Fixed issue with updates after builds and restarting process

agent_manager.py

@@ -15,6 +15,8 @@ import requests
 from build_agent import build_agent_version
 from zeroconf_server import ZeroconfServer
 
+TIMEOUT = 10
+
 def find_server_ips():
     ZeroconfServer.configure("_crosspybuilder._tcp.local.", socket.gethostname(), 9001)
     hostnames = []
@@ -47,12 +49,13 @@ def get_all_servers_status():
 def get_worker_status(hostname):
     """Fetch worker status from the given hostname."""
     try:
-        response = requests.get(f"http://{hostname}:9001/status", timeout=5)
+        response = requests.get(f"http://{hostname}:9001/status", timeout=TIMEOUT)
         status = response.json()
         if status['hostname'] != hostname and status['ip'] != hostname:
             status['ip'] = socket.gethostbyname(hostname)
         return status
-    except requests.exceptions.RequestException:
+    except requests.exceptions.RequestException as e:
+        print(f"EXCEPTION: {e}")
         return {"hostname": hostname, "status": "offline"}
 
 
@@ -71,8 +74,7 @@ def zip_project(source_dir, output_zip):
 def send_build_request(zip_file, server_ip, download_after=False):
     """Uploads the zip file to the given server."""
     upload_url = f"http://{server_ip}:9001/upload"
-    print(f"Submitting build request to URL: {upload_url}")
-    print("Please wait. This may take a few minutes...")
+    print(f"Submitting build request to URL: {upload_url} - Please wait. This may take a few minutes...")
     with open(zip_file, 'rb') as f:
         response = requests.post(upload_url, files={"file": f})
 
@@ -189,7 +191,7 @@ def update_worker(server):
         response_json = response.json()
         if response_json.get('updated_files') and not response_json.get('error_files'):
             try:
-                requests.get(f"http://{server['ip']}:9001/restart", timeout=2)
+                requests.get(f"http://{server['ip']}:9001/restart", timeout=TIMEOUT)
             except requests.exceptions.ConnectionError:
                 pass
             return server
@@ -223,7 +225,7 @@ def update_build_workers(server_data):
         while unverified_servers and datetime.now() < end_time:
             for server_ip in list(unverified_servers.keys()):  # Iterate over a copy to avoid modification issues
                 try:
-                    response = requests.get(f"http://{server_ip}:9001/status", timeout=2)
+                    response = requests.get(f"http://{server_ip}:9001/status")
                     response.raise_for_status()
                     server_info = unverified_servers[server_ip]  # Get full server details
                     agent_version = response.json().get('agent_version')
@@ -254,7 +256,7 @@ def main():
     parser.add_argument("-restart",  type=str, help="Hostname to restart")
     parser.add_argument("-restart-all",  action="store_true", help="Restart all agents")
     parser.add_argument("-shutdown",  type=str, help="Hostname to shutdown")
-    # parser.add_argument("-worker",  type=str, help="Update Agents")
+    parser.add_argument("-shutdown-all",  action="store_true", help="Restart all agents")
     args = parser.parse_args()
 
     if args.status:
@@ -267,11 +269,11 @@ def main():
         for server_ip in find_server_ips():
             restart_agent(server_ip)
     elif args.shutdown:
-        print(f"Shutting down hostname: {args.shutdown}")
-        try:
-            requests.get(f"http://{args.shutdown}:9001/shutdown")
-        except requests.exceptions.ConnectionError:
-            pass
+        shutdown_agent(args.shutdown)
+    elif args.shutdown_all:
+        print("Shutting down all agents...")
+        for server_ip in find_server_ips():
+            shutdown_agent(server_ip)
     elif args.delete_cache:
         delete_cache(get_all_servers_status())
     elif args.path:
@@ -281,12 +283,18 @@ def main():
     else:
         print("No path given!")
 
+def shutdown_agent(hostname):
+    print(f"Shutting down hostname: {hostname}")
+    try:
+        requests.get(f"http://{hostname}:9001/shutdown", timeout=TIMEOUT)
+    except (requests.exceptions.ConnectionError, TimeoutError):
+        pass
 
 def restart_agent(hostname):
     print(f"Restarting agent: {hostname}")
     try:
-        requests.get(f"http://{hostname}:9001/restart")
-    except requests.exceptions.ConnectionError:
+        requests.get(f"http://{hostname}:9001/restart", timeout=TIMEOUT)
+    except (requests.exceptions.ConnectionError, TimeoutError):
         pass