Netplan and high availability

[Mathieu Trudel-Lapierre]

On 2018-10-16 05:46 PM, cdmiller at adams.edu (cdmiller) wrote:
> Hello,
>
> We have some systems still using keepalived or corosync/pacemaker
> for high availability with IP fail over.  Easiest case would be a
> haproxy or nginx fail over pair.
>
> Currently netplan removes interfaces it does not manage on any
> change (netplan apply).
I'll guess you really mean addresses rather than interfaces.

systemd-networkd largely expects to be authoritative as to the network 
configuration, so as pointed out by others, it will do some things 
people might not expect.

I looked, but I can't find any way to tell it not to do that. I think 
the best would be that we file a bug on github for systemd; discuss the 
situation with upstream, so that we can arrive to something that will 
work for everyone.

Now, I do have one untested idea though: using "critical: true" in 
netplan config for an interface, you can tell systemd that the 
connection is critical (CriticalConnection=true in systemd.network). 
It's worth a try, as this is one feature that will avoid 
releasing/renewing DHCP leases on the interface.

If it doesn't work, I'm still of the opinion that it probably should be 
made to work in the way you describe, where systemd-networkd won't touch 
"foreign addresses".

That's my $0.02CAD on the matter ;)

> Please point me to some recommendations for implementing a high
> availability cluster or pair in Bionic with IP fail over.

I don't know that there is a document on what's recommended, but I agree 
it would be a good idea.

/ Matt