postgresql unique non-sequential id for url - postgresql

Question

I've looked a few methods of creating alphanumeric IDs on Stackoverflow, but they all had their weaknesses, some did not check for collision and others used sequences which are not a good option when using logical replication.
After some Googling I found this website that has the following script which checks for collisions and does not use sequences. However this is done as a trigger when a row is inserted into the table.
-- Create a trigger function that takes no arguments.
-- Trigger functions automatically have OLD, NEW records
-- and TG_TABLE_NAME as well as others.
CREATE OR REPLACE FUNCTION unique_short_id()
RETURNS TRIGGER AS $$
-- Declare the variables we'll be using.
DECLARE
key TEXT;
qry TEXT;
found TEXT;
BEGIN
-- generate the first part of a query as a string with safely
-- escaped table name, using || to concat the parts
qry := 'SELECT id FROM ' || quote_ident(TG_TABLE_NAME) || ' WHERE id=';
-- This loop will probably only run once per call until we've generated
-- millions of ids.
LOOP
-- Generate our string bytes and re-encode as a base64 string.
key := encode(gen_random_bytes(6), 'base64');
-- Base64 encoding contains 2 URL unsafe characters by default.
-- The URL-safe version has these replacements.
key := replace(key, '/', '_'); -- url safe replacement
key := replace(key, '+', '-'); -- url safe replacement
-- Concat the generated key (safely quoted) with the generated query
-- and run it.
-- SELECT id FROM "test" WHERE id='blahblah' INTO found
-- Now "found" will be the duplicated id or NULL.
EXECUTE qry || quote_literal(key) INTO found;
-- Check to see if found is NULL.
-- If we checked to see if found = NULL it would always be FALSE
-- because (NULL = NULL) is always FALSE.
IF found IS NULL THEN
-- If we didn't find a collision then leave the LOOP.
EXIT;
END IF;
-- We haven't EXITed yet, so return to the top of the LOOP
-- and try again.
END LOOP;
-- NEW and OLD are available in TRIGGER PROCEDURES.
-- NEW is the mutated row that will actually be INSERTed.
-- We're replacing id, regardless of what it was before
-- with our key variable.
NEW.id = key;
-- The RECORD returned here is what will actually be INSERTed,
-- or what the next trigger will get if there is one.
RETURN NEW;
END;
$$ language 'plpgsql';
I have have a table which already contains data, I have added a new column called pid would it be possible to modify this and use the function call as default so all my prior data gets a short id?

Answer 1

Suppose you have a table test:
DROP TABLE IF EXISTS test;
CREATE TABLE test (foo text, bar int);
INSERT INTO test (foo, bar) VALUES ('A', 1), ('B', 2);
You could add an id column to it:
ALTER TABLE test ADD COLUMN id text;
and attach the trigger:
DROP TRIGGER IF EXISTS unique_short_id_on_test ON test;
CREATE TRIGGER unique_short_id_on_test
BEFORE INSERT ON test
FOR EACH ROW EXECUTE PROCEDURE unique_short_id();
Now make a temporary table, temp, with the same structure as test (but with no data):
DROP TABLE IF EXISTS temp;
CREATE TABLE temp (LIKE test INCLUDING ALL);
CREATE TRIGGER unique_short_id_on_temp
BEFORE INSERT ON temp
FOR EACH ROW EXECUTE PROCEDURE unique_short_id();
Pouring test into temp:
INSERT INTO temp (foo, bar)
SELECT foo, bar
FROM test
RETURNING *
yields something like:
| foo | bar | id |
|------------+-----+----------|
| A | 1 | 9yt9XQwm |
| B | 2 | LCeiA-P8 |
If other tables have foreign key references on the test table or if test must remain online,
it may not be possible to drop test and rename temp to test.
Instead, it is safer to update test with the ids from temp.
Assuming test has a primary key (for concreteness, let's call it, testid), then
you could update test with the ids from temp using:
UPDATE test
SET id = temp.id
FROM temp
WHERE test.testid = temp.testid;
Then you could drop the temp table:
DROP TABLE temp;