I'd like to seed ntp in both server and cloud-image in Vivid. Servers
should maintain the correct time by default. Please make any objections
now.
Right now, ntpdate is seeded in minimal. It makes little sense to have
both ntpdate and ntp installed.
So is there some mechanism I can use to have ntpdate not appear in
server and cloud images? I think we need to move ntpdate out of minimal
for this. Would it be acceptable to move ntpdate to the desktop-common
seed or something?
To try and confirm that this is sensible for all use cases, here's a
brainstromed list. It seems to me that the only downside is a little
extra used space in the LXC case.
* "Traditional" bare metal server installation from server ISO
It makes sense to be running ntp here.
* MAAS-deployed server using d-i
It makes sense to be running ntp here. Additionally, MAAS can choose to
configure the system to point at a MAAS-managed NTP server.
* MAAS-deployed server using curtin
It makes sense to be running ntp here. Additionally, MAAS can choose to
configure the system to point at a MAAS-managed NTP server.
* Hand (or other tool) -deployed VM using a cloud image
It makes sense to be running ntp here[3].
* Hand (or other tool) -deployed VM using d-i or a debootstrapped image
It makes sense to be running ntp here[3].
* Juju-deployed VM using a cloud image
It makes sense to be running ntp here[3]. There may be some interference
with existing charms. When these charms are updated to Vivid (or to X),
then they can start assuming that ntp is already available. A
hypothetical "ntp" subordinate charm could be used in cases where some
other NTP server should be used and the deployer wants to manage it more
directly via a relation; in this case, the subordinate charm could just
rewrite /etc/ntp.conf to change the default accordingtly.
* Hand (or other tool) -deployed LXC using a cloud image
It probably doesn't make sense to run ntp here. We could amend the ntp
init script to skip startup when in a container by default.
* Hand (or other tool) -deployed LXC using a debootstrapped image
It probably doesn't make sense to run ntp here. We could amend the ntp
init script to skip startup when in a container by default.
* Juju-deployed LXC using a cloud image
It probably doesn't make sense to run ntp here. We could amend the ntp
init script to skip startup when in a container by default.
Are there any use cases I've missed here?
Previous discussions:
[1] https://lists.ubuntu.com/archives/ubuntu-server/2013-April/006556.html
[2] https://lists.ubuntu.com/archives/ubuntu-devel/2013-December/037874.html
[3] https://help.ubuntu.com/community/KVM/FAQ#Should_NTP_be_used_for_time_synchronisation.3F
