Removing a node from Ceph cluster
Contents
17. Removing a node from Ceph cluster¶
17.1. Remove OSD from the cluster¶
Take OSD out of the cluster and bring it down, for example osd.3:
ceph osd out osd.3
ceph osd down osd.3
Stop the OSD deamon on the node which runs the OSD:
systemctl stop ceph-osd@3.service
Remove the OSD completely from the cluster:
ceph osd purge osd.3 --yes-i-really-mean-it
17.2. Stop the metadata and manager daemons on the node¶
ceph mds ok-to-stop node10
systemctl stop ceph-mds@node10.service
systemctl stop ceph-mgr@node10.service
17.3. Remove the monitor¶
On the surviving monitor, remove the other monitor:
ceph mon remove node10
Stop the monitor daemon on the node:
systemctl stop ceph-mon@node10
17.4. Remove the node from the configuration file¶
Edit /etc/ceph/ceph.conf file and remove the node IP address from
mon_hostline.Restart the monitor
systemctl restart ceph-mon@node06
17.5. Remove the node from the crash map¶
In the dashboard, the node still shows up in the crash map.
To remove it, we need to decompile the crash map, then modify it.
Dump the crash map in file /tmp/crash:
ceph osd getcrushmap -o /tmp/crush
Decompile the file:
crushtool -d /tmp/crush -o /tmp/crush_decomp
Edit file /tmp/crash_decomp and remove all the entries and blocks related to the removed node.
Compile the modified crash map:
crushtool -c /tmp/crush_decomp -o /tmp/crush_new
Import the modified crash map:
ceph osd setcrushmap -i /tmp/crush_new
Check the ceph status:
ceph -s