Re: display offset along with block number in vacuum errors

On Fri, Jul 31, 2020 at 04:55:14PM -0500, Justin Pryzby wrote:
> Bcc: 
> Subject: Re: display offset along with block number in vacuum errors
> Reply-To: 
> In-Reply-To: <CAKYtNApLJjAaRw0UEBBY6G1o0LRZKS7rA5n46BFh+NfwSOycdg@mail.gmail.com>

whoops

> On Wed, Jul 29, 2020 at 12:35:17AM +0530, Mahendra Singh Thalor wrote:
> > > Here:
> > >
> > > @@ -1924,14 +1932,22 @@ lazy_vacuum_page(Relation onerel, BlockNumber blkno, Buffer buffer,
> > >                 BlockNumber tblk;
> > >                 OffsetNumber toff;
> > >                 ItemId          itemid;
> > > +               LVSavedErrInfo loc_saved_err_info;
> > >
> > >                 tblk = ItemPointerGetBlockNumber(&dead_tuples->itemptrs[tupindex]);
> > >                 if (tblk != blkno)
> > >                         break;                          /* past end of tuples for this block */
> > >                 toff = ItemPointerGetOffsetNumber(&dead_tuples->itemptrs[tupindex]);
> > > +
> > > +               /* Update error traceback information */
> > > +               update_vacuum_error_info(vacrelstats, &loc_saved_err_info, VACUUM_ERRCB_PHASE_VACUUM_HEAP,
> > > +                                                                blkno, toff);
> > >                 itemid = PageGetItemId(page, toff);
> > >                 ItemIdSetUnused(itemid);
> > >                 unused[uncnt++] = toff;
> > > +
> > > +               /* Revert to the previous phase information for error traceback */
> > > +               restore_vacuum_error_info(vacrelstats, &loc_saved_err_info);
> > >         }
> > >
> > > I'm not sure why you use restore_vacuum_error_info() at all.  It's already
> > > called at the end of lazy_vacuum_page() (and others) to allow functions to
> > > clean up after their own state changes, rather than requiring callers to do it.
> > > I don't think you should use it in a loop, nor introduce another
> > > LVSavedErrInfo.
> > >
> > > Since phase and blkno are already set, I think you should just set
> > > vacrelstats->offnum = toff, rather than calling update_vacuum_error_info().
> > > Similar to whats done in lazy_vacuum_heap():
> > >
> > >                 tblk = ItemPointerGetBlockNumber(&vacrelstats->dead_tuples->itemptrs[tupindex]);
> > >                 vacrelstats->blkno = tblk;
> > 
> > Fixed.
> 
> I rearead this thread and I think the earlier suggestion from Masahiko was
> right.  The loop around dead_tuples only does ItemIdSetUnused() which updates
> the page, which has already been read from disk.  On my suggestion, your v2
> patch sets offnum directly, but now I think it's not useful to set at all,
> since the whole page is manipulated by PageRepairFragmentation() and
> log_heap_clean().  An error there would misleadingly say "..at offset number
> MM", but would always show the page's last offset, and not the offset where an
> error occured.

This makes me question whether offset numbers are ever useful during
VACUUM_HEAP, since the real work is done a page at a time (not tuple) or by
internal functions that don't update vacrelstats->offno.  Note that my initial
problem report that led to the errcontext implementation was an ERROR in heap
*scan* (not vacuum).  So an offset number at that point would've been
sufficient.
https://www.postgresql.org/message-id/20190808012436.GG11185@telsasoft.com

I mentioned that lazy_check_needs_freeze() should save and restore the errinfo,
so an error in heap_page_prune() (for example) doesn't get the wrong offset
associated with it.

So see the attached changes on top of your v2 patch.

postgres=# DROP TABLE tt; CREATE TABLE tt(a int) WITH (fillfactor=90); INSERT INTO tt SELECT generate_series(1,99999);
VACUUMtt; UPDATE tt SET a=1 WHERE ctid='(345,10)'; UPDATE pg_class SET relfrozenxid=(relfrozenxid::text::int +
(1<<30))::int::text::xidWHERE oid='tt'::regclass; VACUUM tt;
 
ERROR:  found xmin 1961 from before relfrozenxid 1073743785
CONTEXT:  while scanning block 345 of relation "public.tt", item offset 205

Hmm.. is it confusing that the block number is 0-indexed but the offset is
1-indexed ?

-- 
Justin