@agentunion/kite 1.0.6 → 1.2.0

package/core/registry/entry.py

@@ -1,17 +1,249 @@
 """
 Registry entry point.
-Reads launcher_token, starts HTTP server on dynamic port, writes port.txt.
+Reads token from stdin boot_info, starts HTTP server on dynamic port,
+outputs port via stdout structured message, waits for Event Hub to trigger module.ready.
 """
 
+import builtins
+import json
 import os
+import re
 import sys
 import socket
+import threading
+import time
+import traceback
+from datetime import datetime, timezone
 
 import uvicorn
 
-# Ensure project root is on sys.path
+
+# ── Module configuration ──
+MODULE_NAME = "registry"
+
+
+def _fmt_elapsed(t0: float) -> str:
+    """Format elapsed time since t0: <1s → 'NNNms', >=1s → 'N.Ns', >=10s → 'NNs'."""
+    d = time.monotonic() - t0
+    if d < 1:
+        return f"{d * 1000:.0f}ms"
+    if d < 10:
+        return f"{d:.1f}s"
+    return f"{d:.0f}s"
+
+
+# ── Safe stdout/stderr: ignore BrokenPipeError after Launcher closes stdio ──
+
+class _SafeWriter:
+    """Wraps a stream to silently swallow BrokenPipeError on write/flush."""
+    def __init__(self, stream):
+        self._stream = stream
+
+    def write(self, s):
+        try:
+            self._stream.write(s)
+        except (BrokenPipeError, OSError):
+            pass
+
+    def flush(self):
+        try:
+            self._stream.flush()
+        except (BrokenPipeError, OSError):
+            pass
+
+    def __getattr__(self, name):
+        return getattr(self._stream, name)
+
+sys.stdout = _SafeWriter(sys.stdout)
+sys.stderr = _SafeWriter(sys.stderr)
+
+
+# ── Timestamped print + log file writer ──
+# Implements format: [timestamp] HH:MM:SS.mmm +delta [module_name] message
+# Independent implementation per module (no shared code dependency)
+
+_builtin_print = builtins.print
+_start_ts = time.monotonic()
+_last_ts = time.monotonic()
+_ANSI_RE = re.compile(r"\033\[[0-9;]*m")
+_log_lock = threading.Lock()
+_log_latest_path = None
+_log_daily_path = None
+_log_daily_date = ""
+_log_dir = None
+_crash_log_path = None
+
+def _strip_ansi(s: str) -> str:
+    return _ANSI_RE.sub("", s)
+
+def _resolve_daily_log_path():
+    """Resolve daily log path based on current date."""
+    global _log_daily_path, _log_daily_date
+    if not _log_dir:
+        return
+    today = datetime.now().strftime("%Y-%m-%d")
+    if today == _log_daily_date and _log_daily_path:
+        return
+    month_dir = os.path.join(_log_dir, today[:7])
+    os.makedirs(month_dir, exist_ok=True)
+    _log_daily_path = os.path.join(month_dir, f"{today}.log")
+    _log_daily_date = today
+
+def _write_log(plain_line: str):
+    """Write a plain-text line to both latest.log and daily log."""
+    with _log_lock:
+        if _log_latest_path:
+            try:
+                with open(_log_latest_path, "a", encoding="utf-8") as f:
+                    f.write(plain_line)
+            except Exception:
+                pass
+        _resolve_daily_log_path()
+        if _log_daily_path:
+            try:
+                with open(_log_daily_path, "a", encoding="utf-8") as f:
+                    f.write(plain_line)
+            except Exception:
+                pass
+
+def _write_crash(exc_type, exc_value, exc_tb, thread_name=None, severity="critical", handled=False):
+    """Write crash record to crashes.jsonl + daily crash archive."""
+    record = {
+        "timestamp": datetime.now(timezone.utc).isoformat(),
+        "module": MODULE_NAME,
+        "thread": thread_name or threading.current_thread().name,
+        "exception_type": exc_type.__name__ if exc_type else "Unknown",
+        "exception_message": str(exc_value),
+        "traceback": "".join(traceback.format_exception(exc_type, exc_value, exc_tb)),
+        "severity": severity,
+        "handled": handled,
+        "process_id": os.getpid(),
+        "platform": sys.platform,
+        "runtime_version": f"Python {sys.version.split()[0]}",
+    }
+
+    if exc_tb:
+        tb_entries = traceback.extract_tb(exc_tb)
+        if tb_entries:
+            last = tb_entries[-1]
+            record["context"] = {
+                "function": last.name,
+                "file": os.path.basename(last.filename),
+                "line": last.lineno,
+            }
+
+    line = json.dumps(record, ensure_ascii=False) + "\n"
+
+    # 1. Write to crashes.jsonl (current run)
+    if _crash_log_path:
+        try:
+            with open(_crash_log_path, "a", encoding="utf-8") as f:
+                f.write(line)
+        except Exception:
+            pass
+
+    # 2. Write to daily crash archive
+    if _log_dir:
+        try:
+            today = datetime.now().strftime("%Y-%m-%d")
+            archive_dir = os.path.join(_log_dir, "crashes", today[:7])
+            os.makedirs(archive_dir, exist_ok=True)
+            archive_path = os.path.join(archive_dir, f"{today}.jsonl")
+            with open(archive_path, "a", encoding="utf-8") as f:
+                f.write(line)
+        except Exception:
+            pass
+
+def _print_crash_summary(exc_type, exc_tb, thread_name=None):
+    """Print crash summary to console (red highlight)."""
+    RED = "\033[91m"
+    RESET = "\033[0m"
+
+    if exc_tb:
+        tb_entries = traceback.extract_tb(exc_tb)
+        if tb_entries:
+            last = tb_entries[-1]
+            location = f"{os.path.basename(last.filename)}:{last.lineno}"
+        else:
+            location = "unknown"
+    else:
+        location = "unknown"
+
+    prefix = f"[{MODULE_NAME}]"
+    if thread_name:
+        _builtin_print(f"{prefix} {RED}线程 {thread_name} 崩溃: "
+                       f"{exc_type.__name__} in {location}{RESET}")
+    else:
+        _builtin_print(f"{prefix} {RED}崩溃: {exc_type.__name__} in {location}{RESET}")
+    if _crash_log_path:
+        _builtin_print(f"{prefix} 崩溃日志: {_crash_log_path}")
+
+def _setup_exception_hooks():
+    """Set up global exception hooks (sys.excepthook + threading.excepthook)."""
+    _orig_excepthook = sys.excepthook
+
+    def _excepthook(exc_type, exc_value, exc_tb):
+        _write_crash(exc_type, exc_value, exc_tb, severity="critical", handled=False)
+        _print_crash_summary(exc_type, exc_tb)
+        _orig_excepthook(exc_type, exc_value, exc_tb)
+
+    sys.excepthook = _excepthook
+
+    if hasattr(threading, "excepthook"):
+        def _thread_excepthook(args):
+            _write_crash(args.exc_type, args.exc_value, args.exc_traceback,
+                        thread_name=args.thread.name if args.thread else "unknown",
+                        severity="error", handled=False)
+            _print_crash_summary(args.exc_type, args.exc_traceback,
+                                thread_name=args.thread.name if args.thread else None)
+
+        threading.excepthook = _thread_excepthook
+
+def _tprint(*args, **kwargs):
+    """Timestamped print that adds [timestamp] HH:MM:SS.mmm +delta prefix."""
+    global _last_ts
+    now = time.monotonic()
+    elapsed = now - _start_ts
+    delta = now - _last_ts
+    _last_ts = now
+
+    # Format elapsed (cumulative time since start)
+    if elapsed < 1:
+        elapsed_str = f"{elapsed * 1000:.0f}ms"
+    elif elapsed < 100:
+        elapsed_str = f"{elapsed:.1f}s"
+    else:
+        elapsed_str = f"{elapsed:.0f}s"
+
+    # Format delta (time since last print)
+    if delta < 0.001:
+        delta_str = ""
+    elif delta < 1:
+        delta_str = f"+{delta * 1000:.0f}ms"
+    elif delta < 100:
+        delta_str = f"+{delta:.1f}s"
+    else:
+        delta_str = f"+{delta:.0f}s"
+
+    # Current time
+    ts = datetime.now().strftime("%H:%M:%S.%f")[:-3]
+
+    # Print to console (original behavior)
+    _builtin_print(*args, **kwargs)
+
+    # Write to log files with timestamp prefix
+    if _log_latest_path or _log_daily_path:
+        sep = kwargs.get("sep", " ")
+        end = kwargs.get("end", "\n")
+        text = sep.join(str(a) for a in args)
+        prefix = f"[{elapsed_str:>6}] {ts} {delta_str:>8} "
+        _write_log(prefix + _strip_ansi(text) + end)
+
+builtins.print = _tprint
+
+# Ensure project root is on sys.path (set by main.py or cli.js)
 _this_dir = os.path.dirname(os.path.abspath(__file__))
-_project_root = os.path.dirname(os.path.dirname(_this_dir))
+_project_root = os.environ.get("KITE_PROJECT") or os.path.dirname(os.path.dirname(_this_dir))
 if _project_root not in sys.path:
     sys.path.insert(0, _project_root)
 
@@ -19,55 +251,128 @@ from core.registry.store import RegistryStore
 from core.registry.server import RegistryServer
 
 
-def _get_free_port() -> int:
+def _read_module_md() -> dict:
+    """Read preferred_port and advertise_ip from own module.md."""
+    md_path = os.path.join(_this_dir, "module.md")
+    result = {"preferred_port": 0, "advertise_ip": "127.0.0.1"}
+    try:
+        with open(md_path, "r", encoding="utf-8") as f:
+            text = f.read()
+        import re
+        m = re.match(r'^---\s*\n(.*?)\n---', text, re.DOTALL)
+        if m:
+            try:
+                import yaml
+                fm = yaml.safe_load(m.group(1)) or {}
+            except ImportError:
+                fm = {}
+            result["preferred_port"] = int(fm.get("preferred_port", 0))
+            result["advertise_ip"] = fm.get("advertise_ip", "127.0.0.1")
+    except Exception:
+        pass
+    return result
+
+
+def _bind_port(preferred: int, host: str) -> int:
+    """Try preferred port first, fall back to OS-assigned."""
+    if preferred:
+        try:
+            with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+                s.bind((host, preferred))
+                return preferred
+        except OSError:
+            pass
     with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
-        s.bind(("127.0.0.1", 0))
+        s.bind((host, 0))
         return s.getsockname()[1]
 
 
-def _write_port_file(port: int, instance_id: str = ""):
-    from core.data_dir import get_registry_data_dir
-    data_dir = get_registry_data_dir()
-    os.makedirs(data_dir, exist_ok=True)
-    filename = f"port_{instance_id}.txt" if instance_id else "port.txt"
-    port_file = os.path.join(data_dir, filename)
-    with open(port_file, "w") as f:
-        f.write(str(port))
-    print(f"[registry] Port written to {port_file}")
+def main():
+    # Initialize log file paths
+    global _log_dir, _log_latest_path, _crash_log_path
+    module_data = os.environ.get("KITE_MODULE_DATA")
+    if module_data:
+        _log_dir = os.path.join(module_data, "log")
+        os.makedirs(_log_dir, exist_ok=True)
+        suffix = os.environ.get("KITE_INSTANCE_SUFFIX", "")
 
+        # latest.log — truncate on each startup
+        _log_latest_path = os.path.join(_log_dir, f"latest{suffix}.log")
+        try:
+            with open(_log_latest_path, "w", encoding="utf-8") as f:
+                pass  # truncate
+        except Exception:
+            _log_latest_path = None
 
-def main():
-    # Read launcher_token and config from stdin (boot_info JSON pipe)
+        # crashes.jsonl — truncate on each startup
+        _crash_log_path = os.path.join(_log_dir, f"crashes{suffix}.jsonl")
+        try:
+            with open(_crash_log_path, "w", encoding="utf-8") as f:
+                pass  # truncate
+        except Exception:
+            _crash_log_path = None
+
+        _resolve_daily_log_path()
+
+    # Setup exception hooks
+    _setup_exception_hooks()
+
+    _t0 = time.monotonic()
+
+    # Kite environment
+    kite_instance = os.environ.get("KITE_INSTANCE", "")
+    is_debug = os.environ.get("KITE_DEBUG") == "1"
+
+    # Step 1: Read token from stdin boot_info (only token)
     launcher_token = ""
-    bind_host = "127.0.0.1"
-    instance_id = ""
     try:
-        import json
         line = sys.stdin.readline().strip()
         if line:
             boot_info = json.loads(line)
             launcher_token = boot_info.get("token", "")
-            bind_host = boot_info.get("bind", "127.0.0.1")
-            instance_id = boot_info.get("instance_id", "")
     except Exception:
         pass
 
     if not launcher_token:
-        print("[registry] ERROR: No launcher token provided via stdin boot_info")
+        print("[registry] 错误: 未通过 stdin boot_info 提供启动器令牌")
         sys.exit(1)
 
-    print(f"[registry] Launcher token received ({len(launcher_token)} chars)")
+    print(f"[registry] 已收到启动器令牌 ({len(launcher_token)} 字符) ({_fmt_elapsed(_t0)})")
 
-    # Create store and server
+    if is_debug:
+        print("[registry] 调试模式已启用 (KITE_DEBUG=1)，接受所有令牌")
+
+    # Step 2: Read config from own module.md
+    md_config = _read_module_md()
+    advertise_ip = md_config["advertise_ip"]
+    preferred_port = md_config["preferred_port"]
+
+    # Step 3: Create store and server
     store = RegistryStore(launcher_token)
-    server = RegistryServer(store, launcher_token=launcher_token)
+    server = RegistryServer(store, launcher_token=launcher_token, advertise_ip=advertise_ip)
+
+    # Step 4: Bind port
+    bind_host = advertise_ip
+    port = _bind_port(preferred_port, bind_host)
+
+    # Step 5: Output port via stdout structured message (Launcher reads this)
+    # This message proves HTTP is about to start — Launcher uses it as readiness signal
+    print(json.dumps({"kite": "port", "port": port}), flush=True)
 
-    # Bind to dynamic port
-    port = _get_free_port()
-    _write_port_file(port, instance_id)
+    print(f"[registry] 启动中 {bind_host}:{port} ({_fmt_elapsed(_t0)})")
 
-    print(f"[registry] Starting on {bind_host}:{port}")
-    uvicorn.run(server.app, host=bind_host, port=port, log_level="warning")
+    # Store port and advertise_ip for module.ready event later
+    server.port = port
+
+    try:
+        config = uvicorn.Config(server.app, host=bind_host, port=port, log_level="warning")
+        uvi_server = uvicorn.Server(config)
+        server._uvicorn_server = uvi_server
+        uvi_server.run()
+    except Exception as e:
+        _write_crash(type(e), e, e.__traceback__, severity="critical", handled=True)
+        _print_crash_summary(type(e), e.__traceback__)
+        sys.exit(1)
 
 
 if __name__ == "__main__":