LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

RE: dead peer detection, ipvsadm

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: RE: dead peer detection, ipvsadm
From: "Gary W. Smith" <gary@xxxxxxxxxxxxxxx>
Date: Sat, 2 Sep 2006 20:51:07 -0700
I'm looking for dead peer detection for the real servers.  I already
have ipvsadm setup using linux-ha on a cluster.  This part seems to work
fine.  When I take one of the SMTP nodes offline (being a large incoming
SMTP configuration) I would like it to be detected and no traffic to be
routed there.  I have read various howto's on a variety of techniques
(including the use of feedbackd).  

If there is a specific howto, or specific section in the howto, can you
please point me in that direction?

This really isn't a blind question from me per say.  I've been talking
to some of my colleagues that also use ipvsadm and they just didn't have
a good way of doing this out of the box.  But they have also admitted
that they are by no means experts on ipvsadm.

Here is a little more about the configuration:

Schedulers:

sched1:
eth0: 10.0.14.101
eth1: 10.254.254.2 (hb)

Sched2:
eth0: 10.0.14.102
eth1: 10.254.254.3 (hb)

Cluster VIPs:
22.33.44.51/24

SMTP Servers:

node1:
eth0: 10.0.14.12
lo:   22.33.44.51/32

node2:
eth0: 10.0.14.12
lo:   22.33.44.51/32

node3:
eth0: 10.0.14.13
lo:   22.33.44.51/32

node4:
eth0: 10.0.14.14
lo:   22.33.44.51/32

node5:
eth0: 10.0.14.15
lo:   22.33.44.51/32

node6:
eth0: 10.0.14.16
lo:   22.33.44.51/32

node7:
eth0: 10.0.14.17
lo:   22.33.44.51/32

node8:
eth0: 10.0.14.18
lo:   22.33.44.51/32

ipvsadm script

-A -t 22.33.44.51:25 -s wlc
-a -t 22.33.44.51:25 -r 10.0.14.11:25 -g -w 100
-a -t 22.33.44.51:25 -r 10.0.14.12:25 -g -w 100
-a -t 22.33.44.51:25 -r 10.0.14.13:25 -g -w 100
-a -t 22.33.44.51:25 -r 10.0.14.14:25 -g -w 100
-a -t 22.33.44.51:25 -r 10.0.14.15:25 -g -w 100
-a -t 22.33.44.51:25 -r 10.0.14.16:25 -g -w 100
-a -t 22.33.44.51:25 -r 10.0.14.17:25 -g -w 100
-a -t 22.33.44.51:25 -r 10.0.14.18:25 -g -w 100

This works great for balancing the connections.  But it seems to die
miserably when I take one server down (as all of the connections seem to
go there because is has no active connections).



> -----Original Message-----
> 
> if by dead peer you mean failed realserver, there's many
> chapters on it in the HOWTO. Have you looked there?
> 
> Joe

<Prev in Thread] Current Thread [Next in Thread>