incompatibility between named dns and bonding

[Daniele Bernazzi]

Hi, I have a dl380 g3 with suse sles9 i386. In my configuration I teamed the two eths for redundancy, so I have a static ip configured on bond0 which enslaves eth0 and eth1. So far, so good: it works fine. Problems arises after the setup of the dns server (bind-9.2.3-76.14) service: the server has lots of net discontinuity. To overcome the problem I did have to disable bonding and configure just one eth. Now it works fine, but net redundancy is lost and my question is: how to have both features (bond and dns) working toghether?
As a reference, kernel is 2.6.5-7.151-smp, in log there are no entries related to the problem, hp support pack is not installed because it did work with just suse updates and now I am a bit worried to install new drivers on a production server.
Thank you

[Vitaly Karasik_1]

Did you try to stress network by other services (ftp, NFS and so on)?
IMHO, the problem isn't bind + bonding but bonding itself.
Which NICs/driver do you use?

[Daniele Bernazzi]

On the server there are other net services (proxy and mail for about 1000 users, web), so the net is stressed. Bit rate is about 2Mb/s on business hours. The nics used are the broadcom gigabit nc7781 embedded on the main board

[Steven E. Protter]

Properly bonded NIC cards should end up with one single IP address.

I agree with Vitaly Karasik on this one. I have built several servers with bonded NIC cards in the US and none of them have exibited this behavior.

The network connectivity problem may be from HOW you did the bonding.

I'd like to see the (if need be altered) output from

ifconfig bond0

Also ethtool and other utility output may explain the problem.

There may be a problem in the network config files of the two eth cards, or the ifcfg-bond0 config file. Also, there is a needed entry to make this work right that I insert into the /etc/init.d/network script.

Please post the procedure or cookbook that you used.

SEP

[Daniele Bernazzi]

now bond0 is down, so ifconfig does not work.
Anyway in /etc/sysconfig/network I have this script for bond (when active):
BOOTPROTO='static'
BROADCAST='159.213.51.255'
IPADDR='159.213.51.131'
NETMASK='255.255.255.0'
MTU=''
REMOTE_IPADDR=''
STARTMODE='onboot'
BONDING_MASTER='yes'
BONDING_MODULE_OPTS='miimon=100 mode=active-backup'
BONDING_SLAVE0='eth0'
BONDING_SLAVE1='eth1'

and for eth0:
BOOTPROTO='none'
STARTMODE='off'
UNIQUE='GA8e.dR48ZsvS6aD'
_nm_name='bus-pci-0000:02:01.0'

and for eth1:
BOOTPROTO='none'
STARTMODE='off'
UNIQUE='LHB6.dR48ZsvS6aD'
_nm_name='bus-pci-0000:02:02.0'

ifconfig now give:
ulisse:~ # ifconfig
eth0 Link encap:Ethernet HWaddr 00:0D:9D:4D:F4:37
inet addr:159.213.51.131 Bcast:159.213.51.255 Mask:255.255.255.0
inet6 addr: fe80::20d:9dff:fe4d:f437/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:87728714 errors:0 dropped:0 overruns:0 frame:0
TX packets:95781093 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:3402984768 (3245.3 Mb) TX bytes:1846716390 (1761.1 Mb)
Interrupt:29

lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
inet6 addr: ::1/128 Scope:Host
UP LOOPBACK RUNNING MTU:16436 Metric:1
RX packets:22666404 errors:0 dropped:0 overruns:0 frame:0
TX packets:22666404 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:2995445058 (2856.6 Mb) TX bytes:2995445058 (2856.6 Mb)


As an example I have another proliant (dl580) with similar configuration. The out of ifconfig is:
dl580n1:~ # ifconfig
bond0 Link encap:Ethernet HWaddr 00:02:A5:4F:5E:32
inet addr:172.18.10.25 Bcast:172.18.10.255 Mask:255.255.255.0
inet6 addr: fe80::202:a5ff:fe4f:5e32/64 Scope:Link
UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
RX packets:291918978 errors:0 dropped:0 overruns:0 frame:0
TX packets:259961188 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:792101271 (755.4 Mb) TX bytes:3625147679 (3457.2 Mb)

eth0 Link encap:Ethernet HWaddr 00:02:A5:4F:5E:32
inet addr:172.18.10.25 Bcast:172.18.10.255 Mask:255.255.255.0
inet6 addr: fe80::202:a5ff:fe4f:5e32/64 Scope:Link
UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
RX packets:284130508 errors:0 dropped:0 overruns:0 frame:0
TX packets:259961185 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:117464164 (112.0 Mb) TX bytes:3625147457 (3457.2 Mb)
Base address:0x5000 Memory:f7fe0000-f8000000

eth1 Link encap:Ethernet HWaddr 00:02:A5:4F:5E:32
inet addr:172.18.10.25 Bcast:172.18.10.255 Mask:255.255.255.0
inet6 addr: fe80::202:a5ff:fe4f:5e32/64 Scope:Link
UP BROADCAST RUNNING NOARP SLAVE MULTICAST MTU:1500 Metric:1
RX packets:7788470 errors:0 dropped:0 overruns:0 frame:0
TX packets:3 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:674637107 (643.3 Mb) TX bytes:222 (222.0 b)
Base address:0x5040 Memory:f7f60000-f7f80000

lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
inet6 addr: ::1/128 Scope:Host
UP LOOPBACK RUNNING MTU:16436 Metric:1
RX packets:328106 errors:0 dropped:0 overruns:0 frame:0
TX packets:328106 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:69468287 (66.2 Mb) TX bytes:69468287 (66.2 Mb)

[Ivan Ferreira · Accepted solution]

That discontinuity problem is only for the DNS service?

Other services works fine?

If you run the named server in foreground, do you have any output?

Try configuring the options:

options {
listen-on { bond_ip_address; };
interface-interval 0;
};

[Daniele Bernazzi]

The discontinuity is at the network level and impact all services.
I didn't try to run the daemon in foreground, neither with a debug level. Tomorrow I'll made this test.

Please can you specify where to write or use such configuring options?
Thank you

[Florian Heigl (new acc)]

Daniele,

please don't take offense, but... does Your switch know about the bonding configuration??? It appears not!

[Daniele Bernazzi]

I am not the net administrator, but I believe the problem is not there because:
1 - the two nics on the server are attached to two different L3 switches with ospf and other sw detecting redundant paths
2 - bonding is ok without named daemon
3 - another server with the same bonding attached to the same switches is working ok
4 - bonding is working in active-backup, so it keeps one nic active and one nic inactive; when the link on the working nic goes down it make at work the other nic.

does it make sense?

[Florian Heigl (new acc)]

make's sense, won't work as far as I can tell...

talk to the net admin, even with one nic in fallback mode in bonding config the second nic might be visible if the switch doesn't know about the setup that's been done.

I don't have too much practice with this - but I think splitting a host over switches needs either switch-crossing trunk ability (large switches have that) or turning to STP.

[Ivan Ferreira]

What Florian says make sense, but, will the switch know the MAC address if the bonding is in active/passive mode?

I think that the switch learns the MAC when an IP is active on the port.

BTW, the options you should configure in the /etc/named.conf file.

[Daniele Bernazzi]

I am now testing with the suggested options in named.conf. Till now looks good. I'll update some news after more test.
Thank you for now

[Daniele Bernazzi]

I believe the problem is solved, since from wednesday I had no troubles. I guess the solution was setting named daemon to not autodiscover network connections; may be his function to operate on all availables conns has problems with bond (in fact the command ifconfig reports bond0, but also eth0 and eth1 as operative and configurated).
Thank you for sharps suggestions

[Daniele Bernazzi]

as stated in my last reply