Re: logical replication launcher crash on buildfarm

On 01/04/17 04:19, Andres Freund wrote:
> On 2017-03-31 21:30:17 -0400, Robert Haas wrote:
>> On Fri, Mar 31, 2017 at 9:26 PM, Petr Jelinek
>> <petr(dot)jelinek(at)2ndquadrant(dot)com> wrote:
>>>> Hmm, I don't know if there's any good reason not to just use strcmp(),
>>>> but sure, OK. Committed and back-patched.
>>>
>>> Hmm culicidae still fails, this time only in parallel worker code. This
>>> didn't happen on my machine which is strange. Looking at the code, we
>>> are passing the fps->entrypoint as function pointer again so of course
>>> it fails. We have some code to load libraries again but even that gets
>>> initial entrypoint passed as function pointer
>>> (ParallelExtensionTrampoline). I wonder if we'll have to generalize the
>>> InternalBGWorkers even more to some kind of internal function name to
>>> pointer map and add the parallel entry points there as well.
>>
>> Argh, I forgot about that. I think we need to use something like
>> ParallelExtensionTrampoline all the time, not just for libraries.
>> Since effectively, we've determined that postgres itself has the same
>> problem.
>
> I wonder if we shouldn't just introduce a KnownFunctionPointer[] map,
> that can be used by various facilities (bgworkers themselves,
> parallelism, and probably more coming). Otherwise we'll introduce
> mechanisms like 7d8f6986b8 in multiple places.
>

Yes that's what I meant with what I said above.

--
Petr Jelinek http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services