fix: health check now probes HTTP directly with 3-min timeout

The previous approach relied on Docker's container health status, but
Docker's healthcheck (start_period:30s + 3x15s retries = ~75s) marks
the container "unhealthy" before NestJS finishes cold-starting after a
fresh image build (New Relic + TypeORM + Redis + BullMQ init can take
2-3 minutes).

Changes:
- Primary check is now direct wget to localhost:3000/api from the host
- Docker health status used only for informational logging
- Total timeout increased from 130s to 190s (~3 min) for cold starts
- Early exit if container has stopped/exited (no point waiting)
- More backend log lines (30 vs 20) shown on failure for diagnostics

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

scripts/deploy-prod.sh

@@ -40,9 +40,9 @@ DB_USER="${POSTGRES_USER:-hoafinance}"
 DB_NAME="${POSTGRES_DB:-hoafinance}"
 MIGRATION_DIR="$PROJECT_DIR/db/migrations"
 HEALTH_URL="http://localhost:3000/api"
-HEALTH_RETRIES=20
+HEALTH_RETRIES=36
 HEALTH_INTERVAL=5
-HEALTH_START_WAIT=30
+HEALTH_START_WAIT=10
 LOG_DIR="$PROJECT_DIR/logs"
 LOG_FILE="$LOG_DIR/deploy-$(date +%Y%m%d_%H%M%S).log"
 
@@ -359,38 +359,47 @@ fi
 # ====================================================================
 echo ""
 log "--- Step 5/6: Verifying application health ---"
-log "Waiting ${HEALTH_START_WAIT}s for backend to initialize (matches Docker start_period) ..."
+
+# After a fresh image build, NestJS cold-start can take 2-3 minutes:
+#   New Relic init → TypeORM connections → Redis → BullMQ → NestJS bootstrap
+# Docker's own healthcheck (start_period:30s + 3×15s retries = ~75s) is too
+# aggressive and will mark the container "unhealthy" before the app finishes
+# booting. So we do NOT rely on Docker's health status — we probe the HTTP
+# endpoint directly from the host and give it up to ~3 minutes total.
+TOTAL_WAIT=$((HEALTH_START_WAIT + HEALTH_RETRIES * HEALTH_INTERVAL))
+log "Will wait up to ${TOTAL_WAIT}s for backend to respond at $HEALTH_URL ..."
 sleep "$HEALTH_START_WAIT"
 
-# Primary check: Docker's own container health status
-# (docker-compose.prod.yml already defines a healthcheck using wget inside the container)
 HEALTHY=false
 for ((i=1; i<=HEALTH_RETRIES; i++)); do
-  CONTAINER_HEALTH=$($COMPOSE_CMD ps backend --format '{{.Health}}' 2>/dev/null || echo "unknown")
+  # Direct HTTP check from the host using wget (available on Ubuntu)
-  if [ "$CONTAINER_HEALTH" = "healthy" ]; then
-    HEALTHY=true
-    break
-  fi
-
-  # Also try a direct HTTP check from the host as a secondary signal
-  # Use wget (available on Ubuntu) since curl may not be installed
   if wget -qO- --timeout=5 "$HEALTH_URL" >/dev/null 2>&1; then
     HEALTHY=true
     break
   fi
 
-  log "  Health check attempt $i/$HEALTH_RETRIES — container status: ${CONTAINER_HEALTH}, retrying in ${HEALTH_INTERVAL}s ..."
+  # Also check Docker's container health for informational logging
+  CONTAINER_HEALTH=$($COMPOSE_CMD ps backend --format '{{.Health}}' 2>/dev/null || echo "unknown")
+
+  # If the container exited or was removed, fail immediately — no point waiting
+  CONTAINER_STATUS=$($COMPOSE_CMD ps backend --format '{{.Status}}' 2>/dev/null || echo "unknown")
+  if echo "$CONTAINER_STATUS" | grep -qi "exit\|dead\|removed"; then
+    err "Backend container has stopped unexpectedly: $CONTAINER_STATUS"
+    break
+  fi
+
+  log "  Health check attempt $i/$HEALTH_RETRIES — docker: ${CONTAINER_HEALTH}, retrying in ${HEALTH_INTERVAL}s ..."
   sleep "$HEALTH_INTERVAL"
 done
 
 if [ "$HEALTHY" = true ]; then
-  ok "Backend is healthy and responding"
+  ok "Backend is healthy and responding at $HEALTH_URL"
 else
   # Log diagnostics before triggering rollback
-  err "Backend failed to respond after $((HEALTH_START_WAIT + HEALTH_RETRIES * HEALTH_INTERVAL))s"
+  err "Backend failed to respond after ${TOTAL_WAIT}s"
   warn "Container status: $($COMPOSE_CMD ps backend 2>/dev/null || echo 'unknown')"
   warn "Recent backend logs:"
-  $COMPOSE_CMD logs --tail=20 backend 2>/dev/null || true
+  $COMPOSE_CMD logs --tail=30 backend 2>/dev/null || true
   err "Triggering automatic rollback ..."
   exit 1  # trap will handle rollback
 fi