Define What Fails to Where???

[Lee Harris_5]

Hi,

I am having some difficulty understanding how in cmclconfig.ascii you can decide what standby LAN interface will be used when another fails.

There does not seem to be any logic to the layout of that section of the config file...

For example...

# Definition of nodes in the cluster.
# Repeat node definitions as necessary for additional nodes.

NODE_NAME sgnode01
NETWORK_INTERFACE lan0
HEARTBEAT_IP 192.168.241.10
NETWORK_INTERFACE lan1
HEARTBEAT_IP 192.168.240.10
NETWORK_INTERFACE lan10
HEARTBEAT_IP 192,168.254.2
NETWORK_INTERFACE lan5
NETWORK_INTERFACE lan9

In this configuration, LAN0 is for heartbeat, with no failover...LAN1 is the public interface which will failover to LAN5. LAN10 is a Gigibit Private network which will failover to LAN9.

This does work, we have tested this config...it is from a real running cluster, but I cannot see any logic in this...how does LAN1 know it is meant to fail to LAN5 and so on. Could someone please explain this to me?

Many Thanks - Lee

[G. Vrijhoeven · Accepted solution]

Hi Lee,

The cluster software uses linkloop (linkloop -i ) to know that an other interface is on the same vlan. You can also use this command to verify this.
You can check the syslog to see what interface is configured as failover interface. ( grep cmcld /var/adm/syslog/syslog.log)

HTH,

Gideon

[Duncan Edmonstone]

Lee,

Isn't serviceguard clever? It figures it all out on its own!

Actually what its doing is testing the various networks and seeing which are actually bridged (physically connected and able to communicate at the data link level). I'm sure theres a way to get SG to tell you which networks it thinks are bridged (either a cmquerycl or cmscancl option IIRC - I'm not actually at a system so can't check). You can check this manualy by using lanscan to get the MAC addresses of all your NICs and then linkloop to send link level pings from one interface to another.

Generally the logic serviceguard uses works OK as long as your network is truly configured according to the rules in the SG manual - but you do need to be careful - if you have networks bridged where they shouldn't be (e.g. a heartbeat network bridged with a data network) you could get yourself in a situation where a 'less important' (such as one of many HB networks) IP address fails over to your standby card, and effectively prevents a more important card from failing over by dint of it being on a different IP subnet.

HTH

Duncan

[Slawomir Gora]

Hi,

rules for standby interace in MC/SG are:
1. LAN must be defined in cmclconfig as NETWORK_INTERFACE but must be unconfigured
(no ip address in /etc/rc.config.d/netconfig)

(in your example: lan5 and lan9)

2. LAN interace must be in the same phisical network, it must be able to communicate to the primary LAN card.

If your lans lan9 and lan10 are connected to the same network then lan9 is standby for lan10.

The same rules for lan5 and lan0.

You can use linkloop utility to check which lans can communicate.

[Lee Harris_5]

Thanks for the replies, let me see if I get this straight...

The failover is only for NICs on the same network (i.e. those that have layer 2 connectivity).

Therefore, if lan0 and lan1 are on the same VLAN and can all linkloop to one another, then these automatically become failover NICs for one another...hence if lan0 is configured, and fail...lan1 will pickup the pieces?

So this raises another question...

If for example, I have 3 public lan interfaces, say lan1, lan2 and lan3 which are all on the same VLAN and can all linkloop to one another, then how would I define the following...

lan1 = primary public interface.
lan2 = first failover card.
lan3 = second failover card.

Or would SG just use whichever of the other two (lan2 lan3) if lan1 failed.

Many Thanks - Lee

[Slawomir Gora]

Hi,

I think that order in cluster config file is important.

if you will have:
NETWORK_INTERFACE lan2
NETWORK_INTERFACE lan3
standby lan2 will be used first

You can also run cmviewcl -v to check order for standby network interfaces.