prole/infrastructure/roles/k3s/tasks/install.yml
chrisfu d397b3000e Fix installer ctx apply flow and registry namespace
- Add explicit Apply button for kubectl context switching to avoid half-applied changes

- Allow kubeconfig/context-based auth without requiring K3S_TOKEN when kubeconfig is valid

- Prompt cleanup/reset of previous Common Core services namespace to prevent resource collisions

- Remove hardcoded 'common-services' registry namespace; default registry deploy/check to SERVICE_NAMESPACE/REGISTRY_NAMESPACE

- Update mocks and add regression tests for namespace resolution and installer behavior
2026-03-07 10:35:27 -08:00

568 lines
21 KiB
YAML

---
- name: Skip k3s when disabled
ansible.builtin.meta: end_host
when: not (k3s_enabled | default(true) | bool)
- name: Skip k3s when state is absent
ansible.builtin.meta: end_host
when: k3s_state == "absent"
- name: Set k3s service name
ansible.builtin.set_fact:
k3s_service_name: "{{ 'k3s' if k3s_role == 'server' else 'k3s-agent' }}"
- name: Derive server token secret from node-token (server only)
ansible.builtin.set_fact:
k3s_server_token: "{{ k3s_token | regex_replace('^K10[0-9a-f]+::server:', '') }}"
when:
- k3s_role == 'server'
- (k3s_token | default('')) is match('^K10[0-9a-f]+::server:.+')
no_log: true
- name: Set effective k3s data-dir
ansible.builtin.set_fact:
k3s_effective_data_dir: "{{ k3s_data_dir | default('/var/lib/rancher/k3s', true) }}"
- name: Check if k3s service is installed
ansible.builtin.stat:
path: "/etc/systemd/system/{{ k3s_service_name }}.service"
register: k3s_service
- name: Check if k3s binary exists
ansible.builtin.stat:
path: /usr/local/bin/k3s
register: k3s_bin
- name: Determine if k3s needs installation
ansible.builtin.set_fact:
k3s_needs_install: "{{ not k3s_bin.stat.exists or not k3s_service.stat.exists }}"
- name: Gather service facts for health check
ansible.builtin.service_facts:
when: not k3s_needs_install
- name: Check k3s API responsiveness
ansible.builtin.command: k3s kubectl get nodes
register: k3s_api_check
changed_when: false
failed_when: false
when:
- not k3s_needs_install
- (k3s_service_name + ".service") in ansible_facts.services
- ansible_facts.services[k3s_service_name + ".service"].state == 'running'
- name: Restart k3s if API is not responsive (try recovery before reinstall)
ansible.builtin.service:
name: "{{ k3s_service_name }}"
state: restarted
register: k3s_api_restart
failed_when: false
when:
- not k3s_needs_install
- (k3s_service_name + ".service") in ansible_facts.services
- ansible_facts.services[k3s_service_name + ".service"].state == 'running'
- k3s_api_check.rc | default(0) != 0
- name: Re-check k3s API after restart
ansible.builtin.command: k3s kubectl get nodes
register: k3s_api_check_post_restart
changed_when: false
failed_when: false
until: k3s_api_check_post_restart.rc == 0
retries: 6
delay: 10
when:
- not k3s_needs_install
- (k3s_service_name + ".service") in ansible_facts.services
- ansible_facts.services[k3s_service_name + ".service"].state == 'running'
- k3s_api_check.rc | default(0) != 0
- name: Force k3s install if still not responsive after restart
ansible.builtin.set_fact:
k3s_needs_install: true
when:
- not k3s_needs_install
- (k3s_service_name + ".service") in ansible_facts.services
- ansible_facts.services[k3s_service_name + ".service"].state == 'running'
- k3s_api_check.rc | default(0) != 0
- k3s_api_check_post_restart.rc | default(0) != 0
- name: Install k3s dependencies
ansible.builtin.package:
name:
- curl
- iptables
state: present
- name: Address Retropie networking (wifi power save)
ansible.builtin.shell: |
if [ -d /sys/class/net/wlan0 ]; then
/sbin/iwconfig wlan0 power off || /usr/sbin/iw dev wlan0 set power_save off || true
fi
when: inventory_hostname in ['retropie.prole.org', 'pi.prole.org']
changed_when: false
failed_when: false
- name: Configure connman to ignore k8s/cni interfaces
ansible.builtin.shell: |
if [ -f /etc/connman/main.conf ]; then
if ! grep -q "^NetworkInterfaceBlacklist=" /etc/connman/main.conf; then
echo "NetworkInterfaceBlacklist=veth,cni,flannel,docker,virbr" >> /etc/connman/main.conf
echo "RESTART_CONNMAN"
elif ! grep -q "cni" /etc/connman/main.conf; then
# Ensure we have our required prefixes in the blacklist
sed -i 's/^NetworkInterfaceBlacklist=/NetworkInterfaceBlacklist=veth,cni,flannel,docker,virbr,/' /etc/connman/main.conf
# Clean up any potential double commas or wildcards we might have added before
sed -i 's/\*//g' /etc/connman/main.conf
sed -i 's/lo,//g' /etc/connman/main.conf
sed -i 's/,,/,/g' /etc/connman/main.conf
echo "RESTART_CONNMAN"
fi
fi
register: _connman_config
when: inventory_hostname in ['retropie.prole.org', 'pi.prole.org']
changed_when: "'RESTART_CONNMAN' in _connman_config.stdout"
- name: Restart connman to apply blacklist changes
ansible.builtin.service:
name: connman
state: restarted
async: 1
poll: 0
when:
- inventory_hostname in ['retropie.prole.org', 'pi.prole.org']
- _connman_config is changed
ignore_errors: true
- name: Wait for connection to return after network restart
ansible.builtin.wait_for_connection:
delay: 5
timeout: 300
when:
- inventory_hostname in ['retropie.prole.org', 'pi.prole.org']
- _connman_config is changed
- name: Address Retropie routing (wlan0 priority)
ansible.builtin.shell: |
# If eth0 has a default route but we are using wlan0, it might block internet
if [ "{{ k3s_flannel_iface | default('') }}" = "wlan0" ] || [ -d /sys/class/net/wlan0 ]; then
if ip route show default dev eth0 | grep -q default; then
if ip route show default dev wlan0 | grep -q default; then
ip route del default dev eth0 || true
fi
fi
fi
when: inventory_hostname in ['retropie.prole.org', 'pi.prole.org']
changed_when: false
failed_when: false
- name: Preflight - ensure cgroup kernel params are present
ansible.builtin.command: "cat /proc/cmdline"
register: k3s_proc_cmdline
changed_when: false
check_mode: no
- name: Fail if cgroup params are missing (reboot required)
ansible.builtin.fail:
msg: "Missing cgroup params in /proc/cmdline: {{ ['cgroup_memory=1', 'cgroup_enable=memory'] | reject('in', k3s_proc_cmdline.stdout | default('')) | list }}. Reboot required before k3s install."
when:
- not ansible_check_mode
- (['cgroup_memory=1', 'cgroup_enable=memory'] | reject('in', k3s_proc_cmdline.stdout | default('')) | list) | length > 0
- name: Preflight - ensure required mountpoints are mounted
ansible.builtin.shell: |
awk -v mp="{{ item }}" '$5==mp {found=1} END {exit(found?0:1)}' /proc/self/mountinfo
register: k3s_required_mounts_present
changed_when: false
failed_when: k3s_required_mounts_present.rc != 0
check_mode: no
loop: "{{ k3s_required_mounts | default([]) }}"
when: (k3s_required_mounts | default([])) | length > 0
- name: Guardrail - ensure required mountpoints are not on SD
ansible.builtin.shell: |
awk -v mp="{{ item }}" '$5==mp {for(i=1;i<=NF;i++) if($i=="-") {print $(i+2); exit}}' /proc/self/mountinfo
register: k3s_required_mounts_sources
changed_when: false
check_mode: no
loop: "{{ k3s_required_mounts | default([]) }}"
when: (k3s_required_mounts | default([])) | length > 0
- name: Fail if required mountpoint is on SD
ansible.builtin.fail:
msg: "{{ item.item }} is on SD/rootfs ({{ item.stdout | default('') }}). Refusing to proceed."
loop: "{{ k3s_required_mounts_sources.results | default([]) }}"
when:
- not ansible_check_mode
- (item.stdout | default('')) is search("mmcblk0") or (item.stdout | default('')) is search("/dev/mmc")
- name: Read k3s node token for sharing
ansible.builtin.slurp:
src: "{{ k3s_effective_data_dir }}/server/node-token"
register: k3s_node_token_slurp
when:
- k3s_role == "server"
- k3s_cluster_init | bool or k3s_server_url | default('') | length == 0
ignore_errors: true
- name: Set k3s_token_discovered fact
ansible.builtin.set_fact:
k3s_token_discovered: "{{ k3s_node_token_slurp.content | b64decode | trim }}"
when:
- k3s_node_token_slurp is defined
- k3s_node_token_slurp.content is defined
- name: Wait for k3s token from server
ansible.builtin.command: /usr/bin/true
until: hostvars[item].k3s_token_discovered | default('') | length > 0
retries: 60
delay: 10
loop: "{{ groups['k3s_hosts'] }}"
when:
- k3s_role == 'agent'
- hostvars[item].k3s_role | default('agent') == 'server'
- (k3s_token | default('') | length == 0) or (item in ansible_play_hosts)
changed_when: false
- name: Discover k3s token from other hosts
ansible.builtin.set_fact:
k3s_token: "{{ hostvars[item].k3s_token_discovered }}"
loop: "{{ groups['k3s_hosts'] }}"
when:
- hostvars[item].k3s_token_discovered | default('') | length > 0
- hostvars[item].k3s_role | default('agent') == 'server'
- (k3s_token | default('') | length == 0) or (k3s_role == 'agent' and k3s_token != hostvars[item].k3s_token_discovered)
- name: Detect k3s CA hash mismatch on agents (logs)
ansible.builtin.shell: |
journalctl -u {{ k3s_service_name }} --since "1 hour ago" | grep "token CA hash does not match"
register: k3s_ca_mismatch_logs
changed_when: false
failed_when: false
when:
- k3s_role == 'agent'
- k3s_service.stat.exists
- name: Proactively check for CA hash mismatch (files)
block:
- name: Get local agent CA cert hash (DER)
ansible.builtin.shell: |
if [ -f {{ k3s_effective_data_dir }}/agent/server-ca.crt ]; then
openssl x509 -in {{ k3s_effective_data_dir }}/agent/server-ca.crt -outform DER | openssl dgst -sha256 | awk '{print $NF}'
else
echo "MISSING"
fi
register: k3s_agent_ca_hash_cmd
changed_when: false
- name: Extract expected CA hash from token
ansible.builtin.set_fact:
k3s_expected_ca_hash: "{{ (k3s_token | regex_replace('^K10([a-f0-9]+)::.*$', '\\1')) if (k3s_token is search('^K10[a-f0-9]+::')) else '' }}"
- name: Detect mismatch from files
ansible.builtin.set_fact:
k3s_ca_mismatch_files: "{{ k3s_agent_ca_hash_cmd.stdout | trim != k3s_expected_ca_hash }}"
when:
- k3s_agent_ca_hash_cmd.stdout | trim != "MISSING"
- k3s_expected_ca_hash | length > 0
when:
- k3s_role == 'agent'
- k3s_token | default('') | length > 0
- name: Handle k3s CA hash mismatch
block:
- name: Stop k3s-agent for CA repair
ansible.builtin.shell: "systemctl stop {{ k3s_service_name }} || true"
changed_when: false
failed_when: false
- name: Kill any remaining k3s processes
ansible.builtin.shell: "killall -9 k3s || true"
changed_when: false
- name: Wipe stale agent state
ansible.builtin.file:
path: "{{ k3s_effective_data_dir }}/agent"
state: absent
- name: Force k3s re-installation after CA wipe
ansible.builtin.set_fact:
k3s_needs_install: true
when:
- k3s_role == 'agent'
- (k3s_ca_mismatch_logs.rc | default(1) == 0) or (k3s_ca_mismatch_files | default(false) | bool)
- name: Validate join configuration
ansible.builtin.assert:
that:
- k3s_server_url is defined
- k3s_server_url | length > 0
- k3s_token is defined
- k3s_token | length > 0
fail_msg: "k3s_server_url and k3s_token are required for non-init join nodes."
when:
- not k3s_cluster_init | bool
- k3s_role == 'agent' or k3s_server_url | default('') | length > 0
- name: Check for existing k3s server node token
ansible.builtin.stat:
path: "{{ k3s_effective_data_dir }}/server/node-token"
register: k3s_server_node_token
when:
- k3s_guard_token_drift | bool
- k3s_role == "server"
- k3s_cluster_init | bool
- name: Read existing k3s server node token
ansible.builtin.slurp:
src: "{{ k3s_effective_data_dir }}/server/node-token"
register: k3s_server_node_token_raw
when:
- k3s_guard_token_drift | bool
- k3s_role == "server"
- k3s_cluster_init | bool
- k3s_server_node_token.stat.exists
- name: Fail when vault k3s token is missing for existing cluster
ansible.builtin.fail:
msg: >-
vault_k3s_token is empty but {{ k3s_effective_data_dir }}/server/node-token exists.
Refusing to overwrite k3s config. Update vault_k3s.yml or run the k3s sync playbook.
when:
- k3s_guard_token_drift | bool
- k3s_role == "server"
- k3s_cluster_init | bool
- k3s_server_node_token.stat.exists
- (k3s_token | default('') | trim) | length == 0
- name: Fail on k3s token drift from live cluster
ansible.builtin.fail:
msg: >-
vault_k3s_token does not match the live k3s server node-token.
Refusing to overwrite k3s config. Run infrastructure/playbooks/k3s_sync.yml
or update vault_k3s.yml to match the live token.
when:
- k3s_guard_token_drift | bool
- k3s_role == "server"
- k3s_cluster_init | bool
- k3s_server_node_token.stat.exists
- (k3s_token | default('') | trim) | length > 0
- (k3s_server_node_token_raw.content | b64decode | trim) != (k3s_token | default('') | trim)
- name: Ensure k3s config directory exists
ansible.builtin.file:
path: /etc/rancher/k3s
state: directory
mode: "0755"
- name: Load registry defaults from prole.cfg
ansible.builtin.set_fact:
k3s_prole_k3s_server_cfg: "{{ lookup('ansible.builtin.ini', 'PROLE_K3S_SERVER section=Global file=' + playbook_dir + '/../../conf/prole.cfg', default='') }}"
k3s_prole_service_namespace_cfg: "{{ lookup('ansible.builtin.ini', 'SERVICE_NAMESPACE section=Global file=' + playbook_dir + '/../../conf/prole.cfg', default='') }}"
when: k3s_state == "present" and k3s_registry_config_enabled | bool
- name: Resolve k3s registry host
ansible.builtin.set_fact:
k3s_registry_host_resolved: >-
{{ (k3s_registry_host | default('')) if (k3s_registry_host | default('') | length > 0)
else (k3s_prole_k3s_server_cfg | default('')) if (k3s_prole_k3s_server_cfg | default('') | length > 0)
else (k3s_server_url | default('')) if (k3s_server_url | default('') | length > 0)
else inventory_hostname }}
when: k3s_state == "present" and k3s_registry_config_enabled | bool
- name: Normalize k3s registry host
ansible.builtin.set_fact:
k3s_registry_host_resolved: "{{ k3s_registry_host_resolved | regex_replace('^https?://', '') | regex_replace('/.*$', '') | regex_replace(':.*$', '') }}"
when: k3s_state == "present" and k3s_registry_config_enabled | bool
- name: Resolve k3s registry namespace
ansible.builtin.set_fact:
k3s_registry_namespace_resolved: >-
{{ (k3s_registry_namespace | default('')) if (k3s_registry_namespace | default('') | length > 0 and k3s_registry_namespace != '${SERVICE_NAMESPACE}')
else (k3s_prole_service_namespace_cfg | default('')) if (k3s_prole_service_namespace_cfg | default('') | length > 0 and k3s_prole_service_namespace_cfg != '${SERVICE_NAMESPACE}')
else 'knoe-system' }}
when: k3s_state == "present" and k3s_registry_config_enabled | bool
- name: Validate k3s registry host
ansible.builtin.assert:
that:
- k3s_registry_host_resolved | length > 0
fail_msg: "k3s registry host could not be resolved (set k3s_registry_host or PROLE_K3S_SERVER)."
when: k3s_state == "present" and k3s_registry_config_enabled | bool
- name: Render k3s registries config
ansible.builtin.template:
src: registries.yaml.j2
dest: "{{ k3s_registry_config_path }}"
mode: "0644"
notify: Restart k3s
when: k3s_state == "present" and k3s_registry_config_enabled | bool
- name: Ensure kubeconfig group exists
ansible.builtin.group:
name: "{{ k3s_kubeconfig_group }}"
state: present
local: "{{ (ansible_facts['os_family'] | default('')) in ['Darwin', 'FreeBSD', 'OpenBSD', 'NetBSD'] }}"
- name: Ensure kubeconfig users are in group (server)
ansible.builtin.user:
name: "{{ item }}"
groups: "{{ k3s_kubeconfig_group }}"
append: true
loop: "{{ k3s_kubeconfig_users }}"
when: k3s_role == "server" and (k3s_kubeconfig_users | length) > 0
- name: Ensure rancher directories are group-accessible
ansible.builtin.file:
path: "{{ item }}"
state: directory
owner: root
group: "{{ k3s_kubeconfig_group }}"
mode: "0750"
loop:
- /etc/rancher
- /etc/rancher/k3s
- /etc/rancher/k3s/config.yaml.d
- "{{ k3s_effective_data_dir | dirname }}"
when: k3s_state != "absent"
- name: Detect active interface for k3s_node_ip (Retropie/Pi only)
ansible.builtin.shell: |
# Find interface that has the configured k3s_node_ip
ip -4 addr show | grep -B2 "{{ k3s_node_ip }}" | grep -oE "^[0-9]+: [^:]+" | head -n 1 | awk '{print $2}'
register: _detected_k3s_iface
when:
- inventory_hostname in ['retropie.prole.org', 'pi.prole.org']
- k3s_node_ip is defined and k3s_node_ip | length > 0
changed_when: false
- name: Update k3s_flannel_iface based on detection (Retropie/Pi only)
ansible.builtin.set_fact:
k3s_flannel_iface: "{{ _detected_k3s_iface.stdout | trim }}"
when:
- inventory_hostname in ['retropie.prole.org', 'pi.prole.org']
- _detected_k3s_iface.stdout | default('') | length > 0
- _detected_k3s_iface.stdout | trim != k3s_flannel_iface
- name: Validate k3s config args
ansible.builtin.import_tasks: validate_args.yml
when: k3s_state != "absent"
- name: Render k3s config
ansible.builtin.template:
src: config.yaml.j2
dest: /etc/rancher/k3s/config.yaml
owner: root
group: "{{ k3s_kubeconfig_group }}"
mode: "0640"
notify: Restart k3s
- name: Build k3s install environment
ansible.builtin.set_fact:
k3s_install_env: >-
{{ {'INSTALL_K3S_EXEC': k3s_role}
| combine((k3s_version | length > 0) | ternary({'INSTALL_K3S_VERSION': k3s_version}, {})) }}
- name: Ensure k3s service is stopped before install to prevent hangs
ansible.builtin.service:
name: "{{ k3s_service_name }}"
state: stopped
when:
- k3s_needs_install
- k3s_service.stat.exists
failed_when: false
- name: Install k3s
ansible.builtin.shell: |
curl -sfL https://get.k3s.io | sh -
environment:
"{{ k3s_install_env }}"
when: k3s_needs_install
- name: Re-check if k3s service is installed (post-install)
ansible.builtin.stat:
path: "/etc/systemd/system/{{ k3s_service_name }}.service"
register: k3s_service
- name: Re-check if k3s binary exists (post-install)
ansible.builtin.stat:
path: /usr/local/bin/k3s
register: k3s_bin
- name: Ensure k3s service state
ansible.builtin.service:
name: "{{ k3s_service_name }}"
state: "{{ 'started' if k3s_state == 'present' else 'stopped' }}"
enabled: "{{ k3s_state == 'present' }}"
when: k3s_service.stat.exists or k3s_bin.stat.exists
- name: Flush k3s restart handler before Kubernetes operations
ansible.builtin.meta: flush_handlers
when: k3s_state == "present"
- name: Apply node taints
ansible.builtin.command: "k3s kubectl taint node {{ inventory_hostname }} {{ item }} --overwrite"
delegate_to: "{{ (groups['k3s_hosts'] | map('extract', hostvars) | selectattr('k3s_role', 'defined') | selectattr('k3s_role', 'equalto', 'server') | map(attribute='inventory_hostname') | first) | default(groups['k3s_hosts'] | first) }}"
loop: "{{ k3s_node_taints }}"
when:
- k3s_state == "present"
- k3s_node_taints | length > 0
register: _taint_result
changed_when: "'modified' in _taint_result.stdout or 'tainted' in _taint_result.stdout"
failed_when: false # Avoid failing if node is not yet ready
- name: Wait for k3s API readiness
ansible.builtin.command: k3s kubectl get --raw=/readyz
register: k3s_readyz
until:
- k3s_readyz.rc == 0
- (k3s_readyz.stdout | default('') | trim) in ['ok', 'OK']
retries: 60
delay: 5
changed_when: false
when: k3s_state == "present" and k3s_role == "server"
- name: Wait for local node to report Ready
ansible.builtin.shell: |
k3s kubectl get node {{ inventory_hostname }} -o jsonpath='{.status.conditions[?(@.type=="Ready")].status}' 2>/dev/null || true
register: k3s_node_ready
until: (k3s_node_ready.stdout | default('') | trim) == 'True'
retries: 60
delay: 5
changed_when: false
failed_when: false
when: k3s_state == "present" and k3s_role == "server"
- name: Read k3s node token after install for sharing
ansible.builtin.slurp:
src: "{{ k3s_effective_data_dir }}/server/node-token"
register: k3s_node_token_slurp_post
when:
- k3s_role == "server"
- k3s_cluster_init | bool or k3s_server_url | default('') | length == 0
- k3s_token_discovered is not defined
ignore_errors: true
- name: Set k3s_token_discovered fact (post-install)
ansible.builtin.set_fact:
k3s_token_discovered: "{{ k3s_node_token_slurp_post.content | b64decode | trim }}"
when:
- k3s_node_token_slurp_post is defined
- k3s_node_token_slurp_post.content is defined
- name: Sync k3s token to local vault
ansible.builtin.include_tasks: sync_vault.yml
when:
- k3s_role == "server"
- k3s_cluster_init | bool or k3s_server_url | default('') | length == 0
- k3s_token_discovered is defined
- k3s_token_discovered | length > 0
- name: Pre-stage k3s images (deliberate arch)
ansible.builtin.include_tasks: prestage_images.yml
when:
- k3s_state == "present"
- k3s_prestage_images | bool
tags: [images]