Thread: BUG #18517: Dropping a table referenced by an initially deferred foreign key fails with an error

The following bug has been logged on the website:

Bug reference:      18517
Logged by:          Alexander Lakhin
Email address:      exclusion@gmail.com
PostgreSQL version: 17beta1
Operating system:   Ubuntu 22.04
Description:

The following script:
CREATE TABLE pkt (id int PRIMARY KEY);
CREATE TABLE fkt (id int PRIMARY KEY, fk int REFERENCES pkt INITIALLY
DEFERRED);
INSERT INTO pkt VALUES (1);
BEGIN;
INSERT INTO fkt VALUES (101, 1);

DROP TABLE pkt CASCADE;
COMMIT;

fails on commit with a weird error:
ERROR:  relation 16390 has no triggers

BACKTRACE:
afterTriggerInvokeEvents at trigger.c:4700:7
AfterTriggerFireDeferred at trigger.c:5224:6
CommitTransaction at xact.c:2219:8
CommitTransactionCommandInternal at xact.c:3169:18
CommitTransactionCommand at xact.c:3099:9
finish_xact_command at postgres.c:2783:3
exec_simple_query at postgres.c:1299:4
PostgresMain at postgres.c:4684:27
BackendInitialize at backend_startup.c:123:1
postmaster_child_launch at launch_backend.c:269:9
BackendStartup at postmaster.c:3593:8
ServerLoop at postmaster.c:1677:10
PostmasterMain at postmaster.c:1372:11
startup_hacks at main.c:217:1

afterTriggerInvokeEvents() coding is:
                /*
                 * So let's fire it... but first, find the correct relation
if
                 * this is not the same relation as before.
                 */
                if (rel == NULL || RelationGetRelid(rel) !=
evtshared->ats_relid)
                {   
...
                    if (trigdesc == NULL)   /* should not happen */
                        elog(ERROR, "relation %u has no triggers",
                             evtshared->ats_relid);

Reproduced on REL_12_STABLE .. master.

PG Bug reporting form <noreply@postgresql.org> writes:
> The following script:
> CREATE TABLE pkt (id int PRIMARY KEY);
> CREATE TABLE fkt (id int PRIMARY KEY, fk int REFERENCES pkt INITIALLY
> DEFERRED);
> INSERT INTO pkt VALUES (1);
> BEGIN;
> INSERT INTO fkt VALUES (101, 1);

> DROP TABLE pkt CASCADE;
> COMMIT;

> fails on commit with a weird error:
> ERROR:  relation 16390 has no triggers

Oh, interesting.  Furthermore, if I add an unrelated trigger to fkt
so that there's still a trigdesc, then I reach

ERROR:  XX000: could not find trigger 42452
LOCATION:  AfterTriggerExecute, trigger.c:4303

So what this demonstrates is that we should not consider it an
error if afterTriggerInvokeEvents finds an event for a trigger
that no longer exists.

Perhaps an alternative could be to run through the pending events
at DROP time and throw an error if there's an unfired event for the
trigger.  That doesn't sound better to me though.  I see no very
good reason why we shouldn't allow the above sequence of commands
and treat the constraint as no longer to be enforced.

            regards, tom lane