Re: System crash - invalid page header messages in log

On Tue, Sep 28, 2004 at 04:18:44PM -0400, Jim Buttafuoco wrote:
> One of my systems crashed today and when Postgres started it gave the
> following warnings. Is this OK? I am going to find which database
> has these relations and do some checking. It would be nice if the
> startup wal code gave the database oid also and database version.

Is this running on XFS?

The "invalid page header" messages should not cause you any problem:
what happenned is that while recovery was on progress, some pages were
found to be invalid. This is no big deal because they can be recovered
from XLog (which is why they are zeroed out automatically in the first
place).

Why would they be corrupted is another matter ... partial writes may
cause bad pages, but I think the header is written before the contents
when the page is fsync'ed.

However, there is some zeroing behaviour on XFS under certain
conditions. That could explain the heavy corruption you are seeing.
Hopefully no other page should have been zeroed, else it would have had
a snapshot on XLog.

> 2004-09-28 16:10:52 [979] LOG: incorrect resource manager data checksum in record at F4/72BEFA0

This one may be more problematic ... not sure. Maybe it means that a
XLog record wasn't fully written or something.

--
Alvaro Herrera (<alvherre[a]dcc.uchile.cl>)
"No es bueno caminar con un hombre muerto"