Re: Toast issues with OldestXmin going backwards

Greetings,

* Robert Haas (robertmhaas(at)gmail(dot)com) wrote:
> On Fri, Apr 27, 2018 at 11:35 AM, Andrew Gierth
> <andrew(at)tao11(dot)riddles(dot)org(dot)uk> wrote:
> >>>>>> "Robert" == Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> > Robert> One idea that occurred to me is to somehow record -- I guess in
> > Robert> pg_class using non-transactional updates -- the last cutoff XID
> > Robert> used to vacuum any given table. Then we could just make a rule
> > Robert> that you can't vacuum the TOAST table with an XID that's newer
> > Robert> than the last one used for the main table. That would preserve
> > Robert> the property that you can vacuum the tables separately while
> > Robert> avoiding dangling pointers. But that's obviously not
> > Robert> back-patchable,
> >
> > The suggestion made previously (in a historical thread) was to use an
> > entry in the reloptions field for this, at least in back branches. It
> > would be necessary for vacuum to add the entry initially in a normal
> > transactional update, after which it could be updated inplace.
>
> Yeah, I suppose. Sounds pretty rickety to me, though. Maybe I'm just
> a pessimist.

I tend to agree.. However, this isn't something that's been happening a
lot, from what I gather, and if we actually add a proper column into
pg_class for future versions (not really sure how I feel about if that
means "v11" or "v12" right now...) and reloptions for back-branches then
perhaps it's not so bad.

As far as a metadata page, it'd be pretty overkill, but maybe a fork for
it..? I'm trying to think if there might be anything else we'd be able
to put into such a fork since adding another inode to every relation
that'll only ever likely be 8k definitely wouldn't win us any fans. Not
sure, just brainstorming.

Thanks!

Stephen