feat: Add initial cross-VPS orchestration framework with deploy-all command (Refs #552)

config.yaml

@@ -165,6 +165,9 @@ security:
   # Empty list = deny all (secure by default)
   # Set via env var TIMMY_AUTHOR_WHITELIST as comma-separated list
   author_whitelist: []
+fleet_management:
+  fleet_manager_script: scripts/fleet_manager.py
+
 _config_version: 9
 session_reset:
   mode: none

fleet_config.yaml

@@ -0,0 +1,4 @@
+hosts:
+  - vps1.example.com
+  - vps2.example.com
+  - vps3.example.com

scripts/fleet_manager.py

@@ -0,0 +1,87 @@
+import argparse
+import os
+import subprocess
+import sys
+import yaml
+
+FLEET_CONFIG_PATH = "fleet_config.yaml"
+
+def _read_fleet_config():
+    if not os.path.exists(FLEET_CONFIG_PATH):
+        print(f"Error: {FLEET_CONFIG_PATH} not found.")
+        return {"hosts": []}
+    with open(FLEET_CONFIG_PATH, 'r') as f:
+        return yaml.safe_load(f)
+
+def deploy_agent_to_host(host, agent_id, config_path):
+    print(f"Deploying agent {agent_id} with config {config_path} to {host} (simulated)...")
+    # In a real scenario, this would involve SSHing to the host and deploying the agent
+    print(f"Agent {agent_id} deployed to {host} (simulated).")
+
+def deploy_all_agents(agent_id, config_path):
+    config = _read_fleet_config()
+    hosts = config.get("hosts", [])
+    if not hosts:
+        print("No hosts configured in fleet_config.yaml. Aborting deploy-all.")
+        return
+
+    print(f"Deploying agent {agent_id} to all configured hosts...")
+    for host in hosts:
+        deploy_agent_to_host(host, agent_id, config_path)
+    print("Fleet-wide deployment initiated (simulated).")
+
+def start_agent(agent_id, config_path):
+    print(f"Starting agent {agent_id} with config {config_path}...")
+    # This is a placeholder for actual agent startup logic
+    # In a real scenario, this would involve SSHing to a VPS and starting a systemd service or docker container
+    print(f"Agent {agent_id} started (simulated).")
+
+def stop_agent(agent_id):
+    print(f"Stopping agent {agent_id}...")
+    # Placeholder for actual agent stopping logic
+    print(f"Agent {agent_id} stopped (simulated).")
+
+def list_agents():
+    print("Listing active agents (simulated)...")
+    # Placeholder for listing active agents
+    active_agents = ["agent-1", "agent-2"]
+    for agent in active_agents:
+        print(f"- {agent}")
+
+def main():
+    parser = argparse.ArgumentParser(description="Fleet Manager for Timmy Agents")
+    subparsers = parser.add_subparsers(dest="command", help="Available commands")
+
+    # Start command
+    start_parser = subparsers.add_parser("start", help="Start an agent")
+    start_parser.add_argument("agent_id", help="ID of the agent to start")
+    start_parser.add_argument("--config", default="default", help="Configuration path for the agent")
+
+    # Stop command
+    stop_parser = subparsers.add_parser("stop", help="Stop an agent")
+    stop_parser.add_argument("agent_id", help="ID of the agent to stop")
+
+    # List command
+    list_parser = subparsers.add_parser("list", help="List active agents")
+
+    # Deploy-all command
+    deploy_all_parser = subparsers.add_parser("deploy-all", help="Deploy an agent to all configured hosts")
+    deploy_all_parser.add_argument("agent_id", help="ID of the agent to deploy")
+    deploy_all_parser.add_argument("--config", default="default", help="Configuration path for the agent")
+
+
+    args = parser.parse_args()
+
+    if args.command == "start":
+        start_agent(args.agent_id, args.config)
+    elif args.command == "stop":
+        stop_agent(args.agent_id)
+    elif args.command == "list":
+        list_agents()
+    elif args.command == "deploy-all":
+        deploy_all_agents(args.agent_id, args.config)
+    else:
+        parser.print_help()
+
+if __name__ == "__main__":
+    main()

scripts/fleet_orchestrator.py

@@ -0,0 +1,130 @@
+
+import subprocess
+import sys
+import os
+
+FLEET_HOSTS = os.environ.get("FLEET_HOSTS", "143.198.27.163 104.131.15.18").split()
+TIMMY_USER = os.environ.get("TIMMY_USER", "root")
+TIMMY_DIR = os.environ.get("TIMMY_HOME", "/root") + "/timmy"
+
+
+def run_remote_command(host, command):
+    """Executes a command remotely on a given host via SSH."""
+    ssh_command = ["ssh", f"{TIMMY_USER}@{host}", command]
+    print(f"Executing on {host}: {' '.join(ssh_command)}")
+    try:
+        result = subprocess.run(ssh_command, capture_output=True, text=True, check=True)
+        print(f"[{host}] STDOUT:\n{result.stdout}")
+        if result.stderr:
+            print(f"[{host}] STDERR:\n{result.stderr}")
+        return result.stdout
+    except subprocess.CalledProcessError as e:
+        print(f"[{host}] ERROR: Command failed with exit code {e.returncode}")
+        print(f"[{host}] STDOUT:\n{e.stdout}")
+        print(f"[{host}] STDERR:\n{e.stderr}")
+        return None
+    except Exception as e:
+        print(f"[{host}] AN UNEXPECTED ERROR OCCURRED: {e}")
+        return None
+
+
+def deploy_agent(host):
+    """Deploys the agent on a remote host, including its systemd service and code."""
+    print(f"Deploying agent on {host}...")
+
+    # Ensure the remote directory structure exists
+    run_remote_command(host, f"mkdir -p {TIMMY_DIR}/timmy-home/agent")
+    run_remote_command(host, f"mkdir -p /etc/systemd/system")
+
+    # Copy the timmy-agent.service file
+    subprocess.run(["scp", "configs/timmy-agent.service", f"{TIMMY_USER}@{host}:/etc/systemd/system/timmy-agent.service"], check=True)
+
+    # Copy the agent code
+    # Assuming 'agent' directory is parallel to 'scripts' within 'timmy-home'
+    subprocess.run(["scp", "-r", "agent/", f"{TIMMY_USER}@{host}:{TIMMY_DIR}/timmy-home/"], check=True)
+
+    # Reload systemd, enable and start the service
+    commands = [
+        "systemctl daemon-reload",
+        "systemctl enable timmy-agent.service",
+        "systemctl start timmy-agent.service"
+    ]
+    for cmd in commands:
+        run_remote_command(host, cmd)
+
+
+
+def retire_agent(host):
+    """Retires the agent on a remote host by stopping, disabling, and removing its systemd service and code."""
+    print(f"Retiring agent on {host}...")
+    commands = [
+        "systemctl stop timmy-agent.service",
+        "systemctl disable timmy-agent.service",
+        "rm -f /etc/systemd/system/timmy-agent.service",
+        "systemctl daemon-reload", # Reload daemon to remove the service from memory
+        f"rm -rf {TIMMY_DIR}/timmy-home/agent" # Optional: remove agent code
+    ]
+    for cmd in commands:
+        run_remote_command(host, cmd)
+
+
+def start_agent(host):
+    """Starts the timmy-agent.service on a remote host."""
+    print(f"Starting agent on {host}...")
+    run_remote_command(host, f"systemctl start timmy-agent.service")
+
+
+def stop_agent(host):
+    """Stops the timmy-agent.service on a remote host."""
+    print(f"Stopping agent on {host}...")
+    run_remote_command(host, f"systemctl stop timmy-agent.service")
+
+
+def update_agent(host):
+    """Pulls the latest timmy-home repo and restarts the agent on a remote host."""
+    print(f"Updating agent on {host}...")
+    commands = [
+        f"cd {TIMMY_DIR}/timmy-home && git pull",
+        f"systemctl restart timmy-agent.service"
+    ]
+    for cmd in commands:
+        run_remote_command(host, cmd)
+
+
+def status_agent(host):
+    """Checks the status of the timmy-agent.service on a remote host."""
+    print(f"Checking agent status on {host}...")
+    run_remote_command(host, f"systemctl status timmy-agent.service --no-pager")
+
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: python fleet_orchestrator.py <command> [host]")
+        print("Commands: deploy, start, stop, update, status")
+        sys.exit(1)
+
+    action = sys.argv[1]
+    target_host = sys.argv[2] if len(sys.argv) > 2 else None
+
+    hosts_to_target = [target_host] if target_host else FLEET_HOSTS
+
+    for host in hosts_to_target:
+        if action == "deploy":
+            deploy_agent(host)
+        elif action == "start":
+            start_agent(host)
+        elif action == "stop":
+            stop_agent(host)
+        elif action == "update":
+            update_agent(host)
+        elif action == "status":
+            status_agent(host)
+        elif action == "retire":
+            retire_agent(host)
+        else:
+            print(f"Unknown command: {action}")
+            sys.exit(1)
+
+
+if __name__ == "__main__":
+    main()

tests/test_fleet_manager.py

@@ -0,0 +1,58 @@
+import unittest
+from unittest.mock import patch, MagicMock
+import sys
+import os
+
+# Adjust the path to import fleet_manager from scripts directory
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '..', 'scripts')))
+import fleet_manager
+
+class TestFleetManager(unittest.TestCase):
+
+    @patch('fleet_manager._read_fleet_config')
+    @patch('builtins.print')
+    def test_deploy_all_agents(self, mock_print, mock_read_config):
+        mock_read_config.return_value = {"hosts": ["host1.example.com", "host2.example.com"]}
+        
+        # Mock sys.argv to simulate command line arguments
+        with patch.object(sys, 'argv', ['fleet_manager.py', 'deploy-all', 'test-agent', '--config', 'test-config']):
+            fleet_manager.main()
+
+        mock_print.assert_any_call("Deploying agent test-agent with config test-config to host1.example.com (simulated)...")
+        mock_print.assert_any_call("Deploying agent test-agent with config test-config to host2.example.com (simulated)...")
+        mock_print.assert_any_call("Fleet-wide deployment initiated (simulated).")
+
+    @patch('fleet_manager._read_fleet_config')
+    @patch('builtins.print')
+    def test_deploy_all_agents_no_hosts(self, mock_print, mock_read_config):
+        mock_read_config.return_value = {"hosts": []}
+
+        with patch.object(sys, 'argv', ['fleet_manager.py', 'deploy-all', 'test-agent']):
+            fleet_manager.main()
+
+        mock_print.assert_any_call("No hosts configured in fleet_config.yaml. Aborting deploy-all.")
+
+    @patch('builtins.print')
+    def test_start_agent(self, mock_print):
+        with patch.object(sys, 'argv', ['fleet_manager.py', 'start', 'agent-x', '--config', 'config-x']):
+            fleet_manager.main()
+        mock_print.assert_any_call("Starting agent agent-x with config config-x...")
+        mock_print.assert_any_call("Agent agent-x started (simulated).")
+
+    @patch('builtins.print')
+    def test_stop_agent(self, mock_print):
+        with patch.object(sys, 'argv', ['fleet_manager.py', 'stop', 'agent-y']):
+            fleet_manager.main()
+        mock_print.assert_any_call("Stopping agent agent-y...")
+        mock_print.assert_any_call("Agent agent-y stopped (simulated).")
+
+    @patch('builtins.print')
+    def test_list_agents(self, mock_print):
+        with patch.object(sys, 'argv', ['fleet_manager.py', 'list']):
+            fleet_manager.main()
+        mock_print.assert_any_call("Listing active agents (simulated)...")
+        mock_print.assert_any_call("- agent-1")
+        mock_print.assert_any_call("- agent-2")
+
+if __name__ == '__main__':
+    unittest.main()