Re: Ipv6 and preferred_lft

On 14/10/2013 17:26, Mark Schouten wrote:
> On 10/14/2013 05:11 PM, Ryan O'Hara wrote:> First, the setting is not
> optional. It should be. This problem, which
>> I am unable to recreate, only exists with you are using keepalived
>> with both 'vrrp' and 'check' enabled. I understand this a common case,
>> but not everybody uses keepalived's IPVS capabilities. I'm not sure
>> that changing this behavior will not cause problems for other valid
>> configurations.
>
> Well, it is, imho.
>
> If you're not using the VRRP-addresses for IPVS stuff, you'd probably
> want to specify the netmask. Once you specified a netmask (other than
> /128, obviously), the preferred_lft is left alone.
>
> It's quite easy to reproduce the issues of not setting preferred_lft to
> deprecated. Just ' ip addr add $ipv6address dev $interface' and see how
> that address is used for all outgoing traffic. In both cases, whether
> you use keepalived for vrrp only, or with IPVS, you (or at least I) do
> not want the machine to use virtual addresses for it's own
> communication. I want the machine to use it's own address for
> communicating with the outside world.
>
> AFAICS, this patch doesn't break anything if you're not using IPVS
> stuff. But it does FIX a lot if you do.


Agreed.


>> Also, the DaD optimization is better addressed using the sysctl
>> settings:
>>
>> net.ipv6.conf.*.optimistic_dad
>>
>> Which can be set for all interfaces or per interface.
>
> Why? If VRRP decides that the other end is gone, it probably is. Which
> means that you are safe to assume that the address should be up and
> running directly, instead of having to wait for DAD to give the ok.
> Also, you may want to add some addresses of your own to the interface,
> for which you do want to enable DAD. So disabling DAD just for these
> addresses seems more finetuned than for the whole interface.


Agreed.


I will merge this patch.


Regs,
Alexandre

------------------------------------------------------------------------------
October Webinars: Code for Performance
Free Intel webinars can help you accelerate application performance.
Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from 
the latest Intel processors and coprocessors. See abstracts and register >
http://pubads.g.doubleclick.net/gampad/clk?id=60135031&iu=/4140/ostg.clktrk