Weird network problem

[Phil Dobbin]

On 05/26/2013 02:54 PM, Mikkel L. Ellertson wrote:

> On 05/26/2013 07:18 AM, Phil Dobbin wrote:
>> On 05/26/2013 11:54 AM, Mikkel L. Ellertson wrote:
>>> On 05/25/2013 08:00 PM, Phil Dobbin wrote:
>>>> Hi, all.
>>>>
>>>> I've got several machines on a LAN behind a NAT with DHCP assigning
>>>> always the same addresses from a dynamic IP.
>>>>
>>>> A couple of days ago the IP changed & since then, one of the machines
>>>> running Fedora 17 always fails first time to connect to the network:
>>>> launch Thunderbird, no start screen, first attempt to check mail, it
>>>> tells me that there's no network connection, second attempt it connects.
>>>>
>>>> The scheduled DejaDup backup always fails with no network but will run
>>>> manually no problem. Firefox can't find Google but the Nagios
>>>> web interface is fine as is all the cli stuff (ping, ssh, etc).
>>>>
>>>> Most annoyingly, yum update goes through every mirror before partially
>>>> downloading part of the updates & if the updates are large, it takes
>>>> about three attempts to get them all installed.
>>>>
>>>> I'd like to clear this up naturally especially as in the next couple of
>>>> weeks I'll be upgrading this box to Fedora 18 & the last thing I need is
>>>> a dodgy network connection.
>>>>
>>>> All the machines below are on the same LAN & they all work fine after
>>>> the IP address change, it's only the Fedora box that's causing problems.
>>>>
>>>> Any help appreciated. I'm stuck.
>>>>
>>>> Cheers,
>>>>
>>>>   Phil...
>>>>
>>> Check your name server settings. Does /etc/resolv.conf have a name
>>> server that from the old IP address? Do you have one machine on the
>>> network that runs a catching name server and the rest of the Fedora
>>> machines are looking for it at the old address? Or are you running
>>> something like dnsmasq on the machines, and have the old IP address
>>> in the config file?
>> I've never used any name server settings on any of these machines. The
>> lease is automatically assigned by DHCP so therefore there is no need to.
>>
>> The external link comes into a NAT router then onto a HP ProCurve switch
>> & then via cat5 cables to each machine (there's no wireless involved
>> anywhere). Then each machine in Network Settings uses the automatic
>> setting to assign each address & DNS (192.168.1.254) & address mask
>> (255.255.255.0).
>>
>> As I've said, all other machines (all 16 of them) are fine. It's just
>> the Fedora box which leads me to suspect that Fedora's doing, or not
>> doing something, to cause this.
>>
>> Cheers,
>>
>>   Phil...
>>
> Dumb question - have you checked the network connection? See if
> changing the cable or the port on the switch helps. It seams strange
> that the external IP address changing would cause this, but for some
> strange reason hardware problems totally unrelated to the change
> seam to pick that time to happen.

I put a brand new cat 5 into another HP ProCurve that I have racked as a
backup & still the same:

'Could not get metalink
https://mirrors.fedoraproject.org/metalink?repo=updates-released-f17&arch=x86_64
error was
14: curl#6 - "Could not resolve host: mirrors.fedoraproject.org; Name or
service not known"'

Thunderbird still needs three manual attempts to connect too (as you can
see, it does work eventually).

I've rebooted, looked in logs. I'm at a loss...

Cheers,

  Phil...

-- 
currently (ab)using
CentOS 5.9 & 6.4, Debian Squeeze & Wheezy, Fedora Beefy, Spherical &
That Damn Cat, Lubuntu 12.10, OS X Snow Leopard & Ubuntu Precise,
Quantal & Raring
GnuPG Key : http://www.horse-latitudes.co.uk/publickey.asc