Bug#297607: Long pause when exim daemon is being stopped (was: Re: exim4 prone to break d-i, bug #297607)

[Marc Haber]

On Sun, Mar 27, 2005 at 12:09:57AM +0100, Frans Pop wrote:
> On Saturday 26 March 2005 06:47, Marc Haber wrote:
> > > The only thing I did notice is that, if I run 'dpkg-reconfigure
> > > exim4-config' _after_ the installation is completed, "restarting MTA"
> > > takes a very long time (something like 30 seconds), but only the
> > > first time I do it.
> >
> > That sounds like a DNS issue. Exim tries to resolve its own hostname,
> > and if that cannot be done, it waits for a DNS timeout. Usual remedy
> > is making sure that the local hostname (hostname _and_ FQDN) is
> > resolvable via /etc/hosts, since for a new install the hostname is
> > unlikely to be available via DNS. See the second FAQ question in
> > README.Debian
> 
> No, that's not it. Both systems are known on my DNS server. It could still 
> well be a DNS issue, but nothing that easy.

An ipv6 issue, maybe? How does your system behave when you search for
an AAAA Record for your hostname?  Does /etc/hostname have an FQDN or
only a name? What does exim do when you set "dns_ipv4_lookup = *" in
the main configuration setup? What does exim do when you set
"primary_hostname = your_hostname" in the main configuration setup?

> > > During recent normal installations from Sarge (on different archs),
> > > I've seen a similar problem: on the first shutdown after installation
> > > it takes a very long time to "stop MTA".
> >
> > That's a new one. Can you try setting EX4DEBUG to a non-empty value
> > and see where the delay is happening?
> 
> I've reproduced it just now during an installation using the sparc64 RC3 
> netinst CD. The pause during power off is at the "start-stop-daemon 
> --stop" command.
> The pause is about the same as when exim is being restarted during 
> reconfiguration in the first situation.

I see. Is this reproducible in a system state where stracing both
start-stop-daemon and the actual daemon is possible? Does the delay
also happen when you manually determine the PID and sent it a SIGTERM
directly?

> During previous installations with "local mail only", the default for the 
> hostname would mostly be the system name. Now I see 
> "localhost.localdomain" in some (all?) installations. Could that be the 
> root cause for both these issues?
> 
> This could be something from the d-i installation.
> On this new Sparc installation I get in /etc/hosts (DHCP with fixed 
> address):
> 127.0.0.1       localhost.localdomain   localhost       gimli
> 
> On another installation I have:
> 127.0.0.1       localhost.localdomain   localhost
> 10.19.66.2      elrond.fjphome.nl       elrond
> 
> So on the new install the FQN is missing in /etc/hosts, but it is also 
> readily supplied by my DNS server:
> fjp@gimli:~$ ping gimli.fjphome.nl
> PING gimli.fjphome.nl (10.19.66.9) 56(84) bytes of data.
> 64 bytes from gimli.fjphome.nl (10.19.66.9): icmp_seq=1 ttl=64 time=0.273 
> ms
> 64 bytes from gimli.fjphome.nl (10.19.66.9): icmp_seq=2 ttl=64 time=0.083 
> ms

Is ipv6 part of the game?

Greetings
Marc

-- 
-----------------------------------------------------------------------------
Marc Haber         | "I don't trust Computers. They | Mailadresse im Header
Mannheim, Germany  |  lose things."    Winona Ryder | Fon: *49 621 72739834
Nordisch by Nature |  How to make an American Quilt | Fax: *49 621 72739835