Hi,

I am trying to use heartbeat to manage one virtual IP address accros two hosts 
which are interconnected only via the normal network. Configuration files are 
included in the end of this post.

My problem is that gratious ARP is not always sent when the IP address is moved 
between the hosts, resulting in the other equipment being unable to connect to 
the virtual IP, even though it is available.

I do the following:
* Initially the IP address is located on the slave host (deb-hp-mla-hlrtest). I 
assume the same would happen if it started on the master, but have not tried 
that.
* I disconnect ethernet to the slave host
* The IP address is moved to the master host, which broadcasts some gratious 
ARP. Everything work fine after 10 seconds.
* I reconnect the slave host.
* The master host detect this:
  Nov  7 10:12:08 localhost heartbeat[20683]: WARN: Cluster node 
deb-hp-mla-hlrtest returning 
  after partition.
  Nov  7 10:12:08 localhost heartbeat[20683]: WARN: Deadtime value may be too 
small.
and then it heartbeat restarts

* The slave hosts never released the virtual IP, so it is running as before, 
and since heartbeat restarted on the master host, it is the only host with the 
IP address.
  Nov  7 10:12:07 localhost ipfail[4431]: info: Link Status update: Link nw/nw 
now
  has status up

* And everything looks ok, BUT no gratious ARP are sent! So my other equipment 
still thinks the IP address is at the master host, and cannot connect until 
they time out their ARP table.

Am I doing something wrong? I realize the problem could probably be fixed by 
using a serial interconnect, but the 2 hosts may not be located the same place. 
The easiest for me would be if a host always send gratious arp after the link 
has been down.

I am using version 1.2.3 on Debian 3.1 (Sarge)

Thanks for your help
Morten


**ha.cf**
logfacility daemon         # Log to syslog as facility "daemon"
node mla-test2 deb-hp-mla-hlrtest # List our cluster members
keepalive 2                # Send one heartbeat each second
warntime 3
deadtime 10                # Declare nodes dead after 10 seconds
bcast eth0
# Ping our router and other stuff on the local network to monitor ethernet 
# connectivity
ping_group nw 10.10.10.254 10.10.10.10 10.10.10.11         
auto_failback off          # Don't fail back automatically
respawn hacluster /usr/lib/heartbeat/ipfail  # Failover on network failure

**haresources**
mla-test2 10.10.130.37/16



Venlig Hilsen / Regards 
Morten

-- 
Morten Laursen, M.Sc.S.E.
RTX Telecom A/S - http://www.rtx.dk/
Direct phone: (+45) 96 32 24 03

_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to