Service Guard Startup

[Michael Elleby III]

I am relatively new to ServiceGuard configuration and Package Startup, and have a question about the primary and standby lan interfaces.

Can Package startup (on a server boot/reboot) be configured to use the standby lan interface if the primary one is unavailable?

Example: We have 2 N Class servers in a SG cluster. On one of the servers that was rebooted during our normal maintenance window, the primary lan interface cable was unplugged, but the secondary was available, but when the servers came up, the package would not start until the primary cable had been plugged in. I guess SG said that since he did not see the primary, he would not start the package.

Now prior to this, the primary had been unplugged and the server failed over to the secondary interface, because we were experiencing packet loss on the primary interface.

Can anyone help?

[melvyn burnard]

If I remember correctly, the behaviour you have seen is to be expected, as for the node to start cluster activities correctly, all the networks that the cluster binary knows about should be ok at the time the node tries to join the cluster.
So in this case, the promary lan is down, the cmcld sees this and automatically assumes there is a problem that could cause issues with the package(s) and so will not start.
Once the node is running normal cluster activities and all was om, the failure caused by pulling the cable on the primary lan will be seen as a "true" failure, and the normal behaviour kicks in, allowing the network switch.
HTH
them.

[Michael Elleby III]

So, you are saying that there is no way a package can be configured to "failover the lan interface" when the server boots up?

I know what you stated, but it just seems to me that it would possible to do this.

Mike Elleby

[David Navarro]

Hi,
I'm agree with Melvyn, at startup, if something goes wrong, cluster doesn't starts. If you want to startup with any lan down, try cmruncl -f -n node_name, I don't know if this will startup a node with a lan down, but, you can test it.

[melvyn burnard]

SG makes the assumption that something is wrong at startup, so tries to prevent issues occuring by starting up a package with an already existing problem.

[Michael Elleby III]

Melvin-

I wasn't trying to disagree with you, I was just curious. I do understand what you stated, however.

With that said, let me ask this quesion, is there any type of 'teaming' driver that I could use to facilitate the startup of a package with a lan down..

Mike

[melvyn burnard]

I do not believe there is something to allow this. I am unsure what the result would be of having the package running on th node that is 100%, starting the node with the missing cable, and then attempt to move the package across.

[Bryan Payne]

What would be the effect of using one of those load balancing multi NIC's as your primary?

[Magdi KAMAL]

Hi Michael,

MC SG is desgnied to use the standby lan as soon as the primary interface do not response.

But it still can use only one if the other is not present ( If one of the two lan is not working and a failure occurs, trouble starts !!!).

I just got, now, a problem with a FDDI cable and the second lan was DOWN. I replaced the bad cable and it works fine.

I would recommand to check the /etc/rc.config.d/netconf and to keep only one instance according to one lan ( not two instances ). MC SG will now how to configure that only instance to each individual interface.


Magdi