Switch disconnection + SG.

[Prashant Zanwar_4]

HI, hI have a outage on switch where my host in cluster connects, and it doesnt have redundant lan. Outage will be 15MIns.
I am wondering what happens to
- packages running on node (All Oracle DB packages, just VG's activated thru it)..?
- What happens to node itself, as it wont see itself on network, will it panic ? need to know bit urgent ? Dont want oracle DB's to crash..

Thanks
Prashant

[Duncan Edmonstone]

Prashant,

Difficult to say without knowing the cluster configuration. Can you post the cluster configuration and package configuration details, and indicate which LAN interfaces are attached to the switch in question.

HTH

Duncan

[Steven E. Protter]

Shalom,

If there is a second heartbeat configured the cluster will run through the outage, even if the system is not accessible.

It is prudent with serviceguard to have more than one heartbeat network, preferably a private one to avoid unnecessary TOC reboots.

SEP

[David Child_1]

Are your heartbeats using that switch or do you have a cross-over cable as your primary?

Is this a 2-node cluster?

This is MC/SG?

If you have all your heartbeats connected to this switch then yes, your cluster will have a problem.

[Prashant Zanwar_4]

2 node cluster, HB as crossover, public nic is going down !

[David Child_1]

I think we would need some detailed configuration information to be sure, but at a high level if your heartbeat is over a cross-over and all you are loosing is your public network then I don't think the node will go down. Hopefully you know whats set up in your packages, etc. as you could have rules in there to cause problems if the public interface(s) went down.

Is there a standby lan? If so, is that on a different switch?

Is if an Oracle RAC configuration? If so, is oracle using the public network for interconnect traffic? If that is the case then there would still be problems.

[Prashant Zanwar_4]

And public nic, doesnt have redundancy ...

Will it be issue, and should i take DBs down, or ??
Thanks

[Prashant Zanwar_4]

Sorry unable to upload file, cmviewcl -v is big, and weired issue on my laptop.

[Duncan Edmonstone]

I have to question the value of implementing a HA cluster, and then leaving the network as a single point of failure... but you are where you are, so what you need to determine is whether any of your packages have subnet monitoring for the public LAN connection - if they do, then the package is going to halt as soon as you bring down the switch.

You can check this using:

# cmviewcl -vl package

Look through the output and you should see if your subnet is being monitored. Here's an example from my workstation, you should be able to see that the 192.168 subnet is being monitored. If I disconnected the workstation from the 192.168 LAN then the dbsrc package would halt.

PACKAGE STATUS STATE AUTO_RUN NODE
dbsrc up running enabled c3000

Policy_Parameters:
POLICY_NAME CONFIGURED_VALUE
Failover configured_node
Failback manual

Script_Parameters:
ITEM STATUS MAX_RESTARTS RESTARTS NAME
Subnet up 192.168.0.0


HTH

Duncan

[Stephen Doud]

Serviceguard requires at least one heartbeat network to status the active state of each node. More HB networks can be configured. If all are severed at the same time, a cluster reformation will occur and at least one node in the cluster will reboot itself. If you don't want that interruption, insure that at least one HB network is active at all times.

As for packages, Serviceguard will only respond to a network failure if the package configuration contains a SUBNET parameter.

If the following command results in at least one entry, you will know that the package that lists the entry will be affected by severing that network:

# cmviewconf | grep "package subnet"

If the NIC that supports that subnet has a standby NIC assigned to it, the standby NIC SHOULD be wired to a different switch. Otherwise, it is not truly a highly available environment. If the data NIC doesn't have a standby NIC, it is not truly a highly available environment.