Hi,
just filled a bug on the ceph bug tracker. If you reboot an PVE-node, which has ceph mon + osds running, it's takes more time to create an new correct osdmap, because the osd marked itself down, but due the mon lost at the same time it's takes min. 20 sec longer to reconised the new situation by the other mons.
For less IO-trouble it's makes sense to stop the osds first before do an reboot (this should later do the systemd-script automaticly).
http://tracker.ceph.com/issues/18516
Udo
just filled a bug on the ceph bug tracker. If you reboot an PVE-node, which has ceph mon + osds running, it's takes more time to create an new correct osdmap, because the osd marked itself down, but due the mon lost at the same time it's takes min. 20 sec longer to reconised the new situation by the other mons.
For less IO-trouble it's makes sense to stop the osds first before do an reboot (this should later do the systemd-script automaticly).
http://tracker.ceph.com/issues/18516
Udo