Re: [BUG] Logical replication failure "ERROR: could not map filenode "base/13237/442428" to relation OID" with catalog modifying txns

On Tue, Jul 19, 2022 at 7:28 PM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
>
> On Tue, Jul 19, 2022 at 9:25 PM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> >
> > On Tue, Jul 19, 2022 at 1:10 PM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
> > >
> >
> > > BTW on backbranches, I think that the reason why we add
> > > initial_running_xacts stuff to ReorderBuffer is that we cannot modify
> > > SnapBuild that could be serialized. Can we add a (private) array for
> > > the initial running xacts in snapbuild.c instead of adding new
> > > variables to ReorderBuffer?
> > >
> >
> > While thinking about this, I wonder if the current patch for back
> > branches can lead to an ABI break as it changes the exposed structure?
> > If so, it may be another reason to change it to some other way
> > probably as you are suggesting.
>
> Yeah, it changes the size of ReorderBuffer, which is not good.
>

So, are you planning to give a try with your idea of making a private
array for the initial running xacts? I am not sure but I guess you are
proposing to add it in SnapBuild structure, if so, that seems safe as
that structure is not exposed.

> Changing the function names and arguments would also break ABI. So
> probably we cannot do the above idea of removing
> ReorderBufferInitialXactsSetCatalogChanges() as well.
>

Why do you think we can't remove
ReorderBufferInitialXactsSetCatalogChanges() from the back branch
patch? I think we don't need to change the existing function
ReorderBufferXidHasCatalogChanges() but instead can have a wrapper
like SnapBuildXidHasCatalogChanges() similar to master branch patch.

--
With Regards,
Amit Kapila.