Drazen Kacar <drazen.kacar@oradian.com> writes:
> It turned out that pg_filedump reported errors for 34 database files, but
> that was because all those files had one or more database pages (8k) at the
> end of the file zeroed out.
This can definitely happen after a database crash; it's not a problem
since the wasted space will be recovered by VACUUM.
I don't recall at the moment if there are any code paths that can cause
it without a crash. However, if you're looking directly at the disk
files it's definitely possible for there to be zeroes in a page that
has been obtained from the OS but never yet flushed out of shared
buffers.
regards, tom lane