Wallet config debugging. More.

clore_onboarding.py

@@ -11,6 +11,7 @@ import re
 import os
 import socket
 import asyncio
+import traceback
 from urllib.parse import urlparse
 import subprocess
 from functools import partial
@@ -20,22 +21,34 @@ class logger:
     GREEN = '\033[92m'
     BLUE = '\033[94m'
     RESET = '\033[0m'
+    LOG_FILE = '/opt/clore-hosting/clore_onboarding.log'
     
     @staticmethod
     def _get_current_time():
         return datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
     
+    @staticmethod
+    def _log_to_file(level, message):
+        try:
+            with open(logger.LOG_FILE, 'a') as f:
+                f.write(f"{logger._get_current_time()} | {level} | {message}\n")
+        except Exception:
+            pass
+    
     @staticmethod
     def error(message):
         print(f"{logger.RED}{logger._get_current_time()} | ERROR | {message}{logger.RESET}")
+        logger._log_to_file("ERROR", message)
     
     @staticmethod
     def success(message):
         print(f"{logger.GREEN}{logger._get_current_time()} | SUCCESS | {message}{logger.RESET}")
+        logger._log_to_file("SUCCESS", message)
     
     @staticmethod
     def info(message):
         print(f"{logger.BLUE}{logger._get_current_time()} | INFO | {message}{logger.RESET}")
+        logger._log_to_file("INFO", message)
 
 if os.geteuid() != 0:
     logger.error("This script must be run as root!")
@@ -382,7 +395,13 @@ async def post_request(url, body, headers=None, timeout=15):
         return status_code, response_data
 
     except (http.client.HTTPException, TimeoutError) as e:
-        print(f"Request failed: {e}")
+        logger.error(f"Request failed: {e}")
+        return None, None
+    except socket.gaierror as e:
+        logger.error(f"DNS resolution failed for {url}: {e}")
+        return None, None
+    except Exception as e:
+        logger.error(f"Unexpected error in post_request: {e}")
         return None, None
     finally:
         conn.close()
@@ -439,8 +458,11 @@ if args.write_linux_config:
         sys.exit(1)
 
 async def main(machine_specs):
+    logger.info("Started onboarding")
     global next_retry_reached_server_limit
     last_used_config = None
+    logger.info("logger.info(last_used_config) 1")
+    logger.info(last_used_config)
     ever_pending_creation = False
     machine_id = get_machine_id()
 
@@ -454,9 +476,13 @@ async def main(machine_specs):
         logger.error("Can't load default power limits of nVidia GPU(s)")
         sys.exit(1)
 
+    logger.info("logger.info(last_used_config) 2")
+    logger.info(last_used_config)
+
     oc_config = {}
     while True:
         try:
+            logger.info("Looping")
             if args.mode == "linux":
                 clore_config = await async_read_file(clore_conf_path)
                 clore_config = json.loads(clore_config)
@@ -466,7 +492,10 @@ async def main(machine_specs):
                 machine_name, clore_config, oc_config = await hive_load_configs(default_power_limits, static_clore_config)
                 #print(f"Machine Name: {machine_name}")
 
+            logger.info(args.mode)
+
             config_validation = validate_clore_config(clore_config)
+
             if config_validation == "Validation successful":
                 if "save_config" in clore_config and args.mode == "hive":
                     verify_or_update_file(clore_conf_path, json.dumps(clore_config))
@@ -475,6 +504,12 @@ async def main(machine_specs):
                         clore_config["clear_oc_override"] = True
                     else:
                         clore_config["stock_oc_override"] = oc_config
+
+                logger.info("Is config different")        
+                logger.info(clore_config != last_used_config)
+                logger.info(clore_config)
+                logger.info(last_used_config)
+
                 if clore_config != last_used_config or (time.time() > next_retry_reached_server_limit and next_retry_reached_server_limit > 0):
                     last_used_config = clore_config.copy()
                     if type(clore_config) == dict and "hostname_override" in clore_config:
@@ -521,10 +556,13 @@ async def main(machine_specs):
                 logger.error(f"Could not parse config - {' | '.join(config_validation)}")
                 
         except Exception as e:
-            print(e)
-            last_used_config = None
+            logger.error(f"Exception: {e}")
+            logger.error(f"Traceback: {traceback.format_exc()}")
         await asyncio.sleep(5)
 
+        logger.info("logger.info(last_used_config) 3")
+        logger.info(last_used_config)
+
 if __name__ == "__main__":
     machine_specs = specs.get(benchmark_disk=True, mock=args.mock)
     asyncio.run(main(machine_specs))