how does HACMP detect a change in IP? 2

[Ukyo]

Hi all,

I read somewhere that hacmp shuts down a node to prevent data corruption when it detects a change in IP address in a cluster. I've been having problems where my nodes shutdown by themselves and errpt shows that clexit.rc has been called. The reason why I ask how HACMP detects change is IP is to see if there is a work around.
I have two clusters. Cluster "A" and "B", each cluster has two nodes, "1" and "2".
I've noticed the following behaviour:
With cluster A is stable, with node 1 and 2 up.
I start cluster B node 2, I noticed that Cluster A node 1 drops dead, powers off.
With cluster B stable, with node 1 and 2 up.
I start cluster A node 1, I noticed that Cluster B node 2 drops dead, powers off.
I've checked all MAC address, they are all unique. All boot and services IP's are unique. However, the boot/service IP on both clusters are on the same network (subnet).
Cluster A can run fine without cluster B, and vice versa.

Very weird, my theory is that somehow the stable node detects a change in IP on the other cluster so it decides to drop dead. Does it detect because they are on the same network even though they have different IP?

Would the version I'm running cause this behaviour? I'm running:
cluster.base.server.rte 4.3.1.16
The reason I suspect this is because I have a third cluster with boot/service IP on the same network, but running on cluster.base.server.rte 4.3.1.15. That cluster seems to be fine.

Is there anything I can do besides pulling out a sniffer or turning HACMP debug on to further investigate?

thanks in advance!

Ukyo

 

[DSMARWAY]

Hi,


The 2 clusters can be on the same network as long as not the same ip address . Have you configured 2 resource groups
1 for cluster A containing 2 nodes and 1 for Cluster B containing 2 nodes ?

What is your failover method i.e. cascading,rotating ,mutual

What other messages do you get in hacmp.out , & cluster.log
when you ran a synchronise of resource group ( emulate)
did you get any errors ?

 

[Ukyo]

Hi all,

Thanks for the pointers..

DSMARWAY,

Yes, the two clusters do all have unique boot, service and standby IPs.
I don't seem to get any msgs in hacmp.out , & cluster.log
when I ran a synchronise of resource group ( emulate)
and it showed no errors.

Yes, two resource groups configured. The failover is also set to cascading.

thanks in advance..

Ukyo

 

[axelsche]

Hi Ukyo,
do you have unique cluster id's in your Network?
Just an idea!

HTH
Axel

 

[DSMARWAY]

HI,

You can set an alernative dummy MAC address in HACMP , this is used say if your server fails over , the end users would have to update their arp cache on their PCs so that they can connect to the MAC address of the card on the other server , ( depending on the nwtwork, this could take some time) therefore you have the option to add a dummy MAC address in HACMP note it must be a dummy MAC address ,
so when the server fails over , the connection is immediate because they don't have to update the users arp cache.
Have you got this is place ?

 

[Ukyo]

axelsche,

YOU LEGEND! everything, MACs, node names, cluster names, IPs and all were unique.. except the cluster ID number. We used scripts and a config file to create the cluster and it seems we overlooked that parameter that was left to default as "0" in our config file on both our conflicting clusters.
Weird thing is, we still have another two clusters(a total of three clusters, two were probablematic) on the same network (physically seperate rooms) with the same cluster ID (0) but they seem to be all cheery. I guess its because they are on slightly different versions of HACMP.

DSMARWAY,

thanks for the tip.. is this dummy MAC address the one thats setup in alternative MAC address? My understanding is that when it does the take over, the MAC address gets taken over as well by default?

thank you all! ^_^

Ukyo