Re: some last patches breaks plan cache

On 04/04/2018 07:54 PM, Tom Lane wrote:
> Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com> writes:
>> This should do the trick - I've failed to realize exec_stmt_call may
>> exit by calling elog(ERROR) too, in which case the plan pointer was not
>> reset.
>
>> This does fix the failures presented here, but I don't think it's the
>> right solution
>
> No, it's completely unacceptable.
>

Yes, I realize that and I was not really suggesting this as a proper
fix. It was meant more to demonstrate that it's still the same issue
with the same dangling pointer.

> If there's really no other way, you could use a PG_TRY block to
> ensure that the pointer gets reset on the way out. But I question
> why we've got a design that requires that in the first place. It's
> likely to have more problems than this.
>

I agree it needs a solution that does not require us to track and
manually reset pointers on random places. No argument here.

regards

--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services