Patrick Welche <prlw1@newn.cam.ac.uk> writes:
> Indeed, the plain -d dump says that I have a chunk of /var/mail/prlw1
> in 1000-13ff. No wonder postgres complained!
Yipes. We have seen this sort of thing once or twice in the past.
I don't know whether you are looking at a disk drive fault (dropping
the right data onto the wrong sector) or a kernel fault (getting
confused about which buffer holds which file block), but I believe
it's one or the other. It's hard to see how anything at the application
level could have gotten those two files switched. You might look to see
if there are any updates available for your kernel.
As for recovery, you probably want to drop and reload at least that one
table. Whether it's worth a complete reload is your call.
regards, tom lane