AIX

Would anyone please kindly help to solve this problem...

An LPAR with the below network configuration. ent0 and ent1 are logical lan (virtual ethernet) from VIO SEA.

en0
1.2.3.4 <- boot ip
192.168.1.1 <- persistent ip
192.168.1.10 <- service ip
en1
11.22.33.44 <- boot ip

When I unplug the en0 cable (i.e. unplug the corresponding VIO ethernet cable), persistent and service will not failover to en1.

I tried the case for dedicated machine, when en0 is unplugged, in error report it will show a failure of ethernet, and the persistent and service IP will failover to en1. This will not happen at VIO LPAR since it is logical LAN and it will not detect a failure.

I found a solution from IBM to add !REQD <source IP> <destination IP> in the netmon.cf file, but not successful. I also added the IPs for the variable ping_client_list in clrginfo.rc file, but still failed.

Please kindly help how to make the LPAR know there is a failure in adapter, and will failover the persistent and service IP to the other interface.

Thank you very much!!

how is the system supposed to know that you want them to failover to the other adapter - as this is a hacmp cluster I would expect the cluster to fail it over to the other node rather than to the other adapter?

I mean - there is no reason for it to do it ? If you want an adapter failover than use etherchannel ...

well hacmp switches ips from one adapter to another (see the cspoc option swap adapter to trigger this manually)

when you just use one etherchannel then hacmp complains about the missing second boot interface. but I don't care, that's the way my clusters are configured

in Power ha 7 boot ips can be in the same subnet as service/persistent ips and you just need one


but afaik netmon.cf is the way you configure the ip failover. first the adapter swaps, then the whole resource group


how did you simulate the interface down? hacmp will not react to ifconfig xxx down, at least in hacmp 5.4/5.5

Perhaps I should use "interface" instead of "adapter".
interface "en0" and "en1" is backed by adapter "ent0" and "ent1".
Since both en0 and en1 are communication ports for PowerHA, when either en0 or en1 failure, the persistent IP and service IP should move to the other interface first.
In VIO environment, PCI hot plug is not available i.e. when you unplug the ethernet cable for ent0, in AIX it is still considered "up". You will only see an adapter failure at VIO, but not AIX.
In this case, IBM proposed using netmon.cf. But I failed following the this guildlines...
[img]www.ibm.com/developerworks/aix/library/au-powerha/?S_TACT=105AGY20&S_CMP=HP[/img]

---------- Post updated at 10:56 PM ---------- Previous update was at 10:52 PM ----------

That is what I want to know. I am using PowerHA 6.1 Sp5, VIOS 2.2
PCI Hot plug is not available for AIX with virtual ethernet. What I want to know is how will it detect an interface down? If it is not possible, how to make sure the IP will failover to another interface, when the interface is malfunction (e.g. unplugged ethernet cable in VIO)?

what's wrong with netmon.cf?

I would use it the following way in your case:

Code:

!REQD en0 1.2.3.254
!REQD en1 11.22.33.254

assuming .254 is your default gateway in all subnets
maybe add some additional ips pingable from your service ip

Dear funksen,

Yes I did exactly the same thing as you have written
I also tried !REQD 1.2.3.4 1.2.3.254 and !REQD !ALL 1.2.3.254
but still failed.

When I unplug the cable, clstat -a shows all UP
hacmp.out and cluster.log did not react
it just seems not aware that the network is malfunction.
I doubt if the netmon.cf function properly.

strange behaviour, probably works as designed, I never tested hacmp with netmon.cf in an environment like yours.

do you need two boot subnets?
what about:

both interfaces in the same vlan, etherchannel, and one ip:




en0 + en1 etherchannel (smitty etherchannel)

ent2 Available EtherChannel / IEEE 802.3ad Link Aggregation
en2: 1.2.3.4


otherwise I would open an ibm call