network_down: HACMP

[jprabaker]

I'm running AIX 4.3.3 ML 08 with HACMP 4.4 Classic. I have 2 cascading resource groups running on 2 nodes. If one node has a "network_down" event, all the other machine does is log this in the hacmp.out file.

Has anyone any experience in customising this event so that it forces a takeover?

Any advice will be apriciated.

JP

 

[PSD]

JP,

Well if a network down event is occuring on one node this is a strange phenomenon they are connected to the same n/w after all. The reason failover is not occuring is because via the heartbeat serial link the server is still up. If you wish to force a failover you need to write a post-event script as a post event of network_down to force a failover. i.e.

#! /bin/ksh
clstop -y -N -gr


My argument is that the network should be down on both nodes at this point anyway. If you are getting a lot of network_down events then consider changing the failure detection rate of tcp/ip to slow.

Hope that helps


PSD
IBM Certified Specialist - AIX V4.3 Systems Support
IBM Certified Specialist - AIX V4 HACMP

 

[jprabaker]

PSD,

I was thinking about the unlikely event were both NICs in one of my machines would fail. (Our testing chaps will test this by pulling out both the cables!)

I am in the process of writing 2 scripts to handle this, a network_down_complete_post script and a network_up_pre script.

Firstly the network_down_complete_post script checks if the network_down event is global (if [ $3 = "-1" ];then "do nothing....)

If it is not a global failure then it basically check the local_node against the "nodename" passed from the network_down_complete event ($3).

If they are both the same it will touch a tempory file and test a number of times to see if this file still exists. (The network_up_pre script will remove this file!).

If the file gets removed by the network_up_pre script then the script will exit, otherwise after the timeout period it will run a graceful takeover.

Does this make sense or am I on completely the wrong lines here?

Regards,

JP

 

[vickersdc]

Isn't the point of cluster that it should failover if the active node cannot see the network, but the standby can (ie. network down to active node only) - but that in the event that neither node can see the network no failover takes place?

I'm from a SunCluster background rather than HACMP, but surely the basics of cluster hold true... dont't they?!

 

[jprabaker]

Nope. I'm using HACMP classic 4.4 and it states in the manuals that in the event of a network_down it will do nothing. Its an event that needs to be tailored to the needs of the particalur site which is why I've done a couple of custom events.