|
Hi Julian,
I added 'ipchains -M -S 0 0 10' to my script. It's better now. If the main
node is down, the traffic is switched to the backup in +/- 2 mins (which is
acceptable).
Coudld you explain the "0 0 10" parameters?
Now, I've another problems: If the main node comes back to life, the traffic
is not balanced back... This is not a big problem as the service is "up" but...
:)
Regards,
X
--
Xavier Mertens &Wanadoo Belgium "Contrary to popular belief, Unix is
NOC Manager a subsidiary of userfriendly. It just happens to be
XM1-6BONE XM3-RIPE France Telecom selective about who it makes friends with"
On Fri, 27 Apr 2001, Julian Anastasov wrote:
>
> Hello,
>
> On Fri, 27 Apr 2001, Xavier Mertens wrote:
>
> > Hi,
> >
> > I've the following config:
> >
> > RealServer (Linux box 2.2.16-22) with two nodes configured to reveive
> > RADIUS traffic. Weight are set to force a kind of fail over:
> >
> > UDP radius-vs:rad-auth rr
> > -> hades:rad-auth Route 1 0 1
> > -> virgo:rad-auth Route 65535 0 0
> > UDP radius-vs:rad-acct rr
> > -> hades:rad-acct Route 1 0 1
> > -> virgo:rad-acct Route 65535 0 0
> >
> > I wrote a small daemon which monitors the 2 RADIUS servers and issues
> > "ipvsadm -(a|d)" in case of troubles with one of the RADIUS.
> >
> > Problem: once the daemon removed the node "hades" (weight 1), the traffic
> > is NOT forwarded to virgo (weight 65535)! Any idea? If a few minutes later,
> > hades is back, the traffic is still send to this host.
>
> Any effect from ipchains -M -S 0 0 10 executed in the LVS box?
> May be you can reduce the UDP timeout even more in your setup.
> May be there are still UDP entries that are not timed out after you
> delete the real server.
>
> > Regards,
> > Xavier
> >
> > --
> > Xavier Mertens &Wanadoo Belgium "Contrary to popular belief, Unix is
> > NOC Manager a subsidiary of userfriendly. It just happens to be
> > XM1-6BONE XM3-RIPE France Telecom selective about who it makes friends
> > with"
>
>
> Regards
>
> --
> Julian Anastasov <ja@xxxxxx>
>
|
