feat: カーネル接続の確認機能を追加し、タイムアウト設定を導入

Author: shima004

adf_core_python/core/launcher/agent_launcher.py

@@ -1,4 +1,5 @@
 import importlib
+import socket
 import threading
 from typing import Optional
 
@@ -59,13 +60,27 @@ def init_connector(self) -> None:
     def launch(self) -> None:
         kernel_host: str = self.config.get_value(ConfigKey.KEY_KERNEL_HOST, "localhost")
         kernel_port: int = self.config.get_value(ConfigKey.KEY_KERNEL_PORT, 27931)
-        self.logger.info(
-            f"Start agent launcher (host: {kernel_host}, port: {kernel_port})"
-        )
 
         component_launcher: ComponentLauncher = ComponentLauncher(
             kernel_host, kernel_port, self.logger
         )
+        timeout: int = self.config.get_value(
+            ConfigKey.KEY_KERNEL_TIMEOUT,
+            30,
+        )
+        if component_launcher.check_kernel_connection(timeout=timeout):
+            self.logger.info(
+                f"Kernel is running (host: {kernel_host}, port: {kernel_port})"
+            )
+        else:
+            self.logger.error(
+                f"Kernel is not running (host: {kernel_host}, port: {kernel_port})"
+            )
+            return
+
+        self.logger.info(
+            f"Start agent launcher (host: {kernel_host}, port: {kernel_port})"
+        )
 
         gateway_launcher: Optional[GatewayLauncher] = None
         gateway_flag: bool = self.config.get_value(ConfigKey.KEY_GATEWAY_FLAG, False)
@@ -104,3 +119,14 @@ def connect() -> None:
 
         for thread in self.agent_thread_list:
             thread.join()
+
+    def check_kernel_connection(self, host: str, port: int, timeout: int = 5) -> bool:
+        try:
+            sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+            sock.settimeout(timeout)
+            result = sock.connect_ex((host, port))
+            sock.close()
+            return result == 0
+        except Exception as e:
+            self.logger.error(f"カーネルへの接続確認中にエラーが発生しました: {e}")
+            return False

adf_core_python/core/launcher/config_key.py

@@ -6,6 +6,7 @@ class ConfigKey:
     KEY_LOADER_CLASS: Final[str] = "adf_core_python.launcher.loader"
     KEY_KERNEL_HOST: Final[str] = "kernel.host"
     KEY_KERNEL_PORT: Final[str] = "kernel.port"
+    KEY_KERNEL_TIMEOUT: Final[str] = "kernel.timeout"
     KEY_TEAM_NAME: Final[str] = "team.name"
     KEY_DEBUG_FLAG: Final[str] = "adf.debug.flag"
     KEY_DEVELOP_FLAG: Final[str] = "adf.develop.flag"

adf_core_python/core/launcher/connect/component_launcher.py

@@ -1,4 +1,5 @@
 import socket
+import time
 
 from structlog import BoundLogger
 
@@ -53,3 +54,48 @@ def connect(self, agent: Agent, _request_id: int) -> None:
     def generate_request_id(self) -> int:
         self.request_id += 1
         return self.request_id
+
+    def check_kernel_connection(
+        self, timeout: int = 30, retry_interval: float = 5.0
+    ) -> bool:
+        """Attempts to connect to the kernel multiple times within the specified timeout period.
+
+        Args:
+            timeout (int): Total timeout duration in seconds
+            retry_interval (float): Interval between retry attempts in seconds
+
+        Returns:
+            bool: True if connection successful, False otherwise
+        """
+        start_time = time.time()
+        attempt = 1
+
+        while True:
+            try:
+                sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+                sock.settimeout(retry_interval)
+                result = sock.connect_ex((self.host, self.port))
+                sock.close()
+
+                if result == 0:
+                    self.logger.info(
+                        f"Successfully connected to kernel (attempt: {attempt})"
+                    )
+                    return True
+
+                elapsed_time = time.time() - start_time
+                if elapsed_time >= timeout:
+                    self.logger.error(
+                        f"Timeout: Could not connect to kernel within {timeout} seconds (attempts: {attempt})"
+                    )
+                    return False
+
+                self.logger.debug(
+                    f"Connection attempt {attempt} failed - retrying in {retry_interval} seconds"
+                )
+                time.sleep(retry_interval)
+                attempt += 1
+
+            except Exception as e:
+                self.logger.error(f"Error while checking kernel connection: {e}")
+                return False

adf_core_python/launcher.py

@@ -12,11 +12,8 @@ def __init__(
         self,
         launcher_config_file: str,
     ) -> None:
-        resource.setrlimit(resource.RLIMIT_NOFILE, (8192, 9223372036854775807))
+        resource.setrlimit(resource.RLIMIT_NOFILE, (8192, 1048576))
 
-        configure_logger()
-
-        self.logger = get_logger(__name__)
         self.launcher_config = Config(launcher_config_file)
 
         parser = argparse.ArgumentParser(description="Agent Launcher")
@@ -80,6 +77,12 @@ def __init__(
             action="store_true",
             help="precompute flag",
         )
+        parser.add_argument(
+            "--timeout",
+            type=int,
+            help="timeout in seconds",
+            metavar="",
+        )
         parser.add_argument("--debug", action="store_true", help="debug flag")
         parser.add_argument(
             "--java",
@@ -98,6 +101,7 @@ def __init__(
             ConfigKey.KEY_FIRE_STATION_COUNT: args.firestation,
             ConfigKey.KEY_POLICE_OFFICE_COUNT: args.policeoffice,
             ConfigKey.KEY_PRECOMPUTE: args.precompute,
+            ConfigKey.KEY_KERNEL_TIMEOUT: args.timeout,
             ConfigKey.KEY_DEBUG_FLAG: args.debug,
             ConfigKey.KEY_GATEWAY_FLAG: args.java,
         }
@@ -106,6 +110,9 @@ def __init__(
             if value is not None:
                 self.launcher_config.set_value(key, value)
 
+        configure_logger()
+        self.logger = get_logger(__name__)
+
         self.logger.debug(f"launcher_config: {self.launcher_config}")
 
     def launch(self) -> None: