runner

1# D&D session agent stack.
2#
3# One compose project so the whole system has a single on/off switch:
4#   dnd_enabled: false  ->  every part of it goes away.
5#
6# Three services with deliberately different lifecycles:
7#
8#   dnd-stt   always up. A generic OpenAI-compatible speech endpoint that
9#             anything on the runner can use, not just this stack.
10#   dnd-agent started per session and stopped afterwards. It only makes sense
11#             while people are in the room, and leaving it idling would keep a
12#             participant in the conference all week.
13#   dnd-post  runs to completion and exits. Under a compose profile so a plain
14#             `up` never starts it.
15
16services:
17  dnd-stt:
18    container_name: dnd-stt
19    image: "{{ container_images.dnd_stt }}"
20    restart: unless-stopped
21
22    ports:
23      - "{{ dnd_stt_host_port }}:8000"
24
25    env_file:
26      - stt.env
27
28    volumes:
29      - /etc/localtime:/etc/localtime:ro
30      # Whisper weights are several hundred MB and fetched on first use, so
31      # this must persist or every restart re-downloads them.
32      - {{ dnd_stt_model_dir }}:/home/ubuntu/.cache/huggingface
33
34{% if gpu_enabled %}
35    runtime: nvidia
36    deploy:
37      resources:
38        reservations:
39          devices:
40            - driver: nvidia
41              count: 1
42              capabilities: [gpu]
43{% endif %}
44
45    healthcheck:
46      test: ["CMD", "curl", "-fsS", "http://localhost:8000/health"]
47      interval: 30s
48      timeout: 10s
49      start_period: {{ dnd_stt_start_period }}
50      retries: 3
51
52    networks:
53      {{ runner_docker_network }}:
54        aliases:
55          - dnd-stt
56
57  dnd-agent:
58    container_name: dnd-agent
59    build:
60      context: {{ dnd_config_dir }}/src
61      dockerfile: Dockerfile.agent
62      args:
63        UID: "{{ dnd_uid }}"
64        GID: "{{ dnd_gid }}"
65    image: dnd-agent:{{ dnd_agent_version }}
66
67    # Never restarted automatically. A session ends when people leave, and a
68    # restarting agent would silently rejoin an empty room and record silence.
69    restart: "no"
70    profiles: ["session"]
71
72    env_file:
73      - agent.env
74
75    depends_on:
76      dnd-stt:
77        condition: service_healthy
78
79    volumes:
80      - /etc/localtime:/etc/localtime:ro
81      - {{ dnd_media_dir }}:/media/dnd
82
83    networks:
84      {{ runner_docker_network }}:
85        aliases:
86          - dnd-agent
87
88  dnd-post:
89    container_name: dnd-post
90    build:
91      context: {{ dnd_config_dir }}/src
92      dockerfile: Dockerfile.post
93      args:
94        UID: "{{ dnd_uid }}"
95        GID: "{{ dnd_gid }}"
96    image: dnd-post:{{ dnd_agent_version }}
97
98    restart: "no"
99    # Only started explicitly, so `docker compose up -d` brings up the STT
100    # service alone and never kicks off an hour of summarisation.
101    profiles: ["post"]
102
103    env_file:
104      - post.env
105
106    volumes:
107      - /etc/localtime:/etc/localtime:ro
108      - {{ dnd_media_dir }}:/media/dnd
109      # pyannote weights, kept out of the image so a rebuild does not refetch
110      # them and so the gated download happens exactly once.
111      - {{ dnd_model_dir }}:/models
112
113{% if gpu_enabled %}
114    runtime: nvidia
115    deploy:
116      resources:
117        reservations:
118          devices:
119            - driver: nvidia
120              count: 1
121              capabilities: [gpu]
122{% endif %}
123
124    networks:
125      {{ runner_docker_network }}:
126        aliases:
127          - dnd-post
128
129networks:
130  {{ runner_docker_network }}:
131    external: true
132