From: Gebhardt Thomas (gebhardt_at_hrz.uni-marburg.de)
Date: Mon 08 Mar 2004 - 10:02:57 GMT
we had some problems with nasty side effects when stopping
one specific vserver: all other vservers on the same master lost
their network connectivity.
Herbert helped me to trace down the problem: We are running
the master server and the vservers in different ip subnets:
the vservers have a public ip address; the master has an
"intranet" address which is not routed beyond our intranet.
The kernel ip stack treats the first ip address within a "scope"
as primary and deletes all secondary ip addresses within
this scope when the primary address is taken down.
(ifconfig doesn't show the primary/secondary feature,
but ip from iproute2 does)
To get around this problem I
1. explicitely added an ip route for the public ip
subnet of the vservers:
route add -net <public subnet> netmask <netmask> gw <master ip addr> dev eth0
2. defined all vservers with subnet mask 255.255.255.255, i.e. host scope, so
every vserver is primary within its own scope.
IPROOT="eth0:<vserver ip address>"
IPROOTBCAST=<vserver ip address>
I don't think that these vservers should use some ip broadcast protocol,
so I also set IPROOTBCAST to the vserver ip address.
It seems that this setup works for me.
Vserver mailing list