Cray System Management Documentation > Cray System Management (CSM) Administration Guide > utility storage > Shrink the Ceph Cluster

Shrink the Ceph Cluster

This procedure describes how to remove a Ceph node from the Ceph cluster. Once the node is removed, the cluster is also rebalanced to account for the changes. Use this procedure to reduce the size of a cluster.

Prerequisites

This procedure requires administrative privileges and three ssh sessions.
- One to monitor the cluster.
- One to perform cluster wide actions from a ceph-mon node.
- One to perform node only actions on the node being removed.

IMPORTANT NOTES

Permanent removal of ncn-s001, ncn-s002, or ncn-s003 is NOT SUPPORTED. They can only be rebuilt in place or replaced with new hardware. This is due to the Ceph mon and mgr processes running on them.
Always ensure you have the free capacity to remove the node(s) prior to performing this task.
When removing a node other than ncn-s001, ncn-s002, or ncn-s003, the SMF pools quotas must be adjusted accordingly.
Removal of more than one node at a time is NOT SUPPORTED because the SMA telemetry pool only has 2 copies.

Procedure

Log in as root on ncn-s001, ncn-s002, ncn-s003, or a master node.
Monitor the progress of the OSDs that have been added.
```
watch ceph -s
```

View the status of each OSD and see where they reside.

ncn-s00(1/2/3)# ceph osd tree

Example output:

ID  CLASS  WEIGHT    TYPE NAME          STATUS  REWEIGHT  PRI-AFF
-1         31.43875  root default
-3         10.47958      host ncn-s001
 4    ssd   1.74660          osd.4          up   1.00000  1.00000
 5    ssd   1.74660          osd.5          up   1.00000  1.00000
10    ssd   1.74660          osd.10         up   1.00000  1.00000
12    ssd   1.74660          osd.12         up   1.00000  1.00000
13    ssd   1.74660          osd.13         up   1.00000  1.00000
14    ssd   1.74660          osd.14         up   1.00000  1.00000
-5          6.98639      host ncn-s002
 0    ssd   1.74660          osd.0          up   1.00000  1.00000
 3    ssd   1.74660          osd.3          up   1.00000  1.00000
 6    ssd   1.74660          osd.6          up   1.00000  1.00000
 9    ssd   1.74660          osd.9          up   1.00000  1.00000
-7          6.98639      host ncn-s003
 2    ssd   1.74660          osd.2          up   1.00000  1.00000
 7    ssd   1.74660          osd.7          up   1.00000  1.00000
 8    ssd   1.74660          osd.8          up   1.00000  1.00000
11    ssd   1.74660          osd.11         up   1.00000  1.00000
-9          6.98639      host ncn-s004
 1    ssd   1.74660          osd.1          up   1.00000  1.00000
15    ssd   1.74660          osd.15         up   1.00000  1.00000
16    ssd   1.74660          osd.16         up   1.00000  1.00000
17    ssd   1.74660          osd.17         up   1.00000  1.00000

Set the NODE variable.
```
export NODE=<node being removed>
```
Reweigh the OSD(s) on the node being removed to rebalance the cluster.
1. Change the weight and CRUSH weight of the OSD being removed to 0.
```
ncn-s00(1/2/3)# for osd in $(ceph osd ls-tree $NODE); do
          ceph osd reweight osd.$osd 0;
          ceph osd crush reweight osd.$osd 0;
       done
```
2. Watch the ceph -s output until the cluster status is HEALTH_OK and the Rebalancing has completed.

Remove the OSD after the reweighing work is complete.

ncn-s00(1/2/3)# for osd in $(ceph osd ls-tree $NODE); do
          ceph osd down osd.$osd;
          ceph osd destroy osd.$osd --force;
          ceph osd purge osd.$osd --force;
       done

Remove any weight 0 orphaned OSDs.

If orphaned OSDs from the host $NODE remain that have weight 0, then remove those OSDs.

ncn-s00(1/2/3)# ceph osd tree

Example output:

ID  CLASS  WEIGHT    TYPE NAME          STATUS  REWEIGHT  PRI-AFF
-1         24.45236  root default
-3         10.47958      host ncn-s001
 4    ssd   1.74660          osd.4          up   1.00000  1.00000
 5    ssd   1.74660          osd.5          up   1.00000  1.00000
10    ssd   1.74660          osd.10         up   1.00000  1.00000
12    ssd   1.74660          osd.12         up   1.00000  1.00000
13    ssd   1.74660          osd.13         up   1.00000  1.00000
14    ssd   1.74660          osd.14         up   1.00000  1.00000
-5          6.98639      host ncn-s002
 0    ssd   1.74660          osd.0          up   1.00000  1.00000
 3    ssd   1.74660          osd.3          up   1.00000  1.00000
 6    ssd   1.74660          osd.6          up   1.00000  1.00000
 9    ssd   1.74660          osd.9          up   1.00000  1.00000
-7          6.98639      host ncn-s003
 2    ssd   1.74660          osd.2          up   1.00000  1.00000
 7    ssd   1.74660          osd.7          up   1.00000  1.00000
 8    ssd   1.74660          osd.8          up   1.00000  1.00000
11    ssd   1.74660          osd.11         up   1.00000  1.00000
-9                0      host ncn-s004
 1                0  osd.1                  up   1.00000  1.00000  <--- orphan

ceph osd down osd.1; ceph osd destroy osd.1 --force; ceph osd purge osd.1 --force

Regenerate Rados-GW Load Balancer Configuration.

Update the existing HAProxy configuration to remove the node from the configuration.

ncn-s00(1/2/3)# vi /etc/haproxy/haproxy.cfg

This example removes node ncn-s004 from the backend rgw-backend.

...
backend rgw-backend
    option forwardfor
    balance static-rr
    option httpchk GET /
        server server-ncn-s001-rgw0 10.252.1.6:8080 check weight 100
        server server-ncn-s002-rgw0 10.252.1.5:8080 check weight 100
        server server-ncn-s003-rgw0 10.252.1.4:8080 check weight 100
        server server-ncn-s004-rgw0 10.252.1.13:8080 check weight 100   <--- Line to remove
...

Copy the HAproxy configuration from ncn-s001 to all the storage nodes. Adjust the command based on the number of storage nodes.
```
pdcp -w ncn-s00[2-(end node number)] /etc/haproxy/haproxy.cfg /etc/haproxy/haproxy.cfg
```

Restart HAproxy on all the storage nodes, and stop HAproxy and KeepAlived on the node that is being removed.

pdsh -w ncn-s00[1-(end node number)] -f 2 'systemctl restart haproxy.service'
pdsh -w $NODE 'systemctl stop haproxy.service; systemctl stop keepalived.service'

Redeploy the Rados Gateway containers to adjust the placement group.

ceph orch apply rgw site1 zone1 --placement="<num-daemons> <node1 node2 node3 node4 ... >" --port=8080

For example:

ceph orch apply rgw site1 zone1 --placement="3 ncn-s001 ncn-s002 ncn-s003" --port=8080

Verify that the Rados Gateway is running on the desired nodes.

ceph orch ps --daemon_type rgw

Example output:

NAME                             HOST      STATUS         REFRESHED  AGE  VERSION  IMAGE NAME                         IMAGE ID      CONTAINER ID
rgw.site1.zone1.ncn-s001.kvskqt  ncn-s001  running (41m)  6m ago     41m  15.2.8   registry.local/ceph/ceph:v15.2.8   553b0cb212c   6e323878db46
rgw.site1.zone1.ncn-s002.tisuez  ncn-s002  running (41m)  6m ago     41m  15.2.8   registry.local/ceph/ceph:v15.2.8   553b0cb212c   278830a273d3
rgw.site1.zone1.ncn-s003.nnwuqy  ncn-s003  running (41m)  6m ago     41m  15.2.8   registry.local/ceph/ceph:v15.2.8   553b0cb212c   a9706e6d7a69

Remove the node from the cluster.

ncn-s00(1/2/3)# ceph orch host rm $NODE

Remove the Ceph configuration from the node.

On the node being removed
```
cephadm rm-cluster --fsid $(cephadm ls|jq -r .[1].fsid) --force
```

Remove the node from the CRUSH map.

ncn-s00(1/2/3)# ceph osd crush rm $NODE

In the output from ceph -s, verify that the status is HEALTH_OK.

NOTE If ncn-s001, ncn-s002, or ncn-s003 has been temporarily removed, HEALTH_WARN will be seen until the storage node is added back to the cluster.

 health: HEALTH_WARN
         1 stray daemons(s) not managed by cephadm
         1 stray host(s) with 1 daemon(s) not managed by cephadm
         1/3 mons down, quorum ncn-s003,ncn-s002
         Degraded data redundancy:...