[Dnsmasq-discuss] Disappearing hosts from local DNS resolution

kyron at neuralbs.com kyron at neuralbs.com
Fri Oct 3 20:49:40 BST 2008


> Eric Thibodeau wrote:
>> Simon Kelley wrote:
>>> Eric Thibodeau wrote:
>>>> Hello,
>>>>
>>>>    I am using dnsmasq as a core component of my cluster. I'm having a
>>>> weird problem where the nodes are disappearing (can't use hostnames
>>>> anymore). So I am forced to connect to them using the IP addr. and
>>>> forcing dhcpcd to request an IP for the hosts to reappear in dnsmasq.
>>>> The logs show the following on the disappeared hosts:
>>>>
>>>> Oct  1 15:29:41 headless dnsmasq[7726]: DHCPREQUEST(eth1) 10.0.1.11
>>>> 00:01:03:df:ca:44
>>>> Oct  1 15:29:41 headless dnsmasq[7726]: DHCPNAK(eth1) 10.0.1.11
>>>> 00:01:03:df:ca:44 lease not found
>>>> ...
>>>> Oct  1 15:29:42 headless dnsmasq[7726]: DHCPDISCOVER(eth1)
>>>> 00:01:03:df:ca:44
>>>> Oct  1 15:29:42 headless dnsmasq[7726]: DHCPOFFER(eth1) 10.0.1.11
>>>> 00:01:03:df:ca:44
>>>> Oct  1 15:29:42 headless dnsmasq[7726]: DHCPDISCOVER(eth2)
>>>> 00:01:03:df:ca:44
>>>> Oct  1 15:29:42 headless dnsmasq[7726]: DHCPOFFER(eth2) 10.0.1.11
>>>> 00:01:03:df:ca:44
>>>> Oct  1 15:29:42 headless dnsmasq[7726]: DHCPREQUEST(eth1) 10.0.1.11
>>>> 00:01:03:df:ca:44
>>>> Oct  1 15:29:42 headless dnsmasq[7726]: DHCPACK(eth1) 10.0.1.11
>>>> 00:01:03:df:ca:44 thinkbig1
>>>>
>>>>   Here are two significant lines from my config file:
>>>>
>>>> dhcp-range=AthlonXP_1,10.0.1.10,10.0.1.126,255.255.255.0,12h
>>>>
>>>> dhcp-host=00:01:03:df:ca:44,net:AthlonXP_1,thinkbig1,10.0.1.11
>>>>
>>>> As you can note, I am defining a range with a lease time of 12h but
>>>> none for the statically defined hosts. Could this be the reason the
>>>> nodes disappear (lease time not sent to nodes if it's not defined in
>>>> the static definition)?
>>>
>>> Not a problem, if the lease time is not given, it defaults to the
>>> value for the range.
>> So I understood from the docs (IIRC).
>>>>
>>>> dnsmasq version is 2.45 and dhcpcd is 4.0.1-rc1
>>>
>>> The reaon the host is not in DNS is because there is no lease for it
>>> in dnsmasq's lease database. The dhcp-host line tells dnsmasq what
>>> address to offer, but it's the existence of a lease that gets
>>> reflected into the DNS,
>> Yes, I did understand this and is why I suspect it's related to the
>> lease time but my last attempt to tcpdump failed due to my ASSumption
>> that it wouldn't truncate packets. I'll try again...in about 12h when
>> nodes start disappearing ;) Once I get a nice trace, I'll poke at Roy
>> Marples at the same time.
>
> If the lease time is 12 hours and node disappear in 12 hours then
> dnsmasq maybe is doing the right thing. The clients should be renewing
> the lease after 6 hours or so. Can you see any evidence of that?

Nope, seems dhcpcd is not requesting a renew of the lease at 1/2 time as
it should (IIRC). http://wiki.neuralbs.com/~kyron/DHCP_reqs_tb17 contains
a grep of the DHCP activity from the disappearing nodes int
/var/log/messages.

> What does your network topology look like? Are you using DHCP relays
> and/or a not-fully-routed IP network? It's possiblr for hosts to be able
> to get a network, but not renew it under some circumstances.

Nope

>>> Are you doing something odd with the leases file?
>> Nope, nothing at all, no touching it.
> No on volatile storage which gets wiped over a reboot?

Yes but no, as in, the nodes are diskless so the entire environment is
volatile (hence the static IP assignements in the conf file) but the nodes
don't reboot between the disappearance.

Cheers,

Eric




More information about the Dnsmasq-discuss mailing list