[Logcheck-devel] Bug#574858: Bug#574858: logcheck: Does not ignore unresolvable hostname

Bob Proulx bob at proulx.com
Fri Apr 16 19:37:09 UTC 2010


Hannes von Haugwitz wrote:
> Bob Proulx wrote:
>  > I believe what was intended was the following:
> Fixed in 1.3.8.

Thanks!

>>   # Hostname either fully qualified or not.
>>   if [ $FQDN -eq 1 ]; then
>>           HOSTNAME=$(hostname --fqdn 2>/dev/null)
>> 	  test -z "$HOSTNAME" && HOSTNAME=$(hostname)
>>   else
>>           HOSTNAME=$(hostname)
>> 	  HOSTNAME=${HOSTNAME%%.*}
>>   fi
>
> With the above fix the error message shouldn't occur any longer. Why
> should logcheck bypass an unresolvable hostname? Wouldn't it be better
> if the administrator fixed the hostname issue instead?

A host has a hostname.  A host has network device adaptors.  Network
device adaptors have network addresses.  There isn't a requirement
that hosts have dns names registered.  There isn't a requirement that
a host have only one network address.  Indeed there are many multiply
homed hosts on the net with many addresses.  The method used by
'hostname --fqdn' uses to produce a fqdn isn't without debate the
correct way.  Actually the process of determining the FQDN of a host
for which it isn't specified in the hostname is complex and subtle.
(The BSD way of having the hostname be the fqdn is definitely simplest
but isn't without problems.)

I was setting up a host /here/ for use /there/.  Installing on one
private local network for use on a different private local network.
Both networks are using RFC1918 private addresses with local dns
servers.  The new machine will in the future on its destination
environment have a hostname that is registered in dns.  But in the
installation environment it does not.  I did not wish to install a
local dns name for it.  It is not needed nor required.  The system
uses DHCP to configure its network to obtain an IP address from the
dynamic address pool.  I type on the physical console.  I have no need
to connect to it from the network and so do not need to know the IP
address that it received.

I set up the machine and provision it.  Let it burn-in for several
days running stress testing applications.  Copy files onto it and get
it ready to replace a machine on the desination network.  During this
time period of days to weeks logcheck complains every hour about the
above hostname issue.

But regardless of how I ran into this problem in general it doesn't
seem good for logcheck to place a requirement that the host's hostname
be registered in dns.  A thin client wouldn't have this in every
configuration.  It just isn't needed and therefore shouldn't be a
dependency.  And in the case that it were a dependency then it
shouldn't produce an error so much as it should detect and report the
dependency failure.  But it isn't needed as a dependency.  Logcheck
otherwise works fine without the hostname registered in DNS.

Thanks,
Bob





More information about the Logcheck-devel mailing list