I have a server with 20 pgsql clusters on it. I am seeing a slew of errors
like the following across 5 of these clusters, all running 7.3.4 on Linux
with kernel 2.4.18-4bigmem:
ERROR: Invalid page header in block N of R
N and R vary among block numbers and relation names, respectively. Looking
at the archives, it sounds like a possible hardware issue, and we're
investigating that. In the meantime, every query that attempts to access
those blocks errors out.
Taking the server down or losing data is very expensive, so I'm looking for
some short-term band-aids to re-enable full production access and recover
data while we sort out the hardware issues. When the corrupted block is
within a user index, I've been able to fix the corruption by dropping and
recreating the index. Not sure what I can do for user tables yet.
Preparing to run e2fsck/badblocks after I'm convinced there's nothing more
to salvage.
Is there anything I can do for invalid page header in a block of
pg_depend_reference_index?
Other suggestions?
TIA.
Ed