Matthias(dot)Pitzl(at)izb(dot)de writes:
> invalid page header in block 437702 of relation "xxxx"
I concur with Scott that this sounds suspiciously like a hardware
problem ... but have you tried dumping out the bad pages with
pg_filedump or even just od? The pattern of damage would help to
confirm or disprove the theory.
You can find pg_filedump source code at
http://sources.redhat.com/rhdb/
regards, tom lane