Netplan and high availability

[cdmiller]

Hello,

Thanks for all the responses, sorry for replying so late.

We utilize puppet so our current work around is to subscribe an exec
of keepalived to the exec for netplan.  A change in the netplan
config triggers cascading reloads, putting back the virtual IP.

We considered going back to ifupdown but we like the netplan yaml
syntax and try to minimize our deviations from baseline installs.

Not surprised systemd-networkd is the culprit.  IMO systemd treats
many server expectations as corner cases.  While sometimes annoying
it might delay the eventual AI take over of all sysadmin functions :)

Thanks,

- cameron

On 10/30/18 3:38 PM, James Hebden wrote:
> On Tue, Oct 30, 2018 at 09:17:58PM +0100, Harald Weidner wrote:
>> Hello,
>>
>> On Wed, Oct 17, 2018 at 11:09:07AM +1100, James Hebden wrote:
>>
>>> Do you get the same behaviour if you use an alias interface in your
>>> keepalived configuration? This configured keepalived to behave in a
>>> similar way to pacemaker/crm, in that it will create an secondary
>>> interface attached to the configured interface, which will be configured
>>> with the VIP.
>>>
>>> For example, if your interface is eth0, and you provide an alias of
>>> eth0:1, eth0:1 will be configured with the virtual IP, and eth0 will
>>> retain whichever IP is configured in your netplan if ifupdown
>>> configuration.
>>
>> I have tried this now, and it didn't help. After "netplan apply", all
>> of keepalived's virtual IP addresses are dropped, regardless of having a
>> label or not.
> 
> Unfortunate - this means that netplan is also likely to clobber VIPs
> configured by other software too, for example, keepalived.
> 
>>
>> Best regards,
>> Harald
>>
>> -- 
>> ubuntu-server mailing list
>> ubuntu-server at lists.ubuntu.com
>> https://lists.ubuntu.com/mailman/listinfo/ubuntu-server
>> More info: https://wiki.ubuntu.com/ServerTeam
>