Documentation and playbook for recovering control plane from node failure (#4146)

docs/recover-control-plane.md

@@ -0,0 +1,57 @@
+
+Recovering the control plane
+============================
+
+To recover from broken nodes in the control plane use the "recover\-control\-plane.yml" playbook.
+
+* Backup what you can
+* Provision new nodes to replace the broken ones
+* Place the surviving nodes of the control plane first in the "etcd" and "kube-master" groups
+* Add the new nodes below the surviving control plane nodes in the "etcd" and "kube-master" groups
+
+Examples of what broken means in this context:
+
+* One or more bare metal node(s) suffer from unrecoverable hardware failure
+* One or more node(s) fail during patching or upgrading
+* Etcd database corruption
+* Other node related failures leaving your control plane degraded or nonfunctional
+
+__Note that you need at least one functional node to be able to recover using this method.__
+
+## If etcd quorum is intact
+
+* Set the etcd member names of the broken node(s) in the variable "old\_etcd\_members", this variable is used to remove the broken nodes from the etcd cluster.
+```old_etcd_members=etcd2,etcd3```
+* If you reuse identities for your etcd nodes add the inventory names for those nodes to the variable "old\_etcds". This will remove any previously generated certificates for those nodes.
+```old_etcds=etcd2.example.com,etcd3.example.com```
+* If you would like to remove the broken node objects from the kubernetes cluster add their inventory names to the variable "old\_kube\_masters"
+```old_kube_masters=master2.example.com,master3.example.com```
+
+Then run the playbook with ```--limit etcd,kube-master```
+
+When finished you should have a fully working and highly available control plane again.
+
+## If etcd quorum is lost
+
+* If you reuse identities for your etcd nodes add the inventory names for those nodes to the variable "old\_etcds". This will remove any previously generated certificates for those nodes.
+```old_etcds=etcd2.example.com,etcd3.example.com```
+* If you would like to remove the broken node objects from the kubernetes cluster add their inventory names to the variable "old\_kube\_masters"
+```old_kube_masters=master2.example.com,master3.example.com```
+
+Then run the playbook with ```--limit etcd,kube-master```
+
+When finished you should have a fully working and highly available control plane again.
+
+The playbook will attempt to take a snapshot from the first node in the "etcd" group and restore from that. If you would like to restore from an alternate snapshot set the path to that snapshot in the "etcd\_snapshot" variable.
+
+```etcd_snapshot=/tmp/etcd_snapshot```
+
+## Caveats
+
+* The playbook has only been tested on control planes where the etcd and kube-master nodes are the same, the playbook will warn if run on a cluster with separate etcd and kube-master nodes.
+* The playbook has only been tested with fairly small etcd databases.
+* If your new control plane nodes have new ip addresses you may have to change settings in various places.
+* There may be disruptions while running the playbook.
+* There are absolutely no guarantees.
+
+If possible try to break a cluster in the same way that your target cluster is broken and test to recover that before trying on the real target cluster.

recover-control-plane.yml

@@ -0,0 +1,37 @@
+---
+- hosts: localhost
+  gather_facts: False
+  tasks:
+    - name: "Check ansible version !=2.7.0"
+      assert:
+        msg: "Ansible V2.7.0 can't be used until: https://github.com/ansible/ansible/issues/46600 is fixed"
+        that:
+          - ansible_version.string is version("2.7.0", "!=")
+          - ansible_version.string is version("2.6.0", ">=")
+      tags:
+        - check
+  vars:
+    ansible_connection: local
+
+- hosts: bastion[0]
+  gather_facts: False
+  roles:
+    - { role: kubespray-defaults}
+    - { role: bastion-ssh-config, tags: ["localhost", "bastion"]}
+
+- hosts: "{{ groups['etcd'] | first }}"
+  roles:
+    - { role: kubespray-defaults}
+    - { role: recover_control_plane/pre-recover }
+    - { role: recover_control_plane/etcd }
+
+- hosts: "{{ groups['kube-master'] | first }}"
+  roles:
+    - { role: recover_control_plane/master }
+
+- include: cluster.yml
+
+- hosts: "{{ groups['kube-master'] }}"
+  roles:
+    - { role: kubespray-defaults}
+    - { role: recover_control_plane/post-recover }

roles/recover_control_plane/etcd/tasks/main.yml

@@ -0,0 +1,7 @@
+---
+- include_tasks: prepare.yml
+
+- include_tasks: recover_lost_quorum.yml
+  when:
+    - has_etcdctl
+    - not etcd_cluster_is_healthy

roles/recover_control_plane/etcd/tasks/prepare.yml

@@ -0,0 +1,47 @@
+---
+- name: Delete old certificates
+  shell: "rm /etc/ssl/etcd/ssl/*{{ item }}* /etc/kubernetes/ssl/etcd/*{{ item }}*"
+  with_items: "{{ old_etcds.split(',') }}"
+  register: delete_old_cerificates
+  ignore_errors: true
+  when: old_etcds is defined
+
+- name: Fail if unable to delete old certificates
+  fail:
+    msg: "Unable to delete old certificates for: {{ item.item }}"
+  loop: "{{ delete_old_cerificates.results }}"
+  changed_when: false
+  when:
+    - old_etcds is defined
+    - "item.rc != 0 and not 'No such file or directory' in item.stderr"
+
+- name: Get etcd cluster members
+  shell: "{{ bin_dir }}/etcdctl member list"
+  register: member_list
+  changed_when: false
+  check_mode: no
+  environment:
+    - ETCDCTL_API: 3
+    - ETCDCTL_CA_FILE: /etc/ssl/etcd/ssl/ca.pem
+    - ETCDCTL_CERT: "/etc/ssl/etcd/ssl/admin-{{ inventory_hostname }}.pem"
+    - ETCDCTL_KEY: "/etc/ssl/etcd/ssl/admin-{{ inventory_hostname }}-key.pem"
+  when:
+    - has_etcdctl
+    - etcd_cluster_is_healthy
+    - old_etcd_members is defined
+
+- name: Remove old cluster members
+  shell: "{{ bin_dir}}/etcdctl --endpoints={{ etcd_access_addresses }} member remove {{ item[1].replace(' ','').split(',')[0] }}"
+  environment:
+    - ETCDCTL_API: 3
+    - ETCDCTL_CA_FILE: /etc/ssl/etcd/ssl/ca.pem
+    - ETCDCTL_CERT: "/etc/ssl/etcd/ssl/admin-{{ inventory_hostname }}.pem"
+    - ETCDCTL_KEY: "/etc/ssl/etcd/ssl/admin-{{ inventory_hostname }}-key.pem"
+  with_nested:
+    - "{{ old_etcd_members.split(',') }}"
+    - "{{ member_list.stdout_lines }}"
+  when:
+    - has_etcdctl
+    - etcd_cluster_is_healthy
+    - old_etcd_members is defined
+    - item[0] == item[1].replace(' ','').split(',')[2]

roles/recover_control_plane/etcd/tasks/recover_lost_quorum.yml

@@ -0,0 +1,54 @@
+---
+- name: Save etcd snapshot
+  shell: "{{ bin_dir }}/etcdctl snapshot save /tmp/snapshot.db"
+  environment:
+    - ETCDCTL_API: 3
+    - ETCDCTL_CA_FILE: /etc/ssl/etcd/ssl/ca.pem
+    - ETCDCTL_CERT: "/etc/ssl/etcd/ssl/member-{{ inventory_hostname }}.pem"
+    - ETCDCTL_KEY: "/etc/ssl/etcd/ssl/member-{{ inventory_hostname }}-key.pem"
+  when: etcd_snapshot is not defined
+
+- name: Transfer etcd snapshot to host
+  copy:
+    src: "{{ etcd_snapshot }}"
+    dest: /tmp/snapshot.db
+  when: etcd_snapshot is defined
+
+- name: Stop etcd
+  systemd:
+    name: etcd
+    state: stopped
+
+- name: Remove etcd data-dir
+  shell: "rm -rf {{ etcd_data_dir }}"
+
+- name: Restore etcd snapshot
+  shell: "{{ bin_dir }}/etcdctl snapshot restore /tmp/snapshot.db --name {{ etcd_member_name }} --initial-cluster {{ etcd_member_name }}={{ etcd_peer_url }} --initial-cluster-token k8s_etcd --initial-advertise-peer-urls {{ etcd_peer_url }} --data-dir {{ etcd_data_dir }}"
+  environment:
+    - ETCDCTL_API: 3
+    - ETCDCTL_CA_FILE: /etc/ssl/etcd/ssl/ca.pem
+    - ETCDCTL_CERT: "/etc/ssl/etcd/ssl/member-{{ inventory_hostname }}.pem"
+    - ETCDCTL_KEY: "/etc/ssl/etcd/ssl/member-{{ inventory_hostname }}-key.pem"
+
+- name: Remove etcd snapshot
+  file:
+    path: /tmp/snapshot.db
+    state: absent
+
+- name: Change etcd data-dir owner
+  file:
+    path: "{{ etcd_data_dir }}"
+    owner: etcd
+    group: etcd
+    recurse: true
+
+- name: Reconfigure etcd
+  replace:
+    path: /etc/etcd.env
+    regexp: "^(ETCD_INITIAL_CLUSTER=).*"
+    replace: '\1{{ etcd_member_name }}={{ etcd_peer_url }}'
+
+- name: Start etcd
+  systemd:
+    name: etcd
+    state: started

roles/recover_control_plane/master/tasks/main.yml

@@ -0,0 +1,28 @@
+---
+- name: Wait for apiserver
+  shell: "{{ bin_dir }}/kubectl get nodes"
+  environment:
+    - KUBECONFIG: /root/.kube/config
+  register: apiserver_is_ready
+  until: apiserver_is_ready.rc == 0
+  retries: 6
+  delay: 10
+  changed_when: false
+
+- name: Delete old kube-master nodes from cluster
+  shell: "{{ bin_dir }}/kubectl delete node {{ item }}"
+  environment:
+    - KUBECONFIG: /root/.kube/config
+  with_items: "{{ old_kube_masters.split(',') }}"
+  register: delete_old_kube_masters
+  failed_when: false
+  when: old_kube_masters is defined
+
+- name: Fail if unable to delete old kube-master nodes from cluster
+  fail:
+    msg: "Unable to delete old kube-master node: {{ item.item }}"
+  loop: "{{ delete_old_kube_masters.results }}"
+  changed_when: false
+  when:
+    - old_kube_masters is defined
+    - "item.rc != 0 and not 'NotFound' in item.stderr"

roles/recover_control_plane/post-recover/tasks/main.yml

@@ -0,0 +1,19 @@
+---
+# TODO: Figure out why kubeadm does not fix this
+- name: Set etcd-servers fact
+  set_fact:
+    etcd_servers: >-
+      {% for host in groups['etcd'] -%}
+        {% if not loop.last -%}
+        https://{{ hostvars[host].access_ip | default(hostvars[host].ip | default(hostvars[host].ansible_default_ipv4['address'])) }}:2379,
+        {%- endif -%}
+        {%- if loop.last -%}
+        https://{{ hostvars[host].access_ip | default(hostvars[host].ip | default(hostvars[host].ansible_default_ipv4['address'])) }}:2379
+        {%- endif -%}
+      {%- endfor -%}
+
+- name: Update apiserver etcd-servers list
+  replace:
+    path: /etc/kubernetes/manifests/kube-apiserver.yaml
+    regexp: "(etcd-servers=).*"
+    replace: "\\1{{ etcd_servers }}"

roles/recover_control_plane/pre-recover/defaults/main.yml

@@ -0,0 +1,2 @@
+---
+control_plane_is_converged: "{{ groups['etcd'] | sort == groups['kube-master'] | sort | bool }}"

roles/recover_control_plane/pre-recover/tasks/main.yml

@@ -0,0 +1,36 @@
+---
+- name: Check for etcdctl binary
+  raw: "test -e {{ bin_dir }}/etcdctl"
+  register: test_etcdctl
+
+- name: Set has_etcdctl fact
+  set_fact:
+    has_etcdctl: "{{ test_etcdctl.rc == 0 | bool }}"
+
+- name: Check if etcd cluster is healthy
+  shell: "{{ bin_dir }}/etcdctl --endpoints={{ etcd_access_addresses }} cluster-health | grep -q 'cluster is healthy'"
+  register: etcd_cluster_health
+  ignore_errors: true
+  changed_when: false
+  check_mode: no
+  environment:
+    ETCDCTL_CERT_FILE: "{{ etcd_cert_dir }}/admin-{{ inventory_hostname }}.pem"
+    ETCDCTL_KEY_FILE: "{{ etcd_cert_dir }}/admin-{{ inventory_hostname }}-key.pem"
+    ETCDCTL_CA_FILE: "{{ etcd_cert_dir }}/ca.pem"
+  when: has_etcdctl
+
+- name: Set etcd_cluster_is_healthy fact
+  set_fact:
+    etcd_cluster_is_healthy: "{{ etcd_cluster_health.rc == 0 | bool }}"
+
+- name: Abort if etcd cluster is healthy and old_etcd_members is undefined
+  assert:
+    that: "{{ old_etcd_members is defined }}"
+    msg: "'old_etcd_members' must be defined when the etcd cluster has quorum."
+  when: etcd_cluster_is_healthy
+
+- name: Warn for untested recovery
+  debug:
+    msg: Control plane recovery of split control planes is UNTESTED! Abort or continue at your own risk.
+  delay: 30
+  when: not control_plane_is_converged