Re: BUG #16139: Assertion fails on INSERT into a postgres_fdw' table with two AFTER INSERT triggers

On Wed, Nov 27, 2019 at 8:35 PM Etsuro Fujita <etsuro(dot)fujita(at)gmail(dot)com> wrote:
> A bit of investigation showed that this is caused by commit
> ff11e7f4b9ae017585c3ba146db7ba39c31f209a. I haven't yet looked at the
> commit in any detail, though.

I spent some time studying the commit. I think AfterTriggerExecute()
should make sure that:

* foreign tables always use zero and save the tuple(s) to a
* tuplestore. AFTER_TRIGGER_FDW_FETCH directs AfterTriggerExecute() to
* retrieve a fresh tuple or pair of tuples from that tuplestore, while
* AFTER_TRIGGER_FDW_REUSE directs it to use the most-recently-retrieved
* tuple(s). This permits storing tuples once regardless of the number of
* row-level triggers on a foreign table.

But the commit modified that function to clear the tg_trigslot and
tg_newslot slots' contents, which are reused when
AFTER_TRIGGER_FDW_REUSE, as shown below:

@@ -4374,12 +4373,17 @@ AfterTriggerExecute(AfterTriggerEvent event,
heap_freetuple(rettuple);

/*
- * Release buffers
+ * Release resources
*/
- if (buffer1 != InvalidBuffer)
- ReleaseBuffer(buffer1);
- if (buffer2 != InvalidBuffer)
- ReleaseBuffer(buffer2);
+ if (should_free_trig)
+ heap_freetuple(LocTriggerData.tg_trigtuple);
+ if (should_free_new)
+ heap_freetuple(LocTriggerData.tg_newtuple);
+
+ if (LocTriggerData.tg_trigslot)
+ ExecClearTuple(LocTriggerData.tg_trigslot);
+ if (LocTriggerData.tg_newslot)
+ ExecClearTuple(LocTriggerData.tg_newslot);

Attached is a patch for fixing that. Maybe I'm missing something, though.

Best regards,
Etsuro Fujita