Re: NM changes DHCP address too easily

[Dan Williams <dcbw redhat com>]

On Fri, 2016-04-22 at 16:40 +0300, matti kaasinen wrote:
> 2016-04-22 16:12 GMT+03:00 Beniamino Galvani <bgalvani redhat com>:
>
> > Do you mean that the board is getting a different address at every
> > boot? In order to reuse the same lease, a persistent connection
> > must
> > be used (i.e. a connection written on disk and not the in-memory
> > one
> > automatically generated by NM at every boot).
>
> Usually it gets that old address at boot but changes it soon after
> that. My
> board is using NTP time, so I suppose this happens after NTP time
> synchronization if the original time was corrupted.

On boot there's likely a valid lease from the previous run which can be
renewed.  When the time jumps forward, dhclient notices this and must
expire the lease because it is no longer valid.  dhclient has no choice
but to enter the DISCOVER state and acquire a completely new one.

One workaround:
http://billauer.co.il/blog/2012/10/dhcp-ip-ntpdate-rtc/

It does seem a bit odd that dhclient cannot re-acquire the lease; I'm
not sure if it's allowed by the standards to renew an expired lease or
not.  But you could argue a chicken/egg problem where DHCP delivers
your NTP servers which then cause the time to jump forward which then
kills the lease.  So it seems like something to check into with
dhclient and whether they have improved things here in recent releases.

Dan