Re: Cache lookup failed for type 70385664

This looks like data corruption on that record. Or possibly on
multiple records.

I would:

a) update to the latest bug-fix release of 8.2 asap. I don't see any
fixed bugs which would cause this specific type of error but there are
a lot of them and I could have missed it.

http://www.postgresql.org/docs/8.2/static/release-8-2-13.html
http://www.postgresql.org/docs/8.2/static/release-8-2-12.html
http://www.postgresql.org/docs/8.2/static/release-8-2-11.html
http://www.postgresql.org/docs/8.2/static/release-8-2-10.html
http://www.postgresql.org/docs/8.2/static/release-8-2-9.html
http://www.postgresql.org/docs/8.2/static/release-8-2-8.html
http://www.postgresql.org/docs/8.2/static/release-8-2-7.html
http://www.postgresql.org/docs/8.2/static/release-8-2-6.html
http://www.postgresql.org/docs/8.2/static/release-8-2-5.html
http://www.postgresql.org/docs/8.2/static/release-8-2-4.html

I would also run a good memory checker on this machine, like memtest86
or something like that. Also, check SMART diagnostics for your drives
or the raid status.

But you'll still have to deal with filtering out the corrupted records
from your data. You should be able to delete records even if
individual data within them is corrupted. If you have corruption in
other places you could have a trickier time though.

If you get the "ctid" column from the bad records you could use dd to
extract the block they're on (the first number in the ctid in 8kB
blocks) and post that. It's possible the nature of the corruption
would be clear -- such as a single bit error which makes bad memory a
prime suspect or random bits of garbage from another type of file
which makes the filesystem a suspect. But it's not going to really
help you fix the problem much.

--
greg
http://mit.edu/~gsstark/resume.pdf