Re: Zeroing damaged pages

[Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us>]

Simon Riggs wrote:
> On Thu, 2006-02-23 at 11:54 -0500, Tom Lane wrote:
> > Simon Riggs <simon(at)2ndquadrant(dot)com> writes:
> > > A patch prototype to make zero_damaged_pages work as advertised is
> > > enclosed, though the current behaviour may well be preferred, in which
> > > case a doc patch is more appropriate. 
> > 
> > I don't think this is a good idea, and even if it were, the proposed
> > patch is a model of obscurantism.
> 
> ;-)
> 
> Just some reflections on a recent db recovery for a client.
> 
> > > However, since autovacuum the window of opportunity for support to
> > > assist with data recovery is smaller and somewhat random.
> > 
> > Hmm .... it'd probably be a good idea to force zero_damaged_pages OFF in
> > the autovac subprocess.  That parameter is only intended for interactive
> > use --- as you say, autovac would be a rather nasty loose cannon if it
> > fired up with this parameter ON.
> 
> We can:
> - disable zero_damaged_pages in autovac

I am wondering if we should prevent autovac from running if
zero_damaged_pages is set in postgresql.conf.  I can imagine autovac
aborting when trying to read a damaged page if zero_damaged_pages is
ignored.

We can't prevent zero_damaged_pages from being set in postgresql.conf
because it is possible for the system to be so corrupted that you can't
even start a backend without it being set.

-- 
  Bruce Momjian   http://candle.pha.pa.us
  SRA OSS, Inc.   http://www.sraoss.com

  + If your life is a hard drive, Christ can be your backup. +