Bug #47126
openmgr plugin rpms unconditionally restart ceph-mgr.target on install/uninstall
0%
Description
In nautilus all of the mgr plugin rpms have a post install and post uninstall scriptlet which unconditionally restarts the ceph-mgr.target, e.g:
%post mgr-dashboard
if [ $1 -eq 1 ] ; then
/usr/bin/systemctl try-restart ceph-mgr.target >/dev/null 2>&1 || :
fi
%postun mgr-dashboard
if [ $1 -eq 1 ] ; then
/usr/bin/systemctl try-restart ceph-mgr.target >/dev/null 2>&1 || :
fi
This makes it difficult or perhaps impossible to get the upgrade ordering procedure correct (i.e. it is difficult to upgrade the mons before the mgrs.)
Updated by Sebastian Wagner over 3 years ago
Hm. In order to properly install a module, you have to restart the mgr.
obviously, containers solve this issues.
Updated by Dan van der Ster over 3 years ago
Sebastian Wagner wrote:
Hm. In order to properly install a module, you have to restart the mgr.
that's clear, but when we should restart is obviously depending on a particular deployment, most important if mon/mgr is cohosted on the same box, which I presume is very very common.
Elsewhere in the ceph.spec daemons are checking the CEPH_AUTO_RESTART_ON_UPGRADE value in /etc/sysconfig/ceph (which defaults to "no", btw), e.g.
%postun mgr
%if 0%{?suse_version}
DISABLE_RESTART_ON_UPDATE="yes"
%service_del_postun ceph-mgr@\*.service ceph-mgr.target
%endif
%if 0%{?fedora} || 0%{?rhel}
%systemd_postun ceph-mgr@\*.service ceph-mgr.target
%endif
if [ $1 -ge 1 ] ; then
# Restart on upgrade, but only if "CEPH_AUTO_RESTART_ON_UPGRADE" is set to
# "yes". In any case: if units are not running, do not touch them.
SYSCONF_CEPH=%{_sysconfdir}/sysconfig/ceph
if [ -f $SYSCONF_CEPH -a -r $SYSCONF_CEPH ] ; then
source $SYSCONF_CEPH
fi
if [ "X$CEPH_AUTO_RESTART_ON_UPGRADE" = "Xyes" ] ; then
/usr/bin/systemctl try-restart ceph-mgr@\*.service > /dev/null 2>&1 || :
fi
fi