This document describes the process to replace a single unhealthy etcd member.

This process depends on whether the etcd member is unhealthy because the machine is not running or the node is not ready, or whether it is unhealthy because the etcd Pod is crashlooping.

If you have lost the majority of your master hosts, leading to etcd quorum loss, then you must follow the disaster recovery procedure to restore to a previous cluster state instead of this procedure.

If the control plane certificates are not valid on the member being replaced, then you must follow the procedure to recover from expired control plane certificates instead of this procedure.

If a master node is lost and a new one is created, the etcd cluster Operator handles generating the new TLS certificates and adding the node as an etcd member.

Prerequisites

  • Take an etcd backup prior to replacing an unhealthy etcd member.

Identifying an unhealthy etcd member

You can identify if your cluster has an unhealthy etcd member.

Prerequisites
  • Access to the cluster as a user with the cluster-admin role.

Procedure
  1. Check the status of the EtcdMembersAvailable status condition using the following command:

    $ oc get etcd -o=jsonpath='{range .items[0].status.conditions[?(@.type=="EtcdMembersAvailable")]}{.message}{"\n"}'
  2. Review the output:

    2 of 3 members are available, ip-10-0-131-183.ec2.internal is unhealthy

    This example output shows that the ip-10-0-131-183.ec2.internal etcd member is unhealthy.

Determining the state of the unhealthy etcd member

The steps to replace an unhealthy etcd member depend on which of the following states your etcd member is in:

  • The machine is not running or the node is not ready

  • The etcd Pod is crashlooping

This procedure determines which state your etcd member is in. This enables you to know which procedure to follow to replace the unhealthy etcd member.

If you are aware that the machine is not running or the node is not ready, but you expect it to return to a healthy state soon, then you do not need to perform a procedure to replace the etcd member. The etcd cluster Operator will automatically sync when the machine or node returns to a healthy state.

Prerequisites
  • You have access to the cluster as a user with the cluster-admin role.

  • You have identified an unhealthy etcd member.

Procedure
  1. Determine if the machine is not running:

    $ oc get machines -A -ojsonpath='{range .items[*]}{@.status.nodeRef.name}{"\t"}{@.status.providerStatus.instanceState}{"\n"}' | grep -v running
    Example output
    ip-10-0-131-183.ec2.internal  stopped (1)
    
    1 This output lists the node and the status of the node’s machine. If the status is anything other than running, then the machine is not running.

    If the machine is not running, then follow the Replacing an unhealthy etcd member whose machine is not running or whose node is not ready procedure.

  2. Determine if the node is not ready.

    If either of the following scenarios are true, then the node is not ready.

    • If the machine is running, then check whether the node is unreachable:

      $ oc get nodes -o jsonpath='{range .items[*]}{"\n"}{.metadata.name}{"\t"}{range .spec.taints[*]}{.key}{" "}' | grep unreachable
      Example output
      ip-10-0-131-183.ec2.internal	node-role.kubernetes.io/master node.kubernetes.io/unreachable node.kubernetes.io/unreachable (1)
      
      1 If the node is listed with an unreachable taint, then the node is not ready.
    • If the node is still reachable, then check whether the node is listed as NotReady:

      $ oc get nodes -l node-role.kubernetes.io/master | grep "NotReady"
      Example output
      ip-10-0-131-183.ec2.internal   NotReady   master   122m   v1.19.0 (1)
      
      1 If the node is listed as NotReady, then the node is not ready.

    If the node is not ready, then follow the Replacing an unhealthy etcd member whose machine is not running or whose node is not ready procedure.

  3. Determine if the etcd Pod is crashlooping.

    If the machine is running and the node is ready, then check whether the etcd Pod is crashlooping.

    1. Verify that all master nodes are listed as Ready:

      $ oc get nodes -l node-role.kubernetes.io/master
      Example output
      NAME                           STATUS   ROLES    AGE     VERSION
      ip-10-0-131-183.ec2.internal   Ready    master   6h13m   v1.19.0
      ip-10-0-164-97.ec2.internal    Ready    master   6h13m   v1.19.0
      ip-10-0-154-204.ec2.internal   Ready    master   6h13m   v1.19.0
    2. Check whether the status of an etcd Pod is either Error or CrashloopBackoff:

      $ oc get pods -n openshift-etcd | grep etcd
      Example output
      etcd-ip-10-0-131-183.ec2.internal                2/3     Error       7          6h9m (1)
      etcd-ip-10-0-164-97.ec2.internal                 3/3     Running     0          6h6m
      etcd-ip-10-0-154-204.ec2.internal                3/3     Running     0          6h6m
      1 Since this status of this Pod is Error, then the etcd Pod is crashlooping.

    If the etcd Pod is crashlooping, then follow the Replacing an unhealthy etcd member whose etcd Pod is crashlooping procedure.

Replacing the unhealthy etcd member

Depending on the state of your unhealthy etcd member, use one of the following procedures:

Replacing an unhealthy etcd member whose machine is not running or whose node is not ready

This procedure details the steps to replace an etcd member that is unhealthy either because the machine is not running or because the node is not ready.

Prerequisites
  • You have identified the unhealthy etcd member.

  • You have verified that either the machine is not running or the node is not ready.

  • You have access to the cluster as a user with the cluster-admin role.

  • You have taken an etcd backup.

    It is important to take an etcd backup before performing this procedure so that your cluster can be restored if you encounter any issues.

Procedure
  1. Remove the unhealthy member.

    1. Choose a Pod that is not on the affected node:

      In a terminal that has access to the cluster as a cluster-admin user, run the following command:

      $ oc get pods -n openshift-etcd | grep etcd
      Example output
      etcd-ip-10-0-131-183.ec2.internal                3/3     Running     0          123m
      etcd-ip-10-0-164-97.ec2.internal                 3/3     Running     0          123m
      etcd-ip-10-0-154-204.ec2.internal                3/3     Running     0          124m
    2. Connect to the running etcd container, passing in the name of a Pod that is not on the affected node:

      In a terminal that has access to the cluster as a cluster-admin user, run the following command:

      $ oc rsh -n openshift-etcd etcd-ip-10-0-154-204.ec2.internal
    3. View the member list:

      sh-4.2# etcdctl member list -w table
      Example output
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      |        ID        | STATUS  |             NAME             |        PEER ADDRS         |       CLIENT ADDRS        |
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      | 6fc1e7c9db35841d | started | ip-10-0-131-183.ec2.internal | https://10.0.131.183:2380 | https://10.0.131.183:2379 |
      | 757b6793e2408b6c | started |  ip-10-0-164-97.ec2.internal |  https://10.0.164.97:2380 |  https://10.0.164.97:2379 |
      | ca8c2990a0aa29d1 | started | ip-10-0-154-204.ec2.internal | https://10.0.154.204:2380 | https://10.0.154.204:2379 |
      +------------------+---------+------------------------------+---------------------------+---------------------------+

      Take note of the ID and the name of the unhealthy etcd member, because these values are needed later in the procedure.

    4. Remove the unhealthy etcd member by providing the ID to the etcdctl member remove command:

      sh-4.2# etcdctl member remove 6fc1e7c9db35841d
      Example output
      Member 6fc1e7c9db35841d removed from cluster baa565c8919b060e
    5. View the member list again and verify that the member was removed:

      sh-4.2# etcdctl member list -w table
      Example output
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      |        ID        | STATUS  |             NAME             |        PEER ADDRS         |       CLIENT ADDRS        |
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      | 757b6793e2408b6c | started |  ip-10-0-164-97.ec2.internal |  https://10.0.164.97:2380 |  https://10.0.164.97:2379 |
      | ca8c2990a0aa29d1 | started | ip-10-0-154-204.ec2.internal | https://10.0.154.204:2380 | https://10.0.154.204:2379 |
      +------------------+---------+------------------------------+---------------------------+---------------------------+

      You can now exit the node shell.

  2. Remove the old secrets for the unhealthy etcd member that was removed.

    1. List the secrets for the unhealthy etcd member that was removed.

      $ oc get secrets -n openshift-etcd | grep ip-10-0-131-183.ec2.internal (1)
      1 Pass in the name of the unhealthy etcd member that you took note of earlier in this procedure.

      There is a peer, serving, and metrics secret as shown in the following output:

      Example output
      etcd-peer-ip-10-0-131-183.ec2.internal              kubernetes.io/tls                     2      47m
      etcd-serving-ip-10-0-131-183.ec2.internal           kubernetes.io/tls                     2      47m
      etcd-serving-metrics-ip-10-0-131-183.ec2.internal   kubernetes.io/tls                     2      47m
    2. Delete the secrets for the unhealthy etcd member that was removed.

      1. Delete the peer secret:

        $ oc delete secret -n openshift-etcd etcd-peer-ip-10-0-131-183.ec2.internal
      2. Delete the serving secret:

        $ oc delete secret -n openshift-etcd etcd-serving-ip-10-0-131-183.ec2.internal
      3. Delete the metrics secret:

        $ oc delete secret -n etcd-serving-metrics-ip-10-0-131-183.ec2.internal
  3. Delete and recreate the master machine. After this machine is recreated, a new revision is forced and etcd scales up automatically.

    If you are running installer-provisioned infrastructure, or you used the Machine API to create your machines, follow these steps. Otherwise, you must create the new master using the same method that was used to originally create it.

    1. Obtain the machine for the unhealthy member.

      In a terminal that has access to the cluster as a cluster-admin user, run the following command:

      $ oc get machines -n openshift-machine-api -o wide
      Example output
      NAME                                        PHASE     TYPE        REGION      ZONE         AGE     NODE                           PROVIDERID                              STATE
      clustername-8qw5l-master-0                  Running   m4.xlarge   us-east-1   us-east-1a   3h37m   ip-10-0-131-183.ec2.internal   aws:///us-east-1a/i-0ec2782f8287dfb7e   stopped (1)
      clustername-8qw5l-master-1                  Running   m4.xlarge   us-east-1   us-east-1b   3h37m   ip-10-0-154-204.ec2.internal   aws:///us-east-1b/i-096c349b700a19631   running
      clustername-8qw5l-master-2                  Running   m4.xlarge   us-east-1   us-east-1c   3h37m   ip-10-0-164-97.ec2.internal    aws:///us-east-1c/i-02626f1dba9ed5bba   running
      clustername-8qw5l-worker-us-east-1a-wbtgd   Running   m4.large    us-east-1   us-east-1a   3h28m   ip-10-0-129-226.ec2.internal   aws:///us-east-1a/i-010ef6279b4662ced   running
      clustername-8qw5l-worker-us-east-1b-lrdxb   Running   m4.large    us-east-1   us-east-1b   3h28m   ip-10-0-144-248.ec2.internal   aws:///us-east-1b/i-0cb45ac45a166173b   running
      clustername-8qw5l-worker-us-east-1c-pkg26   Running   m4.large    us-east-1   us-east-1c   3h28m   ip-10-0-170-181.ec2.internal   aws:///us-east-1c/i-06861c00007751b0a   running
      1 This is the master machine for the unhealthy node, ip-10-0-131-183.ec2.internal.
    2. Save the machine configuration to a file on your file system:

      $ oc get machine clustername-8qw5l-master-0 \ (1)
          -n openshift-machine-api \
          -o yaml \
          > new-master-machine.yaml
      1 Specify the name of the master machine for the unhealthy node.
    3. Edit the new-master-machine.yaml file that was created in the previous step.

      1. Remove the entire status section:

        status:
          addresses:
          - address: 10.0.131.183
            type: InternalIP
          - address: ip-10-0-131-183.ec2.internal
            type: InternalDNS
          - address: ip-10-0-131-183.ec2.internal
            type: Hostname
          lastUpdated: "2020-04-20T17:44:29Z"
          nodeRef:
            kind: Node
            name: ip-10-0-131-183.ec2.internal
            uid: acca4411-af0d-4387-b73e-52b2484295ad
          phase: Running
          providerStatus:
            apiVersion: awsproviderconfig.openshift.io/v1beta1
            conditions:
            - lastProbeTime: "2020-04-20T16:53:50Z"
              lastTransitionTime: "2020-04-20T16:53:50Z"
              message: machine successfully created
              reason: MachineCreationSucceeded
              status: "True"
              type: MachineCreation
            instanceId: i-0fdb85790d76d0c3f
            instanceState: stopped
            kind: AWSMachineProviderStatus
      2. Remove the providerID field:

          providerID: aws:///us-east-1a/i-0fdb85790d76d0c3f
      3. Change the name field to a new name.

        It is recommended to keep the same base name as the old machine and change the ending number to the next available number. In this example, clustername-8qw5l-master-0 is changed to clustername-8qw5l-master-3.

        For example:

        apiVersion: machine.openshift.io/v1beta1
        kind: Machine
        metadata:
          ...
          name: clustername-8qw5l-master-3
          ...
      4. Update the selfLink field to use the new machine name from the previous step.

        apiVersion: machine.openshift.io/v1beta1
        kind: Machine
        metadata:
          ...
          selfLink: /apis/machine.openshift.io/v1beta1/namespaces/openshift-machine-api/machines/clustername-8qw5l-master-3
          ...
    4. Delete the machine of the unhealthy member:

      $ oc delete machine -n openshift-machine-api clustername-8qw5l-master-0 (1)
      1 Specify the name of the master machine for the unhealthy node.
    5. Verify that the machine was deleted:

      $ oc get machines -n openshift-machine-api -o wide
      Example output
      NAME                                        PHASE     TYPE        REGION      ZONE         AGE     NODE                           PROVIDERID                              STATE
      clustername-8qw5l-master-1                  Running   m4.xlarge   us-east-1   us-east-1b   3h37m   ip-10-0-154-204.ec2.internal   aws:///us-east-1b/i-096c349b700a19631   running
      clustername-8qw5l-master-2                  Running   m4.xlarge   us-east-1   us-east-1c   3h37m   ip-10-0-164-97.ec2.internal    aws:///us-east-1c/i-02626f1dba9ed5bba   running
      clustername-8qw5l-worker-us-east-1a-wbtgd   Running   m4.large    us-east-1   us-east-1a   3h28m   ip-10-0-129-226.ec2.internal   aws:///us-east-1a/i-010ef6279b4662ced   running
      clustername-8qw5l-worker-us-east-1b-lrdxb   Running   m4.large    us-east-1   us-east-1b   3h28m   ip-10-0-144-248.ec2.internal   aws:///us-east-1b/i-0cb45ac45a166173b   running
      clustername-8qw5l-worker-us-east-1c-pkg26   Running   m4.large    us-east-1   us-east-1c   3h28m   ip-10-0-170-181.ec2.internal   aws:///us-east-1c/i-06861c00007751b0a   running
    6. Create the new machine using the new-master-machine.yaml file:

      $ oc apply -f new-master-machine.yaml
    7. Verify that the new machine has been created:

      $ oc get machines -n openshift-machine-api -o wide
      Example output
      NAME                                        PHASE          TYPE        REGION      ZONE         AGE     NODE                           PROVIDERID                              STATE
      clustername-8qw5l-master-1                  Running        m4.xlarge   us-east-1   us-east-1b   3h37m   ip-10-0-154-204.ec2.internal   aws:///us-east-1b/i-096c349b700a19631   running
      clustername-8qw5l-master-2                  Running        m4.xlarge   us-east-1   us-east-1c   3h37m   ip-10-0-164-97.ec2.internal    aws:///us-east-1c/i-02626f1dba9ed5bba   running
      clustername-8qw5l-master-3                  Provisioning   m4.xlarge   us-east-1   us-east-1a   85s     ip-10-0-133-53.ec2.internal    aws:///us-east-1a/i-015b0888fe17bc2c8   running (1)
      clustername-8qw5l-worker-us-east-1a-wbtgd   Running        m4.large    us-east-1   us-east-1a   3h28m   ip-10-0-129-226.ec2.internal   aws:///us-east-1a/i-010ef6279b4662ced   running
      clustername-8qw5l-worker-us-east-1b-lrdxb   Running        m4.large    us-east-1   us-east-1b   3h28m   ip-10-0-144-248.ec2.internal   aws:///us-east-1b/i-0cb45ac45a166173b   running
      clustername-8qw5l-worker-us-east-1c-pkg26   Running        m4.large    us-east-1   us-east-1c   3h28m   ip-10-0-170-181.ec2.internal   aws:///us-east-1c/i-06861c00007751b0a   running
      1 The new machine, clustername-8qw5l-master-3 is being created and is ready once the phase changes from Provisioning to Running.

      It might take a few minutes for the new machine to be created. The etcd cluster Operator will automatically sync when the machine or node returns to a healthy state.

  4. Verify that all etcd Pods are running properly:

    In a terminal that has access to the cluster as a cluster-admin user, run the following command:

    $ oc get pods -n openshift-etcd | grep etcd
    Example output
    etcd-ip-10-0-133-53.ec2.internal                 3/3     Running     0          7m49s
    etcd-ip-10-0-164-97.ec2.internal                 3/3     Running     0          123m
    etcd-ip-10-0-154-204.ec2.internal                3/3     Running     0          124m

Replacing an unhealthy etcd member whose etcd Pod is crashlooping

This procedure details the steps to replace an etcd member that is unhealthy because the etcd Pod is crashlooping.

Prerequisites
  • You have identified the unhealthy etcd member.

  • You have verified that the etcd Pod is crashlooping.

  • You have access to the cluster as a user with the cluster-admin role.

  • You have taken an etcd backup.

    It is important to take an etcd backup before performing this procedure so that your cluster can be restored if you encounter any issues.

Procedure
  1. Stop the crashlooping etcd Pod.

    1. Debug the node that is crashlooping.

      In a terminal that has access to the cluster as a cluster-admin user, run the following command:

      $ oc debug node/ip-10-0-131-183.ec2.internal (1)
      1 Replace this with the name of the unhealthy node.
    2. Change your root directory to the host:

      sh-4.2# chroot /host
    3. Move the existing etcd Pod file out of the kubelet manifest directory:

      sh-4.2# mkdir /var/lib/etcd-backup
      sh-4.2# mv /etc/kubernetes/manifests/etcd-pod.yaml /var/lib/etcd-backup/
    4. Move the etcd data directory to a different location:

      sh-4.2# mv /var/lib/etcd/ /tmp

      You can now exit the node shell.

  2. Remove the unhealthy member.

    1. Choose a Pod that is not on the affected node.

      In a terminal that has access to the cluster as a cluster-admin user, run the following command:

      $ oc get pods -n openshift-etcd | grep etcd
      Example output
      etcd-ip-10-0-131-183.ec2.internal                2/3     Error       7          6h9m
      etcd-ip-10-0-164-97.ec2.internal                 3/3     Running     0          6h6m
      etcd-ip-10-0-154-204.ec2.internal                3/3     Running     0          6h6m
    2. Connect to the running etcd container, passing in the name of a Pod that is not on the affected node.

      In a terminal that has access to the cluster as a cluster-admin user, run the following command:

      $ oc rsh -n openshift-etcd etcd-ip-10-0-154-204.ec2.internal
    3. View the member list:

      sh-4.2# etcdctl member list -w table
      Example output
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      |        ID        | STATUS  |             NAME             |        PEER ADDRS         |       CLIENT ADDRS        |
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      | 62bcf33650a7170a | started | ip-10-0-131-183.ec2.internal | https://10.0.131.183:2380 | https://10.0.131.183:2379 |
      | b78e2856655bc2eb | started |  ip-10-0-164-97.ec2.internal |  https://10.0.164.97:2380 |  https://10.0.164.97:2379 |
      | d022e10b498760d5 | started | ip-10-0-154-204.ec2.internal | https://10.0.154.204:2380 | https://10.0.154.204:2379 |
      +------------------+---------+------------------------------+---------------------------+---------------------------+

      Take note of the ID and the name of the unhealthy etcd member, because these values are needed later in the procedure.

    4. Remove the unhealthy etcd member by providing the ID to the etcdctl member remove command:

      sh-4.2# etcdctl member remove 62bcf33650a7170a
      Example output
      Member 62bcf33650a7170a removed from cluster ead669ce1fbfb346
    5. View the member list again and verify that the member was removed:

      sh-4.2# etcdctl member list -w table
      Example output
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      |        ID        | STATUS  |             NAME             |        PEER ADDRS         |       CLIENT ADDRS        |
      +------------------+---------+------------------------------+---------------------------+---------------------------+
      | b78e2856655bc2eb | started |  ip-10-0-164-97.ec2.internal |  https://10.0.164.97:2380 |  https://10.0.164.97:2379 |
      | d022e10b498760d5 | started | ip-10-0-154-204.ec2.internal | https://10.0.154.204:2380 | https://10.0.154.204:2379 |
      +------------------+---------+------------------------------+---------------------------+---------------------------+

      You can now exit the node shell.

  3. Remove the old secrets for the unhealthy etcd member that was removed.

    1. List the secrets for the unhealthy etcd member that was removed.

      $ oc get secrets -n openshift-etcd | grep ip-10-0-131-183.ec2.internal (1)
      1 Pass in the name of the unhealthy etcd member that you took note of earlier in this procedure.

      There is a peer, serving, and metrics secret as shown in the following output:

      Example output
      etcd-peer-ip-10-0-131-183.ec2.internal              kubernetes.io/tls                     2      47m
      etcd-serving-ip-10-0-131-183.ec2.internal           kubernetes.io/tls                     2      47m
      etcd-serving-metrics-ip-10-0-131-183.ec2.internal   kubernetes.io/tls                     2      47m
    2. Delete the secrets for the unhealthy etcd member that was removed.

      1. Delete the peer secret:

        $ oc delete secret -n openshift-etcd etcd-peer-ip-10-0-131-183.ec2.internal
      2. Delete the serving secret:

        $ oc delete secret -n openshift-etcd etcd-serving-ip-10-0-131-183.ec2.internal
      3. Delete the metrics secret:

        $ oc delete secret -n etcd-serving-metrics-ip-10-0-131-183.ec2.internal
  4. Force etcd redeployment.

    In a terminal that has access to the cluster as a cluster-admin user, run the following command:

    $ oc patch etcd cluster -p='{"spec": {"forceRedeploymentReason": "single-master-recovery-'"$( date --rfc-3339=ns )"'"}}' --type=merge (1)
    1 The forceRedeploymentReason value must be unique, which is why a timestamp is appended.

    When the etcd cluster Operator performs a redeployment, it ensures that all master nodes have a functioning etcd Pod.

  5. Verify that the new member is available and healthy.

    1. Connect to the running etcd container again.

      In a terminal that has access to the cluster as a cluster-admin user, run the following command:

      $ oc rsh -n openshift-etcd etcd-ip-10-0-154-204.ec2.internal
    2. Verify that all members are healthy:

      sh-4.2# etcdctl endpoint health --cluster
      Example output
      https://10.0.131.183:2379 is healthy: successfully committed proposal: took = 16.671434ms
      https://10.0.154.204:2379 is healthy: successfully committed proposal: took = 16.698331ms
      https://10.0.164.97:2379 is healthy: successfully committed proposal: took = 16.621645ms