Re: Theory about XLogFlush startup failures - Mailing list pgsql-hackers

From Tom Lane
Subject Re: Theory about XLogFlush startup failures
Date
Msg-id 5854.1011133171@sss.pgh.pa.us
Whole thread Raw
In response to Re: Theory about XLogFlush startup failures  ("Hiroshi Inoue" <Inoue@tpf.co.jp>)
List pgsql-hackers
"Hiroshi Inoue" <Inoue@tpf.co.jp> writes:
> What I have minded is e.g. the following case.
> Undoubtedly the page is corrupted(too big offset number).
> I'm suspicious if other pages are safe under such a situation.

You have a point, but I still don't like slowing down PageGetItem.

How about this instead: whenever we read in a page, check to see
if its page header data is sane.  We could do this right after the
smgrread call in ReadBufferInternal, and follow the "status = SM_FAIL"
exit path if we see trouble.
        regards, tom lane


pgsql-hackers by date:

Previous
From: Jan Wieck
Date:
Subject: Re: FATAL 1: Relation 'pg_shadow' does not exist
Next
From: mlw
Date:
Subject: RServ replication