Re: Sporadic relay failures using a hash in mynetworks.

[Dan Trainor <dan@xxxxxxxxxxxx>]

Wietse Venema wrote:

> Josef Karthauser:
>
> Checking application/pgp-signature: FAILURE
> -- Start of PGP signed section.
>
> > On Sun, Oct 17, 2004 at 04:08:32PM -0400, Wietse Venema wrote:
> >
> > > Josef Karthauser:
> > >
> > >
> > > > I don't see any evidence that the virtual.db map is failing at all so I
> > > > am reluctant to think that it a problem in the db lookup code.
> > > Consider how few lines of Postfix code are between the logging of the
> > > following lines:
> > >
> > >
> > > > Oct 15 13:01:57 jonah postfix/smtpd[45485]: match_hostaddr: 80.0.157.230 ~? hash:/usr/local/etc/postfix/mynetworks(0,100)
> > > > Oct 15 13:01:57 jonah postfix/smtpd[45485]: dict_lookup: 80.0.157.230 = (notfound)
> > > Consider that that same Postfix code does all access DB file lookups.
> > >
> > > Consider that the same Postfix DB module is used for all DB file lookups.
> > >
> > > Consider that no-one else is experiencing this.  
> > >
> > > Consider that I run Postfix on FreeBSD too.
> > >
> > > I think it is a site specific problem.
> > For the same reason considering how many other systems on the O/S use
> > the DB hash code, and that the DB hash code works well specifically for
> > virtual.db and aliases.db I can't see how it's an O/S problem either.
>
> The above considerations eliminate a lot of common elements.  Smart
> search focuses on the difference between maps that fail and maps
> that don't.
>
> One of those differences is that the files are different. Thus,
> the problem could be hardware: anything in the path from disk to
> main memory. For example, the VM system caches frequently accessed
> files. If your machine doesn't have ECC memory then cached content
> could change even when the file stays the same.
>
> This would show up by running MD5 hourly and watching the
> output change.
>
>
> > Do you have any suggestions as to how I might proceed to debug this
> > problem?  Where's the fragment of code in postfix which does the
> > reading? Maybe I can extract this and run it continuously to see how
> > often it fails.
>
> You're on the wrong track.  The same Postfix code is used by ALL
> Berkeley DB lookups. It's behavior can't depend on what file is
> being read.
>
> 	Wietse
Perhaps you can put them in a different location for testing, on the 
filesystem, and reflect those changes in the main.cf file, just for 
testing.
Just my $0.02 in troubleshooting, it might tell you something about the 
disk, if it's near death, having issues, etc etc.
Thanks
-dant