data lans down & heartbeat up will there be a failover?

[Amir Haboosheh]

cluster lan configuration:

NODE_NAME unix1
NETWORK_INTERFACE lan0
HEARTBEAT_IP 10.0.20.16
NETWORK_INTERFACE lan1
NETWORK_INTERFACE lan2
HEARTBEAT_IP 192.168.0.1

NODE_NAME unix2
NETWORK_INTERFACE lan0
HEARTBEAT_IP 10.0.20.18
NETWORK_INTERFACE lan1
NETWORK_INTERFACE lan2
HEARTBEAT_IP 192.168.0.2


both node lan2 are in a cross-over (please avoid the cross-over dabate).

both node lan1 are standbay to lan0 which is data and heartbeat.

1. if both lan1 and lan0 cards go down will my packages failover to node2 even if the lan2 the heartbeat lan is still up (imagine i unplug the cables on lan1 and 2 on node1)?

2. if lan2 is down will the package failover to node2 or stay on node1 since lan0 is still sending heartbeat info?


thanks,
amir

[Masud Parvez · Accepted solution]

Q 1

Yes, your package will failover. Because you are lossing subnet.

Q 2
No, your package will not failover.

[melvyn burnard]

If your package is configured to monitor the subnet (10.0.x.x) and the subnet fails, Serviceguard will halt the package running on th enode, and move it across to the other node, provided all the switching is enabled for the package

[Amir Haboosheh]

how or where do i check if my package is monitoring what subnet?

thanks guys that's what i wanted.

[Masud Parvez]

It is under package control file and you can see your subnet by cmviewcl -v

example

Script_Parameters:
ITEM STATUS MAX_RESTARTS RESTARTS NAME

Subnet up 10.0.x.x

[Deepak Kr]

Yes bernard is right. Cluster takes failover decision based on the settings we make in configuration.

Also:
You can use following command to list all configurable/configured parameters in Cluster and package configuration files


grep ^[A-Z] /etc/cmluster/clustername.config

[melvyn burnard]

In your package configuration file, there is a variable SUBNET
If you have put in a subnet value, the packag emanager will monitor this subnet. If it is left blank, then the package will not monitor the subnet, even though in your control file you may put a SUBNET/IP pair for adding an IP to a SUBNET

[Amir Haboosheh]

melvyn,
that's what i did. i added the subnet value 0f 10.0.0.0 for SUBNET.

another question. if lan0 is disconnected will lan1 on unix1 take over based on the configuration above with no disrupption? or will there be a slight disruption but will still take over?

in a related question, in windows hp has a adapter teaming option where there is a loadbalancing between 2 nics and zero disruption if one lan goes down, is there something similar in hpux or is there only a "standby" nic within the service guard application where the "standby" nic simply takeover if the main nic fails?

thanks,
amir

[melvyn burnard]

>if lan0 is disconnected will lan1 on unix1 take over based on the configuration above with no disrupption? or will there be a slight disruption but will still take over?

There will be a small delay, but it will be transparent to users/applications provided that the physical configuration is correct, and the standby lan is in working condition.

in a related question, in windows hp has a adapter teaming option where there is a loadbalancing between 2 nics and zero disruption if one lan goes down, is there something similar in hpux or is there only a "standby" nic within the service guard application where the "standby" nic simply takeover if the main nic fails?

Yes there is something similar, it is called Auto Port Aggregation (APA) and is additional software that is used to aggregate or "team" two or more physical NIC's to provide loadbalancing and redundancy.
It is a separaetly purchaseable product unless you have the new HP-UX 11.31 OE's.

[Amir Haboosheh]

new scenerio i tried on friday:

package is running on node2.
i unpligged the network cables from node2 in the following order:
lan2 is disconnected (crossover heartbeat)
lan0 is disconnected (heartbeat/data)
lan1 is disconnected (lan0 standby nic)
cluster starts reformation.
node1 crashes and reboots.

why?

i thought node1 should take over the package and if there is any crash/reboots it should be on node2 since it's network cards are the ones that were disconnected (failed).

any way to fix this problem?

just it case it matters i am using a lock disk for arbitration.

thanks,
amir