query to update records count based on relation and constraints

Im trying to update a column (address_count) on my people table that must have the sum of records from another table (people_addresses) plus some constraints.

The idea is that, the record on people must exist on people_addresses and must be present on at least one of the other tables (table a, table b)

SELECT pd.address_id, count (pd.person_id)
FROM person_addresses pd
WHERE EXISTS (SELECT a.person_id 
              FROM table_a a where a.person_id = pd.person_id)
  OR EXISTS (SELECT b.person_id 
             FROM table_b b where b.person_id = pd.person_id)
GROUP BY pd.address_id

With the query above, I get the expected count, for each address that fits my requirements, altough, when I try to use it on a update, I get weird results, the records that don’t fit the requirements end up with wrong values on people_count

query im using:

UPDATE people d
SET people_count = (SELECT count (pd.person_id) 
                    FROM person_addresses pd
                    WHERE EXISTS (SELECT a.person_id 
                                  FROM table_a a where a.person_id = pd.person_id)
  OR EXISTS (SELECT b.person_id 
             FROM table_b b where b.person_id = pd.person_id)
  AND d.id=pd.address_id
  GROUP BY pd.address_id) 

I get the correct result, on records that fit the requirements, but wrong values on the ones that dont fit, when I run the update.

Answer

The problem is in your conditions:

WHERE EXISTS (...) OR EXISTS (...) AND d.id = pd.adress_id

Is actually equivalent to:

WHERE EXISTS (...) OR (EXISTS (...) AND d.id = pd.adress_id)

You need to surround the OR condition with parentheses to make your intent explicit:

WHERE (EXISTS (...) OR EXISTS (...)) AND d.id = pd.adress_id

Also, the GROUP BY clause is superfluous in the subquery. I would suggest removing it (you will get a 0 value if the subquery does not match).