Ceph

Trying to achieve a mimic to nautilus upgrade using ceph-ansible, I realized even when CEPH_AUTO_RESTART_ON_UPGRADE is set:

[root@mon1 ~]# grep CEPH_AUTO_RESTART_ON_UPGRADE /etc/sysconfig/ceph
CEPH_AUTO_RESTART_ON_UPGRADE=no

when a ceph RPM gets upgraded, it restarts the associated service anyway:

Mar 14 19:07:55 mon1 ceph-mon⁴⁸²⁴: 2019-03-14 19:07:55.978 7f1c6fc28700 1 received signal: Terminated from /usr/lib/systemd/systemd --system --deserialize 17 (PID: 1) UID: 0
Mar 14 19:07:55 mon1 ceph-mon⁴⁸²⁴: 2019-03-14 19:07:55.978 7f1c6fc28700 -1 mon.mon1@1(peon) e1 * Got Signal Terminated *
Mar 14 19:07:55 mon1 systemd¹: Stopping Ceph cluster monitor daemon...
- Subject: Unit ceph-mon@mon1.service has begun shutting down

step to reproduce (without ceph-ansible):
1/ install a ceph RPM with ceph/mimic repos enabled on a machine.
2/ switch repo from mimic to chacra (to get latest ceph build from master)
3/ upgrade a ceph RPM (eg: yum install ceph-mon)

in my case, I'm testing with this environment:
- 3x mon (mon0,mon1,mon2)
- 1x mgr (collocated on mon0)
- 2x osd (osd0,osd1)

when RPM ceph-mon gets upgraded on mon0, RPM ceph-mgr gets upgraded too because of RPM dependency.
rolling_update playbook is intended to upgrades the mgrs after all mons are upgraded, but in that case it means
the mgrs are upgraded in parallel of monitors before we reach the mgr upgrade part in the playbook which is an issue in the upgrade process.

"mon": {
        "ceph version 13.2.5 (cbff874f9007f1869bfd3821b7e33b2a6ffd4988) mimic (stable)": 2,
        "ceph version 14.1.1-103-g16560e0 (16560e0706edde9d77198768a80e83468ee29339) nautilus (rc)": 1
    },
    "mgr": {
        "ceph version 14.1.1-103-g16560e0 (16560e0706edde9d77198768a80e83468ee29339) nautilus (rc)": 1
    },

Additional information:
I didn't reproduce this behavior in a luminous to mimic upgrade context