|
>> Hmm... oh yes.... currently keepalived doesn t handle netlink ip check
>> before sending advert. The netlink call to set VRRP VIP is done during
>> bootstrap, then during checkers stats.
>
>I see. Nasty ;)
Agreed :)
>> So the answer is.... It is a "bug" well, a lake of time from my side to
add
> ^^^^^^^^^^^^
> :) so you have time
In my night time :)....
>> The VRRP part need to be enhanced... I am currenlty working on this with
>> netlink. The VIP presence check is needed for checkers because currenlty
if
>> you run both healthcheck & failover, keepalived perform healthcheck even
if
>> VIP is not owned by the BACKUP director... So the extension here is :
>> "perform healthcheck if VRRP VIP or IP is owned and correctly set".
>
>What happens in case of failback?
All became in the init wanted state (state set in the configuration file).
So after failover, failback set the VRRP instance in the init state.
Currently a problems steel persist in some case find the IPSEC
seq_number... Need to worker harder on this point...
Probably using a state machine.... need to thought on it :)
>> I have been very busy with my job so I had to delayed a little my work.
But
>> will be back to strong work :)
>>
>> Currenlty the keepalived framework is cleaner than it was. Healthchecker
>> can be developed much more quickly... This work done, I need to spend
more
>> time on netlink & VRRP....
>
>No need to hurry, you've really done an excellent work so far.
Happy to hear this from you ;) but the work is not over... I have some
interresting discuss with Jan Holmberg to enhance keepalived.... I will
probably embend some compiling stuff to define/manipulate LVS/keepalive
objects... My final goal for LVS is a VTYSH framework dealing with failover
protocol (VRRP), healcheck framework (keepalived checkers) & manipulation
LVS rules... And mix interaction between those elements... But the road is
long to this :) And some make profit company will cut down my interrest on
this... Because I don't think if it is really strategic for us to develop
that kind of stuff to be heavily stolen buy those company without any
feedbacks to us... I think this must be those company that can offer code
extension to us...
>This is more Julian's field I guess.
Yes but I really do not want to perturbe his work... I think that I have
enought elements to start work on this. But if I am missing somethings in
my thought... feel free to correct...
>If I was in your situation I'd have
>a look at the iproute2 source code for the part of:
>
>ip route get <IP> ...
>
>And derive stuff from there.
Yes, it is my way of work... Will try to find some validate time for
this... It became more harder for me to find time... Because of economics
needs... OpenSource is very nice for learning effective things but still an
economic problem to fully support OpenSource projects... It became
philosiphic :)
C you later,
Alexandre
|
