BUG #18531: err when using 'current of' with incremental COMMIT
The following bug has been logged on the website:
Bug reference: 18531
Logged by: Dzmitry Jachnik
Email address: dzja112@gmail.com
PostgreSQL version: 14.2
Operating system: x86_64 GNU/Linux
Description:
I tried using 'current of' syntax (like doc. 43.7.3.3.)
and had have ERROR: cursor "bulk_collection_cur" is held from a previous
transaction
If using regular ` WHERE id = l_rec.id; ` I hadn't any error
EXAMPLE:
```
CREATE OR REPLACE PROCEDURE crm_import.etlp_deals(IN i_id bigint DEFAULT
NULL::bigint, IN i_limit integer DEFAULT null)
LANGUAGE plpgsql
AS $procedure$
DECLARE
result jsonb;
COUNTER int := 0;
bulk_collection_cur CURSOR FOR
SELECT *
FROM crm_import.deals dew
WHERE COALESCE(dew.etl_stage,'NOT_LOADED') not IN ('LOADED',
'IGNORE')
-- AND dew.id = any(l_id)
AND (dew.id = i_id OR i_id IS null)
ORDER BY dew.load_date, dew.dt
FOR UPDATE ;
begin
<<COLLECTION>>
BEGIN
FOR l_rec IN bulk_collection_cur
LOOP
COUNTER := COUNTER + 1;
RESULT := crm_import.etl_deals(row_to_json(l_rec.*)::jsonb);
UPDATE crm_import.deals dsp
SET etl_protocol = RESULT
-------------------
--v.1 SQL Error [24000]: ERROR: cursor "bulk_collection_cur" is
held from a previous transaction
WHERE CURRENT OF bulk_collection_cur;
--v.2 WITHOUT ERROR
--WHERE id = l_rec.id;
-------------------
RAISE NOTICE 'counter= %', counter;
IF mod(counter, i_limit) = 0 THEN
RAISE NOTICE 'COMMIT';
COMMIT;
END IF;
END LOOP;
END COLLECTION;
end $procedure$
;
```
That combine 'current of' and 'commit' at one loop is bad idea?
PG Bug reporting form <noreply@postgresql.org> writes:
That combine 'current of' and 'commit' at one loop is bad idea?
Yeah, that's not going to work. The first COMMIT auto-holds the
cursor [1]https://www.postgresql.org/docs/current/plpgsql-transactions.html, and after that there is not an active cursor query for
the WHERE CURRENT OF to refer to. Even if that data were somehow
retained, the COMMIT would have dropped the tuple locks taken by
the cursor query, so that the semantics would be a lot squishier
than you'd want.
The discussion at [1]https://www.postgresql.org/docs/current/plpgsql-transactions.html fails to mention that there are any such
gotchas implicit in the auto-hold behavior, so I added a little
bit of text [2]https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=a0899c0a97d19b0c330ff885af4b78b3e7efb100.
regards, tom lane
[1]: https://www.postgresql.org/docs/current/plpgsql-transactions.html
[2]: https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=a0899c0a97d19b0c330ff885af4b78b3e7efb100