On Fri, Nov 22, 2024 at 2:26 PM Heikki Linnakangas <[email protected]> wrote:
> On 22/11/2024 15:02, Matthias van de Meent wrote:
> > I think I'd go with a patch like attached, where the bulk writer
> > registers that it started with .relsize pages in the relfork, and use
> > that for smgrextend() decisions. It now also records pages_written as
> > a separate but accurate value.
>
> Looks good to me. Eric, can you confirm that Matthias's patch fixes the
> problem for you?
>
>
Yes, it solves the issue so it looks good.
Just a minor nit: the code uses both blokno as local variable for
pending_writes[i].blkno and directly accessing pending_writes[i].blkno.
Maybe it is better to just use the local variable. For example, change
++ b/src/backend/storage/smgr/bulk_write.c
@@ -304,7 +304,8 @@ smgr_bulk_flush(BulkWriteState *bulkstate)
}
smgrextend(bulkstate->smgr, bulkstate->forknum,
blkno, page, true);
- bulkstate->relsize = pending_writes[i].blkno + 1;
+ bulkstate->relsize++;
+ Assert(bulkstate->relsize == blkno + 1);
Just a suggestion.
Thanks for the quick action!
-Erik
