feat: add health-check to all nodes

roles/vega_core/defaults/main.yaml

@@ -57,3 +57,7 @@ vega_core_block_explorer_postgresql_database: "tendermint"
 vega_core_block_explorer_postgresql_password: "tendermint"
 
 vega_core_data_node_settings: {}
+
+vega_core_with_health_check: true
+vega_core_health_check_repository: "nebula-dex/vega-health-check"
+vega_core_health_check_version: "v0.1.0"

roles/vega_core/handlers/main.yaml

@@ -3,7 +3,6 @@
   ansible.builtin.service:
     state: "{{- 'restarted' if vega_core_run_network else 'stopped' -}}"
     daemon_reload: true
-
     name: vegavisor
   listen: "Restart vegavisor"
   when: not ansible_check_mode
@@ -15,3 +14,11 @@
     name: blockexplorer
   listen: "Restart blockexplorer"
   when: not ansible_check_mode
+
+- name: Restart vega-health-check
+  ansible.builtin.service:
+    state: "{{- 'restarted' if vega_core_run_network else 'stopped' -}}"
+    daemon_reload: true
+    name: vega-health-check
+  listen: "Restart vega-health-check"
+  when: not ansible_check_mode

roles/vega_core/tasks/health-check.yaml

@@ -0,0 +1,16 @@
+---
+- name: Download health-check binary
+  ansible.builtin.get_url:
+    force: true
+    url: "https://github.com/{{- vega_core_health_check_repository -}}/releases/download/{{- vega_core_health_check_version -}}/vega-health-check-linux-amd64"
+    dest: /usr/local/bin/vega-health-check
+    mode: '0755'
+
+- name: Install health-check systemd
+  ansible.builtin.template:
+    src: "lib/systemd/system/vega-health-check.service.j2"
+    dest: "/lib/systemd/system/vega-health-check.service"
+    owner: "root"
+    group: "root"
+    mode: "0644"
+  notify: "Restart vega-health-check"

roles/vega_core/tasks/main.yaml

@@ -26,4 +26,8 @@
   when: vega_core_with_unsafe_reset_all
 
 - name: Restart network
-  ansible.builtin.import_tasks: restart_network.yaml
+  ansible.builtin.import_tasks: restart-network.yaml
+
+- name: Health-check setup
+  ansible.builtin.import_tasks: health-check.yaml
+  when: vega_core_with_health_check

roles/vega_core/templates/lib/systemd/system/vega-health-check.service.j2

@@ -0,0 +1,23 @@
+[Unit]
+Description=healthcheck
+Documentation=https://github.com/vegaprotocol/ansible
+After=network.target network-online.target
+Requires=network-online.target
+
+[Service]
+User=vega
+Group=vega
+{% if vega_core_with_block_explorer | default(false) %}
+ExecStart=/usr/local/bin/vega-health-check blockexplorer --blockexplorer-api-url "http://localhost:1515" --core-url "http://localhost:3003" --http-port 8080
+{% elif vega_core_with_data_node | default(false) %}
+ExecStart=/usr/local/bin/vega-health-check data-node --api-url "http://localhost:3008" --core-url "http://localhost:3003" --http-port 8080
+{% else %}
+ExecStart=/usr/local/bin/vega-health-check vega --core-url "http://localhost:3003" --http-port 8080
+{% endif %}
+TimeoutStopSec=10s
+ProtectSystem=full
+AmbientCapabilities=CAP_NET_BIND_SERVICE
+CPUQuota=10%
+
+[Install]
+WantedBy=multi-user.target