Update app.py

HTTP metrics server kept but disabled by default.
To enable:
export METRICS_PORT=7861
streamlit run app.py
Log output:
# Disabled (default)
2025-12-06 [INFO] Metrics: log_interval=60s http_port=disabled

# Enabled
2025-12-06 [INFO] Metrics: log_interval=60s http_port=7861
2025-12-06 [INFO] Metrics server started on port 7861
Env vars:
| Variable | Default | Description |
|----------|---------|-------------|
| TORCH_THREADS | CPU/4 | PyTorch threads |
| WORKERS | CPU-TORCH | ThreadPool workers |
| METRICS_LOG_INTERVAL | 60 | Log interval (seconds) |
| METRICS_PORT | disabled | Set to enable HTTP server |
When enabled, endpoints:
/health - Health JSON
/metrics - Prometheus
/ready - Readiness probe
/live - Liveness probe

app.py

@@ -2,21 +2,20 @@
 Bilingual NLP + TTS - Nikud · Syntax · NER · Speech
 ====================================================
 
+Title: Bilingual NLP + TTS - Nikud · Syntax · NER · Speech
+
 Architecture:
     Text → pysbd Sentence Breaker → async.gather([process(s) for s in sentences]) → JSON Array
     
 Endpoints:
-    - Streamlit UI: default port (8501)
-    - /health: Health check (configurable via METRICS_PORT, default 80)
-    - /metrics: Prometheus metrics
-    - /ready: Readiness probe
-    - /live: Liveness probe
+    - Streamlit UI: port 7860
+    - /health, /metrics, /ready, /live: disabled (enable via METRICS_PORT env var)
 
 Environment Variables:
     TORCH_THREADS         - PyTorch threads (default: CPU_COUNT // 4)
     WORKERS               - ThreadPool workers (default: CPU_COUNT - TORCH_THREADS)
-    METRICS_PORT          - Metrics server port (default: 80)
     METRICS_LOG_INTERVAL  - Periodic metrics log interval in seconds (default: 60)
+    METRICS_PORT          - Enable HTTP metrics server on this port (disabled by default)
 """
 
 # ============================================================================
@@ -58,8 +57,8 @@ _default_workers = max(1, CPU_COUNT - _default_torch)
 
 TORCH_THREADS = _parse_env_int('TORCH_THREADS', _default_torch, min_val=1, max_val=CPU_COUNT)
 WORKERS = _parse_env_int('WORKERS', _default_workers, min_val=1, max_val=CPU_COUNT)
-METRICS_PORT = _parse_env_int('METRICS_PORT', 80, min_val=1, max_val=65535)
 METRICS_LOG_INTERVAL = _parse_env_int('METRICS_LOG_INTERVAL', 60, min_val=10, max_val=3600)
+METRICS_PORT = os.environ.get('METRICS_PORT')  # None = disabled, set to enable
 
 if TORCH_THREADS + WORKERS > CPU_COUNT:
     logger.warning(f"TORCH_THREADS({TORCH_THREADS}) + WORKERS({WORKERS}) exceeds CPU_COUNT({CPU_COUNT})")
@@ -70,7 +69,7 @@ torch.set_num_threads(TORCH_THREADS)
 # Note: set_num_interop_threads removed - can only be called once before any parallel work
 
 logger.info(f"CPU Config: CPUs={CPU_COUNT} Torch={TORCH_THREADS} Workers={WORKERS}")
-logger.info(f"Metrics: port={METRICS_PORT} log_interval={METRICS_LOG_INTERVAL}s")
+logger.info(f"Metrics: log_interval={METRICS_LOG_INTERVAL}s http_port={METRICS_PORT or 'disabled'}")
 
 # ============================================================================
 # Now safe to import everything else
@@ -289,7 +288,6 @@ class MetricsCollector:
                     "cpu_count": CPU_COUNT,
                     "torch_threads": TORCH_THREADS,
                     "workers": WORKERS,
-                    "metrics_port": METRICS_PORT,
                 },
                 "requests_total": self.requests_total,
                 "errors_total": self.errors_total,
@@ -380,7 +378,6 @@ class MetricsHandler(BaseHTTPRequestHandler):
             self.wfile.write(metrics.to_prometheus().encode())
         
         elif self.path == '/ready':
-            # Readiness probe
             ready = metrics.models_loaded
             self.send_response(200 if ready else 503)
             self.send_header('Content-Type', 'application/json')
@@ -388,7 +385,6 @@ class MetricsHandler(BaseHTTPRequestHandler):
             self.wfile.write(json.dumps({"ready": ready}).encode())
         
         elif self.path == '/live':
-            # Liveness probe
             self.send_response(200)
             self.send_header('Content-Type', 'application/json')
             self.end_headers()
@@ -399,10 +395,11 @@ class MetricsHandler(BaseHTTPRequestHandler):
             self.end_headers()
 
 
-def start_metrics_server(port: int = 80):
+def start_metrics_server(port: int = 7861):
     """Start metrics server in background thread."""
     def run_server():
         server = HTTPServer(('0.0.0.0', port), MetricsHandler)
+        logger.info(f"Metrics server started on port {port}")
         server.serve_forever()
     
     thread = threading.Thread(target=run_server, daemon=True)
@@ -410,8 +407,13 @@ def start_metrics_server(port: int = 80):
     return thread
 
 
-# Start metrics server on import
-_metrics_thread = start_metrics_server(port=METRICS_PORT)
+# Metrics HTTP server - disabled by default, enable via METRICS_PORT env var
+if METRICS_PORT:
+    try:
+        _metrics_port = int(METRICS_PORT)
+        _metrics_thread = start_metrics_server(port=_metrics_port)
+    except ValueError:
+        logger.warning(f"METRICS_PORT={METRICS_PORT} is not a valid port number, server disabled")
 
 
 def detect_script(text: str) -> str:
@@ -947,7 +949,6 @@ class AsyncHebrewNLP:
             "cpu_count": CPU_COUNT,
             "torch_threads": TORCH_THREADS,
             "workers": max_workers,
-            "metrics_port": METRICS_PORT,
         }
 
     async def process(
@@ -1809,7 +1810,7 @@ if models_loaded:
         noise_w = st.slider(t("noise_w"), 0.0, 2.0, 1.0, 0.1)
         sentence_pause = st.slider(t("sentence_pause"), 0.1, 1.0, 0.5, 0.1)
         
-        # Telemetry endpoints
+        # Telemetry
         st.divider()
         with st.expander("📊 Telemetry", expanded=False):
             health = metrics.to_health_json()
@@ -1818,18 +1819,12 @@ if models_loaded:
             st.caption(f"CPUs: {sys.get('cpu_count', '?')} | Torch: {sys.get('torch_threads', '?')} | Workers: {sys.get('workers', '?')}")
             st.caption(f"Requests: {health['requests_total']} | Errors: {health['errors_total']}")
             st.caption(f"Sentences: {health['totals']['sentences']} | Words: {health['totals']['words']}")
-            port = sys.get('metrics_port', 80)
-            st.code(f"""Endpoints (port {port}):
-/health  - Health JSON
-/metrics - Prometheus
-/ready   - Readiness probe
-/live    - Liveness probe
-
-Env vars:
+            metrics_status = f"enabled on port {METRICS_PORT}" if METRICS_PORT else "disabled"
+            st.code(f"""Env vars:
 TORCH_THREADS={TORCH_THREADS}
 WORKERS={WORKERS}
-METRICS_PORT={port}
-METRICS_LOG_INTERVAL={METRICS_LOG_INTERVAL}s""", language=None)
+METRICS_LOG_INTERVAL={METRICS_LOG_INTERVAL}s
+METRICS_PORT={metrics_status}""", language=None)
 
     # ========== MAIN INPUT SECTION ==========