On Mon, Jul 07, 2014 at 08:44:43PM +0200, Mxher wrote: > Hello again, > > I'm doing few more tests and now I'm wondering if this is possible to > disallow CARP to have some resources on serverA and others on serverB?
Have you set the sysctl net.inet.carp.preempt=1? > > Here is my tests (advbase=1 and advskew=0 for every interfaces on both > servers): advskew should be different on master from backkup. Try advskew=200 on obsd2. Please read man carp. The first example is exactly what you need. > * Initial state > root@obsd1:~# ifconfig HA |grep status > status: master > status: master > status: master > status: master > root@obsd2:~# ifconfig HA |grep status > status: backup > status: backup > status: backup > status: backup > > * I unplugged em2 and em3 on obsd2 and em1 on obsd1: > root@obsd1:~# ifconfig HA |grep status > status: master > status: invalid > status: master > status: master > root@obsd2:~# ifconfig HA |grep status > status: backup > status: master > status: invalid > status: invalid > > > obsd2 became master for em1 while obsd1 is master for everything else. > Is there any (proper and automatic) way to avoid that ? > > I know that kind of situation will not happens often but... > > > Thanks again! > > > Le 06/07/2014 13:13, Mxher a écrit : > > Le 06/07/2014 12:05, Otto Moerbeek a écrit : > >> On Sun, Jul 06, 2014 at 10:59:16AM +0200, Janne Johansson wrote: > >> > >>> The sysctl for carp.preempt controls if they should all fail at the same > >>> time. > >> > >> read carp(4). It contains answers to some questions asked. > >> > >> -Otto > >> > > > >>> Den 6 jul 2014 10:12 skrev "Adam Thompson" <athom...@athompso.net>: > >>>> I recall someone pointing out that interface groups of carp interfaces > >>>> will all transition simultaneously. I find ifconfig(8) inconclusive; run > >>>> your own tests and if that works, you have a built-in solution for > >>>> keeping > >>>> all the carp interfaces in sync. > >>>> Then, use ifstated to manage the pppoe interfaces depending on ifstate of > >>>> the relevant wan interface? You could set up a carp interface with no IP > >>>> address bound, set it into the common if group and it would go up/down > >>>> with > >>>> the other carp ifs. > >>>> Maybe. I haven't tried anything like that myself. > >>>> -Adam > > > > I run some tests and this is working as expected! > > > > Only thing I see is that there will be no group failback if this is a > > virtual carp interface which goes down. > > > > To be clear if the parent interface of carp2 goes down the whole group > > will switch but not if carp2 goes down by itself (by an admin mistake > > for example): > > * initial states > > root@obsd1:~# sysctl -a|grep preem > > net.inet.carp.preempt=1 > > root@obsd1:~# ifconfig HA |grep status > > status: master > > status: master > > status: master > > status: master > > > > root@obsd2:~# sysctl -a|grep preem > > net.inet.carp.preempt=1 > > root@obsd2:~# ifconfig HA |grep status > > status: backup > > status: backup > > status: backup > > status: backup > > > > > > * states with carp2 down on obsd1 > > root@obsd1:~# ifconfig carp2 down > > root@obsd1:~# ifconfig HA |grep status > > status: master > > status: master > > status: invalid > > status: master > > > > root@obsd2:~# ifconfig HA |grep status > > status: backup > > status: backup > > status: master > > status: backup > > > > > > * also unfortunately when carp2 goes UP again on obsd1 it still remains > > on obsd2: > > root@obsd1:~# ifconfig carp2 up > > root@obsd1:~# ifconfig HA |grep status > > status: master > > status: master > > status: backup > > status: master > > > > root@obsd2:~# ifconfig HA |grep status > > status: backup > > status: backup > > status: master > > status: backup > > > > > > Anyway I think this is an acceptable risk. > > > > > > @Adam: I will now try to use ifstated to manage pppoe interfaces like > > you suggest. > > > > > > Thanks to everyone of you.
