Re: Theory about XLogFlush startup failures

Tom Lane wrote:
> 
> Hiroshi Inoue <Inoue@tpf.co.jp> writes:
> > One thing I can think of is to prevent a corrupted page
> > from spoiling other pages by jumping the page boundary
> > in the buffer pool.
> >>
> >> We do that already, no?
> 
> > Oh I may be missing something.
> > Where is it checked ?
> 
> I know PageRepairFragmentation is real paranoid about this, because I
> made it so recently.  I suppose it might be worth adding some more
> sanity checks to PageAddItem, maybe PageZero (is that ever called on a
> pre-existing page?), and PageIndexTupleDelete.  Seems like that should
> about cover it --- noplace else inserts items on disk pages or
> reshuffles disk page contents, AFAIK.

What about PageGetItem ? It seems to be able to touch the item
via HeapTupleSatisfies etc. 

regards,
Hiroshi Inoue