Add supervisord to support a single docker run deployment (#29)

* Add suppervisord in dockerfile

* Create supervisord

* Update README.md

* Update README.md

Dockerfile

@@ -32,4 +32,17 @@ RUN --mount=type=cache,target=/root/.cache pip install -i $PYPI_INDEX_URL --extr
 ENV LD_LIBRARY_PATH "$LD_LIBRARY_PATH:/opt/conda/lib"
 RUN ln -s /opt/conda/lib/libcudart.so.11.7.99 /opt/conda/lib/libcudart.so
 
+# vector
+RUN <<EOF
+curl --proto '=https' --tlsv1.2 -sSf https://sh.vector.dev | bash -s -- -y
+mkdir -p /var/lib/vector
+EOF
+ENV PATH "$PATH:/root/.vector/bin"
+COPY deployment/config/vector.toml /etc/vector/vector.toml
+
+# Supervisord
+RUN --mount=type=cache,target=/root/.cache pip install -i $PYPI_INDEX_URL --extra-index-url https://pypi.org/simple supervisor
+COPY deployment/scripts/supervisord.sh /usr/bin
+
 COPY tabby ./tabby
+CMD ["supervisord.sh"]

README.md

@@ -26,11 +26,16 @@ An opensource / on-prem alternative to GitHub Copilot.
 
 ### Docker
 
-The easiest way of getting started is using the `deployment/docker-compose.yml`:
+The easiest way of getting started is using the official docker image:
 ```bash
-docker-compose up
+docker run \
+  -it --rm \
+  -v ./data:/data \
+  -v ./data/hf_cache:/root/.cache/huggingface \
+  -p 5000:5000 \
+  -p 8501:8501 \
+  -e MODEL_NAME=TabbyML/J-350M tabbyml/tabby
 ```
-Note: To use GPUs, you need to install the [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html). We also recommend using NVIDIA drivers with CUDA version 11.8 or higher.
 
 You can then query the server using `/v1/completions` endpoint:
 ```bash
@@ -39,6 +44,12 @@ curl -X POST http://localhost:5000/v1/completions -H 'Content-Type: application/
 }'
 ```
 
+To use the GPU backend (triton) for a faster inference experience, use `deployment/docker-compose.yml`:
+```bash
+docker-compose up
+```
+Note: To use GPUs, you need to install the [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html). We also recommend using NVIDIA drivers with CUDA version 11.8 or higher.
+
 We also provides an interactive playground in admin panel [localhost:8501](http://localhost:8501)
 
 ![image](https://user-images.githubusercontent.com/388154/227792390-ec19e9b9-ebbb-4a94-99ca-8a142ffb5e46.png)

deployment/.env

@@ -1,6 +1,5 @@
 MODEL_NAME=TabbyML/J-350M
 
 # Volumes
+DATA_VOLUME="../data:/data"
 HF_VOLUME="../data/hf_cache:/root/.cache/huggingface"
-
-LOGS_VOLUME="../data/logs:/logs"

deployment/config/vector.toml

@@ -4,7 +4,7 @@ address = "0.0.0.0:8686"
 
 [sources.tabby_server_logs]
 type = "file"
-include = ["/logs/tabby-server/events.*.log"]
+include = ["/data/logs/tabby-server/events.*.log"]
 
 [transforms.process_tabby_server_logs]
 type = "remap"
@@ -16,4 +16,4 @@ type = "file"
 inputs = [ "process_tabby_server_logs" ]
 encoding = { codec = "json" }
 framing = { method = "newline_delimited" }
-path = "/logs/tabby-server/events.%Y-%m-%d.json"
+path = "/data/logs/tabby-server/events.%Y-%m-%d.json"

deployment/docker-compose.yml

@@ -1,40 +1,26 @@
 version: '3.3'
 
 services:
-  init:
+  tabby:
     image: tabbyml/tabby
-    container_name: tabby-init
-    command: dagu start --params=MODEL_NAME=${MODEL_NAME} ./tabby/tasks/init.yaml
-    volumes:
-      - ${HF_VOLUME}
-
-  server:
-    image: tabbyml/tabby
-    container_name: tabby-server
-    command: uvicorn tabby.server:app --host 0.0.0.0 --port 5000
+    container_name: tabby
     environment:
       MODEL_NAME: ${MODEL_NAME}
       MODEL_BACKEND: triton
-      EVENTS_LOG_DIR: /logs/tabby-server
+      EVENTS_LOG_DIR: /data/logs/tabby-server
+      DAGU_DAGS: /app/tabby/tasks
     ports:
       - "5000:5000"
-    volumes:
-      - ${HF_VOLUME}
-      - ${LOGS_VOLUME}
-    depends_on:
-      init:
-        condition: service_completed_successfully
-      triton:
-        condition: service_healthy
-
-  admin:
-    image: tabbyml/tabby
-    container_name: tabby-admin
-    command: streamlit run tabby/admin/Home.py
-    ports:
+      - "8080:8080"
       - "8501:8501"
     volumes:
-      - ${LOGS_VOLUME}
+      - ${DATA_VOLUME}
+      - ${HF_VOLUME}
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:5000"]
+      interval: 2s
+      timeout: 2s
+      start_period: 120s
 
   triton:
     image: tabbyml/fastertransformer_backend
@@ -54,29 +40,5 @@ services:
     environment:
       MODEL_NAME: ${MODEL_NAME}
     depends_on:
-      init:
-        condition: service_completed_successfully
-    healthcheck:
-      test: ["CMD", "curl", "-f", "http://localhost:8002/metrics"]
-      interval: 2s
-      timeout: 2s
-      start_period: 120s
-
-  vector:
-    image: timberio/vector:0.28.1-alpine
-    container_name: tabby-vector
-    volumes:
-      - ./config/vector.toml:/etc/vector/vector.toml:ro
-      - ${LOGS_VOLUME}
-
-  dagu:
-    image: tabbyml/tabby
-    container_name: tabby-dagu
-    command: dagu scheduler
-    volumes:
-      - ${LOGS_VOLUME}
-    environment:
-      DAGU_DAGS: tabby/tasks
-    depends_on:
-      init:
-        condition: service_completed_successfully
+      tabby:
+        condition: service_healthy

deployment/scripts/supervisord.sh

@@ -0,0 +1,46 @@
+#!/bin/bash
+set -e
+
+# Shared environment variables
+export LOGS_DIR="${LOGS_DIR:-/data/logs}"
+export DB_FILE="${DB_FILE:-/data/logs/duckdb/duck.db}"
+
+# server
+export MODEL_NAME="${MODEL_NAME:-TabbyML/J-350M}"
+export MODEL_BACKEND="${MODEL_BACKEND:-python}"
+export EVENTS_LOG_DIR="${LOGS_DIR}/tabby-server"
+
+# dagu
+export DAGU_DAGS="tabby/tasks"
+
+init() {
+python -m tabby.tools.download_models --repo_id=$MODEL_NAME
+}
+
+
+supervisor() {
+supervisord -n -c <(cat <<EOF
+[supervisord]
+logfile = /var/log/supervisord.log
+loglevel = debug
+
+[program:server]
+command=uvicorn tabby.server:app --host 0.0.0.0 --port 5000
+
+[program:admin]
+command=streamlit run tabby/admin/Home.py --server.port 8501
+
+[program:vector]
+command=vector
+
+[program:dagu_scheduler]
+command=dagu scheduler
+
+[program:dagu_server]
+command=dagu server --host 0.0.0.0 --port 8080
+EOF
+)
+}
+
+init
+supervisor

development/docker-compose.dev.yml

@@ -1,30 +1,16 @@
 version: '3.3'
 
 services:
-  init:
+  tabby:
     build:
       context: ..
       args:
         PYPI_INDEX_URL: https://pypi.tuna.tsinghua.edu.cn/simple
-    volumes:
-      - ../:/app
-
-  server:
-    command: uvicorn tabby.server:app --host 0.0.0.0 --port 5000 --reload
-    volumes:
-      - ../:/app
-
-  admin:
-    command: streamlit run --server.runOnSave=true tabby/admin/Home.py
-    volumes:
-      - ../:/app
-
-  vector:
     environment:
-      - VECTOR_WATCH_CONFIG=true
-
-  dagu:
-    ports:
-      - 8080:8080
+      UVICORN_RELOAD: true
+      VECTOR_WATCH_CONFIG: true
+      STREAMLIT_RUN_ON_SAVE: true
     volumes:
       - ../:/app
+      - ../deployment/config/vector.toml:/etc/vector/vector.toml:ro
+      - ../deployment/scripts/supervisord.sh:/usr/bin/supervisord.sh:ro

development/docker-compose.python.yml

@@ -1,7 +1,7 @@
 version: '3.3'
 
 services:
-  server:
+  tabby:
     image: tabbyml/tabby
     environment:
       MODEL_BACKEND: python

tabby/admin/Home.py

@@ -3,7 +3,7 @@ from components import monaco
 from utils.service_info import ServiceInfo
 
 SERVICES = [
-    ServiceInfo(label="server", url="http://server:5000"),
+    ServiceInfo(label="server", url="http://localhost:5000"),
     ServiceInfo(label="triton", url="http://triton:8002/metrics"),
 ]
 

tabby/admin/pages/Metrics.py

@@ -7,7 +7,7 @@ st.set_page_config(page_title="Tabby Admin - Metrics")
 
 
 def query_data():
-    filepath = os.environ.get("DB_FILE", "/logs/duckdb/duck.db")
+    filepath = os.environ.get("DB_FILE", "/data/logs/duckdb/duck.db")
     if not os.path.isfile(filepath):
         return []
 

tabby/tasks/duckdb.yaml

@@ -1,7 +1,9 @@
 schedule: "*/3 * * * *" # Run every 3rd minute
 
 env:
-  - PATH: "/opt/conda/bin:$PATH"
+  - PATH: "$PATH"
+  - LOGS_DIR: "$LOGS_DIR"
+  - DB_FILE: "$DB_FILE"
   - APP_DIR: /app
 steps:
   - name: Collect Tabby

tabby/tasks/init.yaml

@@ -1,8 +0,0 @@
-params: MODEL_NAME
-env:
-  - PATH: "/opt/conda/bin:$PATH"
-  - APP_DIR: /app
-steps:
-  - name: Download models
-    dir: $APP_DIR
-    command: python -m tabby.tools.download_models --repo_id=$MODEL_NAME

tabby/tools/duckdb/main.sh

@@ -1,8 +1,8 @@
 #!/bin/bash
 set -e
 
-DB_FILE=${DB_FILE:-"/logs/duckdb/duck.db"}
-LOGS_DIR=${LOGS_DIR:-"/logs"}
+DB_FILE=${DB_FILE:-"/data/logs/duckdb/duck.db"}
+LOGS_DIR=${LOGS_DIR:-"/data/logs"}
 TABBY_SERVER_LOGS="${LOGS_DIR}/tabby-server/events.*.json"
 
 # Init schema