Re: BUG #16036: Segmentation fault while doing an update

At last, i’ve found the source of the problem.

I have a trigger:

create function test_session_bu() returns trigger
    language plpgsql
as
$$
begin
  if new.session_end is not null and old.session_end is null then
    new.is_finished   := true;
    new.success_count := (select count(*) from tracking.test_result where session_id = new.id and status = 'success');
    new.failing_count := (select count(*) from tracking.test_result where session_id = new.id and status = 'failing');
    new.fail_count    := (select count(*) from tracking.test_result where session_id = new.id and status = 'fail');
  end if;
  return new;
end;
$$;

create trigger test_session_bu
    before update
    on tracking.test_session
    for each row
execute procedure tracking.test_session_bu();

Everything works fine without it, when i return it — problem returns too.

As you can see session_end is not affected in the queries. So i tried to modify the trigger to:

create or replace function test_session_bu() returns trigger
    language plpgsql
as
$$
begin
  return new;
end;
$$;
And problem was still in place.


> 4 окт. 2019 г., в 4:20, Andres Freund <andres@anarazel.de> написал(а):
> 
> Hi, 
> 
> On October 3, 2019 6:14:33 PM PDT, "Антон Власов" <druidvav@gmail.com> wrote:
>> Hello,
>> 
>> Looks like disabling jit didn’t affect backtrace at all:
>> 
>> 0  GetMemoryChunkContext (pointer=0x0) at
>> ./build/../src/include/utils/memutils.h:127
>> #1  pfree (pointer=0x0) at
>> ./build/../src/backend/utils/mmgr/mcxt.c:1033
>> #2  0x0000555d7276aaca in heap_freetuple (htup=<optimized out>) at
>> ./build/../src/backend/access/common/heaptuple.c:1340
>> #3  0x0000555d72918889 in tts_buffer_heap_clear (slot=0x555d73dbb118)
>> at ./build/../src/backend/executor/execTuples.c:652
>> #4  0x0000555d72918cbe in ExecClearTuple (slot=0x555d73dbb118) at
>> ./build/../src/include/executor/tuptable.h:428
>> #5  ExecResetTupleTable (tupleTable=0x555d73db5a10, shouldFree=false)
> 
> That's good - I was only looking at that because of the truncated backtrace. Did you do anything to make that work better this time?
> 
> Are there any details that you can provide? Schema? Any extensions in use? 
> 
> Does the problem happen always, or just under concurrency?
> 
> Andres
> -- 
> Sent from my Android device with K-9 Mail. Please excuse my brevity.