERT INTO cbm.anchor_history (anchor_id, geo) VALUES (NEW.anchor_id, geog);
END IF;
END IF;
-- bad intersection:
INSERT INTO cbm.anchor_review(anchor_id, geo, rwo, rwo_value, addr) VALUES
(NEW.anchor_id, NEW.geo, NEW.rwo, NEW.rwo_value, NEW.addr);
RETURN NULL;
END;
$function$
;