Re: Allow ERROR from heap_prepare_freeze_tuple to be downgraded to WARNING

On Mon, Jul 20, 2020 at 10:14 PM Alvaro Herrera
<alvherre(at)2ndquadrant(dot)com> wrote:
>
> On 2020-Jul-20, Dilip Kumar wrote:
>
> > On Fri, Jul 17, 2020 at 4:16 PM Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:
>
> > > So if the vacuum_tolerate_damage is set then in
> > > all the cases in heap_prepare_freeze_tuple where the corrupted xid is
> > > detected, it will emit a warning and return that nothing is changed in
> > > the tuple and the 'tuple_totally_frozen' will also be set to false.
> > > Since we are returning false the caller will not try to freeze such
> > > tuple and the tuple_totally_frozen is also set to false so that the
> > > page will not be marked to all frozen even if all other tuples in the
> > > page are frozen.
>
> > Robert has mentioned at [1] that we probably should refuse to update
> > 'relfrozenxid/relminmxid' when we encounter such tuple and emit
> > WARNING instead of an error.
>
> Isn't this already happening per your description above?

As per the above description, we are avoiding to set the page as all
frozen. But the vacrelstats->scanned_pages count has already been
increased for this page. Now, right after the lazy_scan_heap, we
will update the pg_class tuple with the new FreezeLimit and
MultiXactCutoff.

>
> > I think we shall do that in some cases
> > but IMHO it's not a very good idea in all the cases. Basically, if
> > the xmin precedes the relfrozenxid then probably we should allow to
> > update the relfrozenxid whereas if the xmin precedes cutoff xid and
> > still uncommitted then probably we might stop relfrozenxid from being
> > updated so that we can stop CLOG from getting truncated.
>
> I'm not sure I understand 100% what you're talking about here (the first
> half seems dangerous unless you misspoke), but in any case it seems a
> pointless optimization. I mean, if the heap is corrupted, you can hope
> to complete the vacuum (which will hopefully return which *other* tuples
> are similarly corrupt) but trying to advance relfrozenxid is a lost
> cause.

I agree with your point. I think we just need to avoid advancing the
relfrozenxid in all such cases.

--
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com