|
On Mon, Sep 04, 2000 at 05:06:49PM -0400, Donald Ball wrote:
> hi guys. i've been running an lvs test web farm for a few months now with
> very few problems. recently, though, i had to reboot my load balancer
> (yeah, i've only got one) and now the load balancer won't distribute my
> http requests. i'm using heartbeat+ldirectord from the 0.4.8-1 rpm and i'm
> pretty sure i'm using direct routing. here are the relevant conf files:
A couple of things you may want to check.
First make sure that your ipvsadm and LVS kernel patches are from
the same release of LVS. The version of LVS can be found (as shown below)
using "ipvsadm -L -n", the version of ipvsadm can be found using
ipvsadm -h. You should inspect the tar balls for the relevant versions
as available on www.linuxvirtualserver.org to make sure that things
these two peices of core are in step.
There was an issue with heartbeat not being able to start ldirectord
if it was not shut down cleanly. From memory, this problem was
fixed for the 0.4.8 release but try the following in any case.
1. Shut down heartbeat
/etc/rc.d/init.d/heartbeat stop
2. Make sure both heartbeat and ldirectord are _not_ running
by examining the output of ps, and using kill as neccesary.
3. Start up heartbeat.
Look at the logs. In particular /var/log/ldirectord.log.
Try updating ldirectord to the version in the heartbeat cvs tree.
The code has undergone a number of bug fixes and enhancements over
the past months. You may well have a buggy version.
Finally, as you only have one director you don't really need to run
heartbeat. It shouldn't do any harm, but really it is one extra piece of
the puzzle, one extra thing to debug. Try running ldirectord by itself.
Ldirectord should come with its own init script so you can run it
at boot time.
An example of this can be found at
http://ultramonkey.sourceforge.net/ultramonkey-1.0.1/topologies/topology-lb-3.html
<snip>
> also, just thought i'd mention - frequently, when heartbeat is starting up
> and is setting up ip address aliases, it seems to get into a race
> condition in which more than one ip address tries to grab the same network
> interface alias - eth0:0 for instance - so only some ip addresses get
> fired up properly.
As suggested above I would consider taking heartbeat out of the equation
and configuing these interfaces manually, using
/etc/sysconfig/network-scripts/ifcfg-eth0:0 and friends under Red Hat.
--
Horms
|
