fix(aca): reliable scale-to-zero wake-up (#47)

.github/workflows/cd.yml

@@ -416,3 +416,62 @@ jobs:
             --min-replicas 0 \
             --output none || true
           echo "Scale-to-zero restored."
+
+      - name: Verify scale-from-zero (real visitor path)
+        if: success()
+        shell: bash
+        env:
+          BACKEND_URL: ${{ needs.deploy-backend.outputs.backend_url }}
+          FRONTEND_URL: ${{ needs.deploy-frontend.outputs.frontend_url }}
+        run: |
+          set -euo pipefail
+          resolve_url() {
+            local name="$1" url="$2"
+            if [[ -n "$url" && "$url" != "https://" ]]; then
+              echo "$url"
+              return
+            fi
+            local fqdn
+            fqdn=$(az containerapp show \
+              --name "$name" \
+              --resource-group "$AZURE_RESOURCE_GROUP" \
+              --query "properties.configuration.ingress.fqdn" -o tsv)
+            echo "https://${fqdn}"
+          }
+          BACKEND=$(resolve_url "$BACKEND_APP_NAME" "${BACKEND_URL:-}")
+          FRONTEND=$(resolve_url "$FRONTEND_APP_NAME" "${FRONTEND_URL:-}")
+          echo "Scale-from-zero check — min-replicas=0, no warm-up..."
+          http="000"
+          echo "  backend: ${BACKEND}/health"
+          for i in $(seq 1 24); do
+            http=$(curl -sk -o /tmp/sfz-be.json -w "%{http_code}" --max-time 25 "${BACKEND}/health" || echo "000")
+            if [[ "$http" == "200" ]]; then
+              echo "    backend OK after ~$((i * 10))s"
+              break
+            fi
+            echo "    backend attempt $i/24 — HTTP $http"
+            sleep 10
+          done
+          if [[ "$http" != "200" ]]; then
+            echo "::error::Backend did not wake from zero within 240s"
+            az containerapp logs show --name "$BACKEND_APP_NAME" --resource-group "$AZURE_RESOURCE_GROUP" --tail 30 || true
+            exit 1
+          fi
+          http="000"
+          echo "  frontend: ${FRONTEND}/nginx-health"
+          for i in $(seq 1 24); do
+            http=$(curl -sk -o /tmp/sfz-fe.json -w "%{http_code}" --max-time 25 "${FRONTEND}/nginx-health" || echo "000")
+            if [[ "$http" == "200" ]]; then
+              echo "    frontend OK after ~$((i * 10))s"
+              break
+            fi
+            echo "    frontend attempt $i/24 — HTTP $http"
+            sleep 10
+          done
+          if [[ "$http" != "200" ]]; then
+            echo "::error::Frontend did not wake from zero within 240s"
+            az containerapp logs show --name "$FRONTEND_APP_NAME" --resource-group "$AZURE_RESOURCE_GROUP" --tail 30 || true
+            az containerapp revision list --name "$FRONTEND_APP_NAME" --resource-group "$AZURE_RESOURCE_GROUP" -o table || true
+            exit 1
+          fi
+          echo "Scale-from-zero wake-up verified."

README.md

@@ -13,6 +13,8 @@
 
 Live: https://insightiq-frontend.orangemeadow-25c8b891.westeurope.azurecontainerapps.io
 
+**Scale-to-zero:** both apps sleep when idle (`min-replicas: 0`). First visit after idle may take **30–90 seconds** while Azure starts containers. Open the [frontend](https://insightiq-frontend.orangemeadow-25c8b891.westeurope.azurecontainerapps.io) before a demo; sign-in wakes the backend automatically.
+
 ---
 
 ## Demo
@@ -80,7 +82,7 @@ Browser
   │     All API calls go directly: browser → backend HTTPS URL.
   │
   └─► Backend Container App (FastAPI + asyncpg + Claude)
-        min-replicas: 1  (always warm — no cold start on demo)
+        min-replicas: 0  (scale to zero — wakes on first HTTP request)
         max-replicas: 3  (HPA on CPU)
              │
              ▼
@@ -272,11 +274,11 @@ All deploys are zero-downtime rolling updates. The CD pipeline applies Neon migr
 
 | Service | Cost |
 |---------|------|
-| Azure Container Apps — backend (min 1 replica) | ~$5–8 |
+| Azure Container Apps — backend (scale to zero) | ~$1–5 |
 | Azure Container Apps — frontend (scale to zero) | ~$1–3 |
 | Neon Serverless PostgreSQL | $0 (free tier) |
 | GitHub Actions | $0 (free tier) |
-| **Total infrastructure** | **~$6–12/month** |
+| **Total infrastructure** | **~$2–8/month** |
 
 Anthropic API cost scales with query volume. With the 2-query lifetime cap per user and JSONB result caching, the marginal cost per unique visitor is bounded at approximately $0.01–0.03.
 

backend/Dockerfile

@@ -63,11 +63,11 @@ EXPOSE 8000
 HEALTHCHECK \
   --interval=30s \
   --timeout=10s \
-  --start-period=20s \
+  --start-period=15s \
   --retries=3 \
   CMD python3 -c "\
 import urllib.request, sys; \
-r = urllib.request.urlopen('http://localhost:8000/ready', timeout=10); \
+r = urllib.request.urlopen('http://localhost:8000/health', timeout=10); \
 sys.exit(0 if r.status == 200 else 1)"
 
 # Production server — 2 workers, no reload

backend/app/db/pool.py

@@ -12,6 +12,7 @@
 - All public methods log timing for observability.
 """
 
+import asyncio
 import logging
 import time
 from typing import Any, List, Optional
@@ -37,9 +38,18 @@ class DatabasePool:
 
     def __init__(self) -> None:
         self._pool: Optional[Pool] = None
+        self._init_lock = asyncio.Lock()
 
     # ── Lifecycle ──────────────────────────────────────────────────────────
 
+    async def ensure_ready(self) -> None:
+        """Create the pool on first use (supports fast /health on cold start)."""
+        if self._pool is not None:
+            return
+        async with self._init_lock:
+            if self._pool is None:
+                await self.initialise()
+
     async def initialise(self) -> None:
         """
         Create the asyncpg pool.
@@ -99,7 +109,7 @@ async def fetch(
         Raises:
             DatabaseError: On any asyncpg or timeout error.
         """
-        self._assert_pool_ready()
+        await self.ensure_ready()
         start = time.perf_counter()
         try:
             async with self._pool.acquire() as conn:
@@ -133,7 +143,7 @@ async def execute(
 
         Returns asyncpg status string e.g. 'INSERT 0 1'.
         """
-        self._assert_pool_ready()
+        await self.ensure_ready()
         try:
             async with self._pool.acquire() as conn:
                 return await conn.execute(
@@ -152,7 +162,7 @@ async def fetchrow(
         Fetch a single row. Returns None if no rows match.
         Useful for INSERT ... RETURNING id queries.
         """
-        self._assert_pool_ready()
+        await self.ensure_ready()
         try:
             async with self._pool.acquire() as conn:
                 record = await conn.fetchrow(query, *args)
@@ -162,14 +172,6 @@ async def fetchrow(
 
     # ── Private helpers ────────────────────────────────────────────────────
 
-    def _assert_pool_ready(self) -> None:
-        """Guard against calls before initialise() has run."""
-        if self._pool is None:
-            raise DatabaseError(
-                "Database pool is not initialised. "
-                "Ensure initialise() is called during app startup."
-            )
-
     @staticmethod
     async def _set_type_codecs(conn: asyncpg.Connection) -> None:
         """

backend/main.py

@@ -26,13 +26,10 @@
 
 @asynccontextmanager
 async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
-    logger.info("InsightIQ starting — initialising database pool...")
-    await database_pool.initialise()
-
-    # Log CORS origins at startup so you can verify in ACA logs
+    # Do not block on DB here — ACA scale-from-zero probes /health before Neon wakes.
     cors_origins = settings.get_all_cors_origins()
+    logger.info("InsightIQ starting — /health live; DB pool connects on first API call")
     logger.info(f"CORS allowed origins: {cors_origins}")
-    logger.info("Startup complete")
 
     yield
 
@@ -88,6 +85,7 @@ async def health_check() -> dict:
 @app.get("/ready", tags=["Health"])
 async def readiness_check() -> dict:
     """Readiness — confirms database connectivity."""
+    await database_pool.ensure_ready()
     await database_pool.execute("SELECT 1")
     return {"status": "ready", "version": "1.0.0"}
 

frontend/Dockerfile

@@ -39,7 +39,7 @@ LABEL org.opencontainers.image.title="insightiq-frontend"
 
 # Render nginx template at start (BACKEND_UPSTREAM from ACA env)
 RUN rm -f /etc/nginx/conf.d/default.conf
-RUN apk add --no-cache gettext
+RUN apk add --no-cache gettext curl
 COPY nginx.conf.template /etc/nginx/templates/default.conf.template
 COPY docker-entrypoint.sh /docker-entrypoint.sh
 RUN chmod +x /docker-entrypoint.sh
@@ -55,7 +55,8 @@ EXPOSE 80
 HEALTHCHECK \
   --interval=30s \
   --timeout=5s \
+  --start-period=10s \
   --retries=3 \
-  CMD wget -qO- http://localhost:80 || exit 1
+  CMD curl -fsS http://localhost/nginx-health || exit 1
 
 ENTRYPOINT ["/docker-entrypoint.sh"]

frontend/docker-entrypoint.sh

@@ -8,4 +8,6 @@ envsubst '${BACKEND_UPSTREAM}' \
   < /etc/nginx/templates/default.conf.template \
   > /etc/nginx/conf.d/default.conf
 
+nginx -t
+
 exec nginx -g 'daemon off;'

infra/fix-scale-to-zero.sh

@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+# =================================================================
+# InsightIQ — Fix Azure scale-to-zero + failed revision activation
+#
+# Run when portal shows "Activation failed" or links hang forever.
+#
+# Prerequisites:
+#   az login
+#   az account set --subscription YOUR_SUBSCRIPTION_ID
+#
+# Usage:
+#   chmod +x infra/fix-scale-to-zero.sh
+#   ./infra/fix-scale-to-zero.sh
+# =================================================================
+
+set -euo pipefail
+
+RESOURCE_GROUP="${AZURE_RESOURCE_GROUP:-rg-insightiq-prod}"
+BACKEND_APP="${BACKEND_APP_NAME:-insightiq-backend}"
+FRONTEND_APP="${FRONTEND_APP_NAME:-insightiq-frontend}"
+
+echo "=================================================="
+echo " InsightIQ — scale-to-zero repair"
+echo "=================================================="
+echo " Resource group : $RESOURCE_GROUP"
+echo " Backend app    : $BACKEND_APP"
+echo " Frontend app   : $FRONTEND_APP"
+echo ""
+
+for app in "$BACKEND_APP" "$FRONTEND_APP"; do
+  echo "--- $app ---"
+  az containerapp show \
+    --name "$app" \
+    --resource-group "$RESOURCE_GROUP" \
+    --query "{url:properties.configuration.ingress.fqdn,scale:properties.template.scale,running:properties.runningStatus,latest:properties.latestRevisionName}" \
+    -o json
+  echo "Recent revisions:"
+  az containerapp revision list \
+    --name "$app" \
+    --resource-group "$RESOURCE_GROUP" \
+    --query "[].{name:name,active:properties.active,traffic:properties.trafficWeight,health:properties.healthState,replicas:properties.replicas}" \
+    -o table
+  echo ""
+done
+
+echo "[1/3] Enforcing min-replicas=0 (idle cost) on both apps..."
+az containerapp update --name "$BACKEND_APP"  --resource-group "$RESOURCE_GROUP" --min-replicas 0 --output none
+az containerapp update --name "$FRONTEND_APP" --resource-group "$RESOURCE_GROUP" --min-replicas 0 --output none
+echo "      ✓ min-replicas=0"
+
+echo ""
+echo "[2/3] Checklist before redeploy:"
+echo "  • GitHub Packages → insightiq-frontend / insightiq-backend → Package settings → Public"
+echo "    (or set GHCR_READ_TOKEN secret with read:packages scope)"
+echo "  • GitHub Secrets: NEON_DATABASE_URL, JWT_SECRET_KEY, ANTHROPIC_API_KEY, GOOGLE_CLIENT_ID"
+echo "  • GitHub → Actions → CD → Run workflow on main"
+echo ""
+echo "[3/3] After CD succeeds, test wake-up (first click may take 30–90s):"
+BACKEND_FQDN=$(az containerapp show --name "$BACKEND_APP"  --resource-group "$RESOURCE_GROUP" --query "properties.configuration.ingress.fqdn" -o tsv)
+FRONTEND_FQDN=$(az containerapp show --name "$FRONTEND_APP" --resource-group "$RESOURCE_GROUP" --query "properties.configuration.ingress.fqdn" -o tsv)
+echo "  Frontend : https://${FRONTEND_FQDN}/"
+echo "  Backend  : https://${BACKEND_FQDN}/health"
+echo ""
+echo "If a revision still shows 'Activation failed' with 100% traffic,"
+echo "open Portal → Revisions → deactivate the failed revision, then re-run CD."
+echo "=================================================="