runner

1---
2# Runner Services - Unified LLM Stack Deployment
3# Single deployment with Ollama, OpenWebUI, and LiteLLM services
4
5- name: Check if LLM stack is enabled
6  fail:
7    msg: "LLM stack is disabled. Set llm_stack_enabled: true to enable."
8  when: not llm_stack_enabled
9  tags: always
10
11- name: Create LLM stack configuration directory structure
12  file:
13    path: "{{ item }}"
14    state: directory
15    owner: "{{ runner_user }}"
16    group: "{{ runner_group }}"
17    mode: '0775'
18  loop:
19    - "{{ llm_stack_config_dir }}"
20    - "{{ llm_stack_ollama_data_dir }}"
21    - "{{ llm_stack_openwebui_data_dir }}"
22    - "{{ llm_stack_litellm_data_dir }}"
23
24- name: Set group sticky bit on LLM stack directories for permission inheritance
25  file:
26    path: "{{ llm_stack_config_dir }}"
27    state: directory
28    mode: "g+s"
29
30- name: Create unified LLM stack Docker Compose file
31  template:
32    src: llm-stack-compose.yml.j2
33    dest: "{{ llm_stack_config_dir }}/docker-compose.yml"
34    owner: "{{ runner_user }}"
35    group: "{{ runner_group }}"
36    mode: '0664'
37  notify: restart llm-stack
38
39- name: Create unified LLM stack environment file
40  template:
41    src: llm-stack.env.j2
42    dest: "{{ llm_stack_config_dir }}/.env"
43    owner: "{{ runner_user }}"
44    group: "{{ runner_group }}"
45    mode: '0664'
46  notify: restart llm-stack
47
48- name: Create LiteLLM environment file
49  template:
50    src: llm-stack-litellm.env.j2
51    dest: "{{ llm_stack_config_dir }}/litellm.env"
52    owner: "{{ runner_user }}"
53    group: "{{ runner_group }}"
54    mode: '0664'
55  notify: restart llm-stack
56
57- name: Create LiteLLM proxy configuration
58  template:
59    src: llm-stack-litellm-config.yml.j2
60    dest: "{{ llm_stack_config_dir }}/litellm-config.yaml"
61    owner: "{{ runner_user }}"
62    group: "{{ runner_group }}"
63    mode: '0664'
64  notify: restart llm-stack
65
66- name: Pull the pinned LLM stack images
67  community.docker.docker_image:
68    name: "{{ item }}"
69    source: pull
70    force_source: true
71  loop:
72    - "{{ container_images.ollama }}"
73    - "{{ container_images.litellm }}"
74    - "{{ container_images.openwebui }}"
75
76- name: Start unified LLM stack service
77  community.docker.docker_compose_v2:
78    project_src: "{{ llm_stack_config_dir }}"
79    state: present
80  register: llm_stack_start_result
81  check_mode: no
82
83- name: Wait for Ollama to be healthy
84  uri:
85    url: "http://localhost:{{ llm_stack_ollama_port }}/api/tags"
86    method: GET
87    status_code: 200
88  register: ollama_health
89  until: ollama_health.status == 200
90  retries: 30
91  delay: 10
92  when: llm_stack_start_result is changed
93  check_mode: no
94
95- name: Wait for OpenWebUI to be healthy
96  uri:
97    url: "http://localhost:{{ llm_stack_openwebui_port }}/api/health"
98    method: GET
99    status_code: 200
100    timeout: 30
101  register: openwebui_health
102  until: openwebui_health.status == 200
103  retries: 60
104  delay: 15
105  when: llm_stack_start_result is changed
106  check_mode: no
107
108- name: Wait for LiteLLM PostgreSQL to be healthy
109  wait_for:
110    host: localhost
111    port: "{{ llm_stack_litellm_port }}"
112    timeout: 30
113  when: llm_stack_start_result is changed
114  check_mode: no
115
116- name: Wait for LiteLLM to be healthy
117  uri:
118    url: "http://localhost:{{ llm_stack_litellm_port }}/"
119    method: GET
120    status_code: 200
121  register: litellm_health
122  until: litellm_health.status == 200
123  retries: 10
124  delay: 10
125  when: llm_stack_start_result is changed
126  check_mode: no
127
128- name: Pull declared Ollama models
129  when: llm_stack_pull_models
130  tags: [llm-models]
131  block:
132    - name: Query models already present in Ollama
133      uri:
134        url: "http://localhost:{{ llm_stack_ollama_port }}/api/tags"
135        method: GET
136        status_code: 200
137      register: ollama_present_models
138      check_mode: no
139
140    - name: Determine which declared models are missing
141      set_fact:
142        llm_stack_missing_models: >-
143          {{ llm_stack_models | map(attribute='name') | difference(
144               ollama_present_models.json.models | default([]) | map(attribute='name')
145             ) | list }}
146
147    - name: Report model pull plan
148      debug:
149        msg: >-
150          {{ (llm_stack_models | length) - (llm_stack_missing_models | length) }} of
151          {{ llm_stack_models | length }} declared models already present.
152          {{ 'Pulling: ' ~ (llm_stack_missing_models | join(', '))
153             if llm_stack_missing_models | length > 0
154             else 'Nothing to pull.' }}
155
156    # Pulls are serialised deliberately: concurrent pulls of multi-gigabyte models
157    # saturate both disk and the Ollama API, and a failure part-way is harder to
158    # attribute. Timeout is generous because a cold pull of a 14B model is slow.
159    - name: Pull missing Ollama models
160      uri:
161        url: "http://localhost:{{ llm_stack_ollama_port }}/api/pull"
162        method: POST
163        body_format: json
164        body:
165          model: "{{ item }}"
166          stream: false
167        status_code: 200
168        timeout: "{{ llm_stack_model_pull_timeout }}"
169      loop: "{{ llm_stack_missing_models }}"
170      loop_control:
171        label: "{{ item }}"
172      register: ollama_pull_result
173      # A model name that does not exist upstream returns 200 with an error body,
174      # so the status code alone is not a sufficient success signal.
175      failed_when: >-
176        ollama_pull_result.json.status | default('') != 'success'
177      check_mode: no
178
179- name: Display unified LLM stack deployment summary
180  debug:
181    msg: |
182      Unified LLM Stack Deployment:
183      - Status: {{ 'Started' if llm_stack_start_result is changed else 'Already running' }}
184      
185      Ollama (Local LLM Server):
186      - Web UI: http://{{ ansible_default_ipv4.address }}:{{ llm_stack_ollama_port }}
187      - Configuration: {{ llm_stack_config_dir }}
188      - Data Storage: {{ llm_stack_ollama_data_dir }}
189      - GPU Support: {{ 'Enabled' if gpu_enabled else 'Disabled' }}
190      - Default Model: {{ llm_stack_ollama_default_model }}
191
192      OpenWebUI (Web Interface):
193      - Web UI: http://{{ ansible_default_ipv4.address }}:{{ llm_stack_openwebui_port }}
194      - Connected to Ollama: ollama:11434
195      - Name: {{ llm_stack_openwebui_name }}
196
197      LiteLLM (Unified Proxy):
198      - API Endpoint: http://{{ ansible_default_ipv4.address }}:{{ llm_stack_litellm_port }}
199      - Connected to Ollama: ollama:11434
200      - Available Models:
201      {% for model in llm_stack_models %}
202        - {{ model.name }} ({{ model.role }}): {{ model.description }}
203      {% endfor %}
204      
205      Management Commands:
206      - cd {{ llm_stack_config_dir }} && docker compose logs -f
207      - cd {{ llm_stack_config_dir }} && docker compose restart
208      - cd {{ llm_stack_config_dir }} && docker compose ps