使用 postgres 触发器有效地更新标签计数

Question

I have a schema where users have tags associated with them:

user_id: int4    tags: text[]
-------------    ------------
1                [ 'apple', 'carrot', 'jelly' ]
2                [ 'jelly', 'zebra' ]

I am looking to create query that returns the list of all tags along with their associated counts. Example:

tag: text      count: int4
---------      -----------
'apple'        1
'carrot'       1
'jelly'        2
'zebra'        1

Triggers seem to be the ideal way to do this, since the application is read-heavy and write-light.

However, I am having difficulty implementing this. The trigger itself seems simple enough I think:

CREATE TRIGGER tags_count_update
AFTER UPDATE OF tags ON person
FOR EACH ROW
EXECUTE PROCEDURE trigger_update_tags_count();

The trigger_update_tags_count is the part I'm having trouble with, because it seems to be a highly complex operation. For example, if the tag doesn't already exist in the tags table then it should be inserted with a count of 1.

Also, I believe you need to do some sort of diff operation, because if someone's tags are updated from [ 'apple', 'carrot', 'jelly' ] to [ 'apple', 'dogs' ] then apple 's count doesn't change, carrot and jelly is decremented by 1, and dogs is created with its count set to 1. This is compounded with the fact that tags are arrays. I currently have something like this:

CREATE OR REPLACE FUNCTION public.trigger_update_tags_count()
  RETURNS trigger
  LANGUAGE plpgsql
AS $function$
BEGIN
   INSERT INTO tags (tag, count) VALUES (new.tag, 1)
   ON CONFLICT (tag) DO UPDATE
   SET count = CASE WHEN (tag.new AND NOT tag.old) THEN count + 1 
                    WHEN (tag.old AND NOT tag.new) THEN count - 1
                    ELSE count
               END
   RETURN NEW;
END;
$function$;

Which is slightly pseudocode because I'm not sure how to integrate the fact that I'm dealing with text arrays, nor how to handle the diffing case.

Answer 1

You have nothing called tag in the data; it is tags . So you need to UNNEST() and to handle both old and new tags. I'm thinking something like:

INSERT INTO tags (tag, count) 
   SELECT COALESCE(ntag, otag),
          (ntag IS NOT NULL)::int - (otag IS NOT NULL)::int
   FROM UNNEST(new.tags) ntag FULL JOIN
        UNNEST(old.tags) otag
        ON ntag = otag
ON CONFLICT (tag) DO UPDATE
   SET count = count + (excluded.ntag IS NOT NULL)::int - (excluded.otag IS NOT NULL)::int

Answer 2

I am unsure that you do need a trigger a complex logic for that. You can get the output that you want with a simple query that leverages Postgres' array function unnest() :

select x.tag, count(*)
from mytable t
cross join lateral unnest(t.tags) x(tag)
group by x.tag

Note that if you find yourself repeatedly running such query, it is an indication that your design does not really fit your use case; you might be better off storing each user_id/tag tuple in a separate table row, separated from the tables where users are stored.