On Mon, 2007-12-31 at 01:27 -0500, Tom Lane wrote:
> Gregory Stark <stark@enterprisedb.com> writes:
> > "Tom Lane" <tgl@sss.pgh.pa.us> writes:
> >> I wonder whether it's just a coincidence that these have the same offset
> >> number...
>
> > I can't imagine any Postgres bug which would depend on the offsets
> > being the same. But what I could imagine is filesystem corruption
> > which copied the block to someplace else in the table or possibly has
> > even mapped the same block into two different places in the table.
>
> That idea was in my mind too, but Mason stated that the rows showed
> different "updated_at" values, so they couldn't be mirror images of that
> sort.
Tom,
I think you misread Mason's post of 20:23 GMT-6 where he says the
created_at values are the *same*, not different. Mason's previous bug
report 3724 also had duplicate rows with matching created_at values.
So mangling block numbers and re-writing blocks in the wrong place is a
possibility. Filesystem corruption is just one way that can occur.
--
Simon Riggs
2ndQuadrant http://www.2ndQuadrant.com