Replace a failed Ceph node¶

After a physical node replacement, you can use the Ceph LCM API to redeploy failed Ceph nodes. The common flow of replacing a failed Ceph node is as follows:

Remove the obsolete Ceph node from the Ceph cluster.
Add a new Ceph node with the same configuration to the Ceph cluster.

Note

Ceph OSD node replacement presupposes usage of a KaaSCephOperationRequest CR. For workflow overview, spec and phases description, see High-level workflow of Ceph OSD or node removal.

Remove a failed Ceph node¶

Open the KaasCephCluster CR of a managed cluster for editing:
```
kubectl edit kaascephcluster -n <managedClusterProjectName>
```
Substitute <managedClusterProjectName> with the corresponding value.

In the nodes section, remove the required device:

spec:
  cephClusterSpec:
    nodes:
      <machineName>: # remove the entire entry for the node to replace
        storageDevices: {...}
        role: [...]

Substitute <machineName> with the machine name to replace.

Save KaaSCephCluster and close the editor.

Create a KaaSCephOperationRequest CR template and save it as replace-failed-<machineName>-request.yaml:

apiVersion: kaas.mirantis.com/v1alpha1
kind: KaaSCephOperationRequest
metadata:
  name: replace-failed-<machineName>-request
  namespace: <managedClusterProjectName>
spec:
  osdRemove:
    nodes:
      <machineName>:
        completeCleanUp: true
  kaasCephCluster:
    name: <kaasCephClusterName>
    namespace: <managedClusterProjectName>

Substitute <kaasCephClusterName> with the corresponding KaaSCephCluster resource from the <managedClusterProjectName> namespace.

Apply the template to the cluster:

kubectl apply -f replace-failed-<machineName>-request.yaml

Verify that the corresponding request has been created:

kubectl get kaascephoperationrequest -n <managedClusterProjectName>

Verify that the removeInfo section appeared in the KaaSCephOperationRequest CR status:

kubectl -n <managedClusterProjectName> get kaascephoperationrequest replace-failed-<machineName>-request -o yaml

Example of system response:

status:
  childNodesMapping:
    <nodeName>: <machineName>
  osdRemoveStatus:
    removeInfo:
      cleanUpMap:
        <nodeName>:
          osdMapping:
            ...
            <osdId>:
              deviceMapping:
                ...
                <deviceName>:
                  path: <deviceByPath>
                  partition: "/dev/ceph-b-vg_sdb/osd-block-b-lv_sdb"
                  type: "block"
                  class: "hdd"
                  zapDisk: true

If needed, change the following values:

<machineName> - machine name where the replacement occurs, for example, worker-1.
<nodeName> - underlying machine node name, for example, kaas-node-5a74b669-7e53-4535-aabd-5b509ec844af.
<osdId> - actual Ceph OSD ID for the device being replaced, for example, 1.
<deviceName> - actual device name placed on the node, for example, sdb.
<deviceByPath> - actual device by-path placed on the node, for example, /dev/disk/by-path/pci-0000:00:1t.9.

Verify that the cleanUpMap section matches the required removal and wait for the ApproveWaiting phase to appear in status:

kubectl -n <managedClusterProjectName> get kaascephoperationrequest replace-failed-<machineName>-request -o yaml

Example of system response:

status:
  phase: ApproveWaiting

Edit the KaaSCephOperationRequest CR and set the approve flag to true:

kubectl -n <managedClusterProjectName> edit kaascephoperationrequest replace-failed-<machineName>-request

For example:

spec:
  osdRemove:
    approve: true

Review the following status fields of the KaaSCephOperationRequest CR request processing:
- status.phase - current state of request processing
- status.messages - description of the current phase
- status.conditions - full history of request processing before the current phase
- status.removeInfo.issues and status.removeInfo.warnings - error and warning messages occurred during request processing, if any

Verify that the KaaSCephOperationRequest has been completed. For example:

status:
  phase: Completed # or CompletedWithWarnings if there are non-critical issues

Remove the device cleanup jobs:

kubectl delete jobs -n ceph-lcm-mirantis -l app=miraceph-cleanup-disks

Deploy a new Ceph node after removal of a failed one¶

Note

You can spawn Ceph OSD on a raw device, but it must be clean and without any data or partitions. If you want to add a device that was in use, also ensure it is raw and clean. To clean up all data and partitions from a device, refer to official Rook documentation.

Open the KaasCephCluster CR of a managed cluster for editing:
```
kubectl edit kaascephcluster -n <managedClusterProjectName>
```
Substitute <managedClusterProjectName> with the corresponding value.

In the nodes section, add a new device:

spec:
  cephClusterSpec:
    nodes:
      <machineName>: # add new configuration for replaced Ceph node
        storageDevices:
        - fullPath: <deviceByID> # Recommended since MCC 2.25.0 (17.0.0), non-wwn by-id symlink
          # name: <deviceByID> # Prior MCC 2.25.0, non-wwn by-id symlink
          # fullPath: <deviceByPath> # if device is supposed to be added with by-path
          config:
            deviceClass: hdd
          ...

Substitute <machineName> with the machine name of the replaced node and configure it as required.

Warning

Since MCC 2.25.0 (17.0.0), Mirantis highly recommends using non-wwn by-id symlinks only to specify storage devices in the storageDevices list.

For details, see Container Cloud documentation: Addressing storage devices.

Verify that all Ceph daemons from the replaced node have appeared on the Ceph cluster and are in and up. The fullClusterInfo section should not contain any issues.

kubectl -n <managedClusterProjectName> get kaascephcluster -o yaml

Example of system response:

status:
  fullClusterInfo:
    clusterStatus:
      ceph:
        health: HEALTH_OK
        ...
    daemonStatus:
      mgr:
        running: a is active mgr
        status: Ok
      mon:
        running: '3/3 mons running: [a b c] in quorum'
        status: Ok
      osd:
        running: '3/3 running: 3 up, 3 in'
        status: Ok

Verify the Ceph node on the managed cluster:

kubectl -n rook-ceph get pod -o wide | grep <machineName>