IBM SG24-5131-00 User Manual

IBM Certification Study Guide  AIX HACMP

  • Verify that failover has occurred (

netstat -i

 and 

ping

 for networks, 

lsvg -o

 

and 

vi

 of a test file for volume groups, and 

ps -U <appuid

> for application 

processes).

  • Power cycle NodeF.  If HACMP is not configured to start from /etc/inittab 

(on restart), start HACMP on NodeF (

smit clstart

).  NodeF will take back 

its cascading Resource Groups.

  • Verify that re-integration has occurred (

netstat -i

 and 

ping

 for networks, 

lsvg -o

 and 

vi

 of a test file for volume groups, and 

ps -U <appuid

> for 

application processes).

6.2.2.3  TCP/IP Subsystem Failure
  • Check, by way of the verification commands, that all the Nodes in the 

cluster are up and running.

  • Optional: Prune the error log on NodeF (errclear 0).

  • Monitor the cluster log files on NodeT.

  • On NodeF, stop the TCP/IP subsystem (

sh /etc/tcp.clean

) or crash the 

subsystem by increasing the size of the sb_max and thewall parameters to 
large values (

no -o sb_max=10000; no -o thewall=10000

) and ping NodeT.  

Note that you should record the values for sb_max and thewall prior to 
modifying them, and, as an extra check, you may want to add the original 
values to the end of /etc/rc.net.

  • The TCP/IP subsystem failure on NodeF will cause a network failure of all 

the TCP/IP networks on NodeF. Unless there has been some 
customization done to promote this type of failure to a node failure, only 
the network failure will occur. The presence of a non-TCP/IP network 
(RS232, target mode SCSI or target mode SSA) should prevent the cluster 
from triggering a node down in this situation.

  • Verify that the network_down event has been run by checking the 

/tmp/hacmp.out file on either node. By default, the network_down script 
does nothing, but it can be customized to do whatever is appropriate for 
that situation in your environment.

  • On NodeF, issue the command 

startsrc -g tcpip

. This should restart the 

TCP/IP daemons, and should cause a network_up event to be triggered in 
the cluster for each of your TCP/IP networks.

  • Check, by way of the verification commands, that all the Nodes in the 

cluster are up and running.

  • Optional: Prune the error log on NodeF (

errclear 0

).