09-09-2011 03:15 AM
Hopefully a quick solution. Is there a way to remove the currently active IP aliases from the TCIP configuration during startup? Preferably prior to starting TCPIP itself.
We have two machines which share aliases, if one is restarted, the application and network alias migrate to the other machine. One the failed machine reboots there is a point where both network aliases are active which may cause some problems.
OpenVMS 8-4 (fully patches)
TCPIP 5-7 ECO2
2 of RX2660s (clustered)
09-09-2011 03:47 AM
how about creating/using SYS$STARTUP:TCPIP$SYSTARTUP.COM and adding
$ ifconfig xx0 -alias ip-alias-address-or-name
This is the only supported way to hook into TCPIP$STARTUP
If you're using TCPIP aliases on clustered nodes, these aliases will only move to the newly booting node, IF they are defined as HOME aliases on that node AND there are no current connections to these aliases on the 2nd node. Otherwise the aliases will stay where they are.
09-09-2011 08:20 AM
If you can, don't use aliases at all and use DNS load balancing.
Hard configuration of a virtual ip address can give you two hosts
with the same (virtual) ip address in some situations, the
consequences are funny on a test environment, but disastrous
on a production environment.
MQ can be configured to have failover functionality like this, and after some adventures,
I have made the configuration conditional on the non-reachability
of the ip address, and if not met a cluster-wide kill of any virtual interface with that address
still up. It gives a bit of protection anyway, but it is a kludge.
To get an idea:
$ pingcount = 0 $PingLoop: $! $ ping -c 1 'p3' $! $ if $status $ then $! $ ifcommand = ifconfig - "$" $ pipe write sys$output "do mc ''ifcommand' ''p4' -alias ''p3'" | mc sysman set env/cluster $! $ pingcount = pingcount + 1 $ if pingcount .lt. 10 $ then $ wait 0:0:10 $ goto PingLoop $ endif $! $ request/to=oper9 "Alias ip address ''p3' still in use somewhere, MQ NOT started on ''f$getsyi("nodename")'" $ exit 2 $! $ endif