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: Start unified LLM stack service
67  community.docker.docker_compose_v2:
68    project_src: "{{ llm_stack_config_dir }}"
69    state: present
70  register: llm_stack_start_result
71  check_mode: no
72
73- name: Wait for Ollama to be healthy
74  uri:
75    url: "http://localhost:{{ llm_stack_ollama_port }}/api/tags"
76    method: GET
77    status_code: 200
78  register: ollama_health
79  until: ollama_health.status == 200
80  retries: 30
81  delay: 10
82  when: llm_stack_start_result is changed
83  check_mode: no
84
85- name: Wait for OpenWebUI to be healthy
86  uri:
87    url: "http://localhost:{{ llm_stack_openwebui_port }}/api/health"
88    method: GET
89    status_code: 200
90    timeout: 30
91  register: openwebui_health
92  until: openwebui_health.status == 200
93  retries: 60
94  delay: 15
95  when: llm_stack_start_result is changed
96  check_mode: no
97
98- name: Wait for LiteLLM PostgreSQL to be healthy
99  wait_for:
100    host: localhost
101    port: "{{ llm_stack_litellm_port }}"
102    timeout: 30
103  when: llm_stack_start_result is changed
104  check_mode: no
105
106- name: Wait for LiteLLM to be healthy
107  uri:
108    url: "http://localhost:{{ llm_stack_litellm_port }}/"
109    method: GET
110    status_code: 200
111  register: litellm_health
112  until: litellm_health.status == 200
113  retries: 10
114  delay: 10
115  when: llm_stack_start_result is changed
116  check_mode: no
117
118- name: Pull declared Ollama models
119  when: llm_stack_pull_models
120  tags: [llm-models]
121  block:
122    - name: Query models already present in Ollama
123      uri:
124        url: "http://localhost:{{ llm_stack_ollama_port }}/api/tags"
125        method: GET
126        status_code: 200
127      register: ollama_present_models
128      check_mode: no
129
130    - name: Determine which declared models are missing
131      set_fact:
132        llm_stack_missing_models: >-
133          {{ llm_stack_models | map(attribute='name') | difference(
134               ollama_present_models.json.models | default([]) | map(attribute='name')
135             ) | list }}
136
137    - name: Report model pull plan
138      debug:
139        msg: >-
140          {{ (llm_stack_models | length) - (llm_stack_missing_models | length) }} of
141          {{ llm_stack_models | length }} declared models already present.
142          {{ 'Pulling: ' ~ (llm_stack_missing_models | join(', '))
143             if llm_stack_missing_models | length > 0
144             else 'Nothing to pull.' }}
145
146    # Pulls are serialised deliberately: concurrent pulls of multi-gigabyte models
147    # saturate both disk and the Ollama API, and a failure part-way is harder to
148    # attribute. Timeout is generous because a cold pull of a 14B model is slow.
149    - name: Pull missing Ollama models
150      uri:
151        url: "http://localhost:{{ llm_stack_ollama_port }}/api/pull"
152        method: POST
153        body_format: json
154        body:
155          model: "{{ item }}"
156          stream: false
157        status_code: 200
158        timeout: "{{ llm_stack_model_pull_timeout }}"
159      loop: "{{ llm_stack_missing_models }}"
160      loop_control:
161        label: "{{ item }}"
162      register: ollama_pull_result
163      # A model name that does not exist upstream returns 200 with an error body,
164      # so the status code alone is not a sufficient success signal.
165      failed_when: >-
166        ollama_pull_result.json.status | default('') != 'success'
167      check_mode: no
168
169- name: Display unified LLM stack deployment summary
170  debug:
171    msg: |
172      Unified LLM Stack Deployment:
173      - Status: {{ 'Started' if llm_stack_start_result is changed else 'Already running' }}
174      
175      Ollama (Local LLM Server):
176      - Web UI: http://{{ ansible_default_ipv4.address }}:{{ llm_stack_ollama_port }}
177      - Configuration: {{ llm_stack_config_dir }}
178      - Data Storage: {{ llm_stack_ollama_data_dir }}
179      - GPU Support: {{ 'Enabled' if gpu_enabled else 'Disabled' }}
180      - Default Model: {{ llm_stack_ollama_default_model }}
181
182      OpenWebUI (Web Interface):
183      - Web UI: http://{{ ansible_default_ipv4.address }}:{{ llm_stack_openwebui_port }}
184      - Connected to Ollama: ollama:11434
185      - Name: {{ llm_stack_openwebui_name }}
186
187      LiteLLM (Unified Proxy):
188      - API Endpoint: http://{{ ansible_default_ipv4.address }}:{{ llm_stack_litellm_port }}
189      - Connected to Ollama: ollama:11434
190      - Available Models:
191      {% for model in llm_stack_models %}
192        - {{ model.name }} ({{ model.role }}): {{ model.description }}
193      {% endfor %}
194      
195      Management Commands:
196      - cd {{ llm_stack_config_dir }} && docker compose logs -f
197      - cd {{ llm_stack_config_dir }} && docker compose restart
198      - cd {{ llm_stack_config_dir }} && docker compose ps