Day 07 — Production wrapper + tests + ops dashboard + sprint close (Sentinel PROJECT COMPLETE)

.github/workflows/ci.yml

@@ -0,0 +1,65 @@
+# Sentinel CI -- Day 7 Phase 6 (2026-05-24).
+#
+# Lightweight CI: validates the DVC pipeline DAG and runs the deterministic
+# unit test surface. The dataset-dependent stages (train/evaluate/benchmark)
+# are intentionally NOT executed -- they need the multi-GB raw data on disk
+# and aren't suited to a 6-minute hosted runner. The DVC DAG check ensures
+# the pipeline definition stays parseable as src/ evolves; the unit tests
+# exercise the production wrapper end-to-end against in-memory fixtures.
+
+name: ci
+
+on:
+  push:
+    branches: [main, dev]
+  pull_request:
+    branches: [main, dev]
+  workflow_dispatch:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+          cache: "pip"
+          cache-dependency-path: requirements.txt
+
+      - name: Install runtime dependencies
+        run: |
+          python -m pip install --upgrade pip
+          # Install the pinned runtime deps. Day-4 added FastAPI / SQLAlchemy
+          # / psycopg2 / httpx into requirements.txt, so a single install
+          # covers both training-side and serving-side tests.
+          pip install -r requirements.txt
+          # Extra test-only deps (pytest plus stdlib-only helpers).
+          pip install pytest pytest-asyncio
+
+      - name: Show installed versions
+        run: |
+          python --version
+          pip list | grep -Ei "dvc|mlflow|xgboost|pandas|dask|fastapi|pydantic|streamlit" || true
+
+      - name: Validate DVC pipeline DAG
+        run: |
+          # `dvc dag` parses dvc.yaml and prints the stage graph -- a cheap
+          # way to fail loudly if a stage definition or path drifts.
+          dvc dag --quiet || true
+          dvc dag
+
+      - name: Run unit tests
+        env:
+          # Keep test runs hermetic: in-memory sqlite for both MLflow and
+          # the telemetry store, no shadow evaluator at app import time.
+          MLFLOW_TRACKING_URI: "sqlite:///${{ github.workspace }}/.ci_mlflow.db"
+          SENTINEL_DISABLE_SHADOW: "1"
+          SENTINEL_BUILD_APP_AT_IMPORT: "0"
+        run: |
+          pytest tests/ -q -m "not requires_data" --maxfail=1 --disable-warnings

docker-compose.yml

@@ -1,9 +1,13 @@
 version: "3.9"
 
-# Sentinel telemetry stack — Day 4 Phase 3 (2026-05-21).
-# Brings up the Postgres backing store the FastAPI service writes to.
-# MLflow remains on the local sqlite store (mlflow.db) for now — Day 7
-# graduates the full stack to docker compose.
+# Sentinel full MLOps stack -- Day 7 Phase 6 (2026-05-24).
+# One `docker compose up` brings the entire stack: Postgres telemetry,
+# MLflow tracking + registry, Redis cache, and the FastAPI serving image.
+#
+# Day-4 introduced this file with Postgres + a profiled API service; Day-7
+# layers MLflow (tracking + registry, backed by its own Postgres database
+# on the same instance) and Redis (per-card prediction caching) so the
+# whole production wrapper boots with a single command.
 
 services:
   postgres:
@@ -13,19 +17,74 @@ services:
     environment:
       POSTGRES_USER: ${POSTGRES_USER:-sentinel}
       POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-sentinel}
+      # Two logical databases on the same Postgres instance: one for the
+      # telemetry store the FastAPI service writes to, one for MLflow's
+      # tracking + registry backend. POSTGRES_DB creates the first; the
+      # second is bootstrapped by ./scripts/postgres-init.sh.
       POSTGRES_DB: ${POSTGRES_DB:-sentinel_telemetry}
+      POSTGRES_MULTIPLE_DATABASES: ${POSTGRES_MULTIPLE_DATABASES:-mlflow}
     ports:
       - "${POSTGRES_PORT:-5432}:5432"
     volumes:
       - sentinel_pgdata:/var/lib/postgresql/data
+      - ./scripts/postgres-init.sh:/docker-entrypoint-initdb.d/00-multi-db.sh:ro
     healthcheck:
       test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER:-sentinel} -d ${POSTGRES_DB:-sentinel_telemetry}"]
       interval: 5s
       timeout: 5s
       retries: 5
 
-  # Convenience: serve the FastAPI app off the same machine that owns Postgres.
-  # Day 7 layers MLflow and Redis on top.
+  mlflow:
+    image: ghcr.io/mlflow/mlflow:v2.18.0
+    container_name: sentinel-mlflow
+    restart: unless-stopped
+    depends_on:
+      postgres:
+        condition: service_healthy
+    environment:
+      MLFLOW_BACKEND_STORE_URI: postgresql+psycopg2://${POSTGRES_USER:-sentinel}:${POSTGRES_PASSWORD:-sentinel}@postgres:5432/mlflow
+      MLFLOW_DEFAULT_ARTIFACT_ROOT: /mlartifacts
+    # ghcr.io/mlflow/mlflow ships without psycopg2; install it once at
+    # container start before launching the server. Artifacts go to a named
+    # volume so they survive container restarts.
+    command: >
+      bash -c "pip install --quiet psycopg2-binary &&
+               mlflow server
+               --host 0.0.0.0
+               --port 5000
+               --backend-store-uri postgresql+psycopg2://${POSTGRES_USER:-sentinel}:${POSTGRES_PASSWORD:-sentinel}@postgres:5432/mlflow
+               --artifacts-destination /mlartifacts"
+    ports:
+      - "${MLFLOW_PORT:-5000}:5000"
+    volumes:
+      - sentinel_mlartifacts:/mlartifacts
+    healthcheck:
+      test: ["CMD-SHELL", "python -c 'import urllib.request,sys; sys.exit(0 if urllib.request.urlopen(\"http://localhost:5000/health\").status==200 else 1)'"]
+      interval: 10s
+      timeout: 5s
+      retries: 10
+      start_period: 30s
+
+  redis:
+    image: redis:7-alpine
+    container_name: sentinel-redis
+    restart: unless-stopped
+    # Cache-only profile: prediction caching is a hot-path optimisation, no
+    # need for AOF/RDB. Bound memory; LRU eviction.
+    command: >
+      redis-server
+      --maxmemory ${REDIS_MAXMEMORY:-128mb}
+      --maxmemory-policy allkeys-lru
+      --save ""
+      --appendonly no
+    ports:
+      - "${REDIS_PORT:-6379}:6379"
+    healthcheck:
+      test: ["CMD", "redis-cli", "ping"]
+      interval: 5s
+      timeout: 3s
+      retries: 5
+
   api:
     profiles: ["serving"]
     build:
@@ -36,12 +95,18 @@ services:
     depends_on:
       postgres:
         condition: service_healthy
+      mlflow:
+        condition: service_healthy
+      redis:
+        condition: service_healthy
     environment:
       SENTINEL_DATABASE_URL: postgresql+psycopg2://${POSTGRES_USER:-sentinel}:${POSTGRES_PASSWORD:-sentinel}@postgres:5432/${POSTGRES_DB:-sentinel_telemetry}
-      MLFLOW_TRACKING_URI: ${MLFLOW_TRACKING_URI:-sqlite:////app/mlflow.db}
+      MLFLOW_TRACKING_URI: ${MLFLOW_TRACKING_URI:-http://mlflow:5000}
+      SENTINEL_REDIS_URL: ${SENTINEL_REDIS_URL:-redis://redis:6379/0}
       SENTINEL_DISABLE_SHADOW: "${SENTINEL_DISABLE_SHADOW:-1}"
     ports:
       - "${API_PORT:-8000}:8000"
 
 volumes:
   sentinel_pgdata:
+  sentinel_mlartifacts: