Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Oh, I never realized it would work with damage that severe.
I'm not sure it will, but it's his best shot.
Actually, I think there is a TODO item here. In most cases where
a page is being touched during WAL replay, we have a complete image of
the page stashed in WAL and are just going to overwrite what we find
in the data file. I think that the zero_damaged_pages code may be
barfing in situations where it has no need to. Someone needs to look
into that and see if we do have a problem; if so, what's the best fix?
regards, tom lane