ceph-fs charm shouldn't install ntp package unconditionally

[Yoshi Kadokawa]

According to this[0], ceph-fs charm will install ntp package,
and with the following condition, nrpe check script will fail to run.

- when ceph-fs is deployed in an LXD container
- when the host machine, where ceph-fs is running in an LXD container, has a relation with ntp and nrpe charm

Here is the related bug[1].
First of all, I believe this ntp package is not necessary for ceph-fs to deploy.

[0] https://opendev.org/openstack/charm-ceph-fs/src/branch/master/src/lib/charm/openstack/ceph_fs.py#L172

[1] https://bugs.launchpad.net/charm-nrpe/+bug/1921392

[Nobuto Murata (nobuto)]

I think there are two issues here:
1. charm always installs "ntp" for both bionic and focal, "ntp" is in universe with focal it should be "chrony" on focal instead
2. charm installs (and activate) ntp servers even inside LXD container.

Like ceph-mon charm, we shouldn't install ntp in LXD, but relies on baremetal.
https://bugs.launchpad.net/charm-ceph-mon/+bug/1852441
https://opendev.org/openstack/charm-ceph-mon/commit/60a9a4f27aa1cddd717431abae7a26c46f5f8bd5

[Alex Kavanagh (ajkavanagh)]

Yes, looking at the source it's doing the wrong thing on focal. It would also be useful to detect if running in a container and then not install the appropriate package.

I'm not sure what it will do on series upgrade, but it's worth checking that ntp migrates to chrony properly.

[Drew Freiberger (afreiberger)]

I have run into this and it's actually causing knock on effect of ntpmon monitoring in the baremetal server that the container is running on. That's a separate bug, but it lead me to +1 this bug as an issue.

Regarding Nobuto's comments, 1. IIRC, bionic and focal should both be chrony, xenial was ntpd. 2. +1k to not run ntpd in a container just as was implemented in the ceph-mon charm.

Regarding ntp and baremetal, perhaps there should just be reliance on cs:ntp to handle the chronyd/ntpd decision and series-upgrade process since this charm installs but does not have a method to configure ntp per the environment's requirements, as it seems to only install the base ntp packages which assume access to global ubuntu NTP server pools.

[Aurelien Lourot (aurelien-lourot)]

Agreed, using cs:ntp as a subordinate instead of installing a package seems to be the way to go.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/charm-ceph-fs/+/807668
Committed: https://opendev.org/openstack/charm-ceph-fs/commit/2cfacc2646dc06d9aa2c9d3667b0a334858a3d30
Submitter: "Zuul (22348)"
Branch: master

commit 2cfacc2646dc06d9aa2c9d3667b0a334858a3d30
Author: Eric Chen <email address hidden>
Date: Tue Sep 7 05:56:03 2021 +0000

    Remove ntp installation in ceph-fs.

    For the case in container, it is not necessary to install ntp.
    For the bare metal, we now expect operators to use a subordinate
    ntp charm instead. Therefore, we remove ntp installation in
    ceph-fs.

    Closes-Bug: #1935013
    Change-Id: I48f1a6a1ee6153bbc86cc9acf9302d8eb9d7db43