Client provided hostname "hostname" which does not match resolved name "hostnameB"

[FLQ]

Hi all,

I have this particular problem regarding name resolution in MC/SG for Linux.

Context:

SUSE Linux Enterprise Server 10 (x86_64)
VERSION = 10
PATCHLEVEL = 2

serviceguard-A.11.18.00-0.sles10

I have 2 pkgs running on a 2-node cluster which are working fine.

The problem is that I see

Client provided hostname "hostname" which does not match resolved name "hostnameB" in my syslog.

"hostname" is the front-end interface from another 2-node HP-UX cluster.

"hostnameB" is the back-end interface from that other 2-node HP-UX cluster.

The Linux cluster is on the same networks as the HP-UX one.

I tcpdump'ed the networks (back-end and front-end) on my Linux cluster and saw that the HP-UX cluster is broadcasting cmclconfd packets on each network containing "hostname" in the payload.

123.234.12.25.59660 > 123.234.12.255.5302: [udp sum ok] UDP
10.1.1.25.59659 > 10.1.1.255.5302: [udp sum ok] UDP

It seems the problem is name resolution related but I can't seem to find an acceptable solution for it.

Things I have tried so far:

1 - echo "123.234.12.25 hostname hostname.fqdn" >> /etc/hosts

2 - echo "10.1.1.25 hostnameB hostnameB.fqdn" >> /etc/hosts

The #1 fixes one of the two occurences in my syslog regarding "hostname" in the payload of the broadcast packet but the other remains since the short name always resolves to "hostname" (the front-end interface) and obviously is not matching "hostnameB".

My question is:

Why is the HP-UX cluster broadcasts the node name "hostname" on the 2 networks?

Is it the right behavior or is there a config problem?

TIA

[Vitaly Karasik_1]

what "hostname" command show?
try to change your hostname with this command.

[John Bigg]

Firstly, you should have the hostname as an alias for all interfaces on Serviceguard nodes, i.e. what you should really do is:

cat >> /etc/hosts
123.234.12.25 hostname hostname.fqdn
10.1.1.25 hostnameB hostnameB.fqdn hostname
^D

since 123.234.12.25 and 10.1.1.25 belong to host "hostname" this should be an alias for both these addresses. This should stop all the messages you are seeing in syslog.

This is described in the section entitled, "Configuring Name Resolution" in the HP-UX and Linux Managing Serviceguard Manual.

With regards to the UDP broadcasts, this is normal and no a misconfig. UDP broadcasts are done for various different reasons at different times including checking for duplicate names on a subnet which are not supported, but also so multiple interfaces can respond to a command so that the commands work even when the primary host based lan has failed. This leads to the situations where one cluster picks up messages broadcast by another cluster. This is quite normal behaviour.

However, if you are seeing many of these messages it does indicate that something or someone is running Serviceguard commands when you do not realise it. Maybe you have a cron job issuing commands, or an application like ITO configured to run SG commands regularly. UDP broadcasts on hacl-cfg are only issued when a command is run and are not used during normal cluster running by the cluster daemons themselves.

[Jov]

Hi Francis,

Did you managed to resolve your issue?


Jov

[Wim Rombauts]

This issue is probably already solved some time ago, but I just wanted to confirm that the solution from John Bigg is correct.
I had the same messages here after an upgrade from 11.17 to 11.19 and adding the short hostname as alias to all configured IP addresses solved the issue.