Re: "invalid page header in block 597621 of relation..."error

Adam Witney <awitney(at)sgul(dot)ac(dot)uk> writes:
> bugasbase2=# vacuum;
> WARNING: relation "mba_data_base" page 597621 is uninitialized --- fixing

This is the expected result of what you did.

> WARNING: relation "mba_data_base" page 640793 is uninitialized --- fixing
> WARNING: relation "mba_data_base" page 640794 is uninitialized --- fixing
> WARNING: relation "mba_data_base" page 640795 is uninitialized --- fixing
> WARNING: relation "mba_data_base" page 640796 is uninitialized --- fixing
> WARNING: relation "mba_data_base" page 640797 is uninitialized --- fixing
> WARNING: relation "mba_data_base" page 640798 is uninitialized --- fixing

That's a bit odd. There are scenarios where all-zero pages can
legitimately appear in a PG file --- specifically, if PG extends
the table and the OS completes that task, but then there's a crash
before PG gets to write any data into the new page. Conceivably a
crash during a bulk data load process could result in half a dozen
such pages together, but it seems improbable. Try looking at the
data on the preceding and following pages --- does it look like there's
something missing?

regards, tom lane