[Mellanox][Smartswitch]Added debug logs to reboot and admin state change (#24263)

[Mellanox][Smartswitch]Added debug logs to reboot and admin state change

Author: gpunathilell

platform/mellanox/mlnx-platform-api/sonic_platform/dpuctlplat.py

@@ -24,6 +24,7 @@
 from contextlib import contextmanager
 from select import poll, POLLPRI, POLLIN
 from enum import Enum
+import signal
 
 try:
     from .inotify_helper import InotifyHelper
@@ -107,7 +108,7 @@ def __init__(self, dpu_name):
         self.pci_dev_path = []
         self.verbosity = False
 
-    def setup_logger(self, use_print=False):
+    def setup_logger(self, use_print=False, use_notice_level=False):
         def print_with_time(msg):
             timestamp = time.strftime("%Y-%m-%d %H:%M:%S")
             print(f"[{timestamp}] {msg}")
@@ -118,8 +119,12 @@ def print_with_time(msg):
             self.logger_warning = print_with_time
             self.logger_debug = print_with_time
             return
-        self.logger_debug = logger.log_debug
-        self.logger_info = logger.log_info
+        if use_notice_level:
+            self.logger_debug = logger.log_notice
+            self.logger_info = logger.log_notice
+        else:
+            self.logger_debug = logger.log_debug
+            self.logger_info = logger.log_info
         self.logger_error = logger.log_error
         self.logger_warning = logger.log_warning
 
@@ -414,20 +419,32 @@ def update_boot_prog_once(self, poll_var):
         read_value = self.read_boot_prog()
         if read_value != self.boot_prog_state:
             self.dpu_boot_prog_update(read_value)
-            self.log_error(f"The boot_progress status is changed to = {self.boot_prog_indication}")
+            self.log_info(f"The boot_progress status is changed to = {self.boot_prog_indication}")
 
     def watch_boot_prog(self):
         """Read boot_progress and update the value in an infinite loop"""
+        def signal_handler(signum, frame):
+            self.log_info("Received termination signal, shutting down...")
+            raise SystemExit("Terminated by signal")
+        
+        # Register signal handler for SIGTERM
+        signal.signal(signal.SIGTERM, signal_handler)
+        
+        file = None
+        file = open(self.boot_prog_path, "r")
+        p = poll()
+        p.register(file.fileno(), POLLPRI)
         try:
-            self.dpu_boot_prog_update()
-            self.log_info(f"The initial boot_progress status is = {self.boot_prog_indication}")
-            file = open(self.boot_prog_path, "r")
-            p = poll()
-            p.register(file.fileno(), POLLPRI)
             while True:
-                self.update_boot_prog_once(p)
-        except Exception:
-            self.log_error(f"Exception occured during watch_boot_progress!")
+                try:
+                    self.update_boot_prog_once(p)
+                except SystemExit:
+                    break  # Exit on termination signal
+        except Exception as e:
+            self.log_error(f"Error during watch_boot_progress: {e}")
+        finally:
+            if file:
+                file.close()
 
     @contextmanager
     def boot_prog_context(self):
@@ -444,6 +461,8 @@ def boot_prog_context(self):
             finally:
                 if self.boot_prog_proc and self.boot_prog_proc.is_alive():
                     self.boot_prog_proc.terminate()
+                    self.boot_prog_proc.join(timeout=3)
+                    self.boot_prog_proc.kill()
                     self.boot_prog_proc.join()
         else:
             yield

platform/mellanox/mlnx-platform-api/sonic_platform/module.py

@@ -262,6 +262,8 @@ def __init__(self, dpu_id):
         self.dpu_id = dpu_id
         self._name = f"DPU{self.dpu_id}"
         self.dpuctl_obj = DpuCtlPlat(self._name.lower())
+        self.dpuctl_obj.setup_logger(use_notice_level=True)
+        self.dpuctl_obj.verbosity = True
         self.fault_state = False
         self.dpu_vpd_parser = DpuVpdParser('/var/run/hw-management/eeprom/vpd_data', self.dpuctl_obj._name.upper())
         self.CONFIG_DB_NAME = "CONFIG_DB"
@@ -334,8 +336,11 @@ def reboot(self, reboot_type=ModuleBase.MODULE_REBOOT_DPU):
         Returns:
             bool: True if the request has been issued successfully, False if not
         """
+        logger.log_notice(f"Rebooting {self._name} with type {reboot_type}")
         # no_wait=True is not supported at this point, because of race conditions with other drivers
-        return self.dpuctl_obj.dpu_reboot(skip_pre_post=True)
+        return_value = self.dpuctl_obj.dpu_reboot(skip_pre_post=True)
+        logger.log_notice(f"Rebooted {self._name} with type {reboot_type} and return value {return_value}")
+        return return_value
 
     def set_admin_state(self, up):
         """
@@ -352,12 +357,16 @@ def set_admin_state(self, up):
         Returns:
             bool: True if the request has been issued successfully, False if not
         """
+        logger.log_notice(f"Setting the admin state for {self._name} to {up}")
         if up:
             if self.dpuctl_obj.dpu_power_on(skip_pre_post=True):
+                logger.log_notice(f"Completed the admin state change for {self._name} to {up}")
                 return True
             logger.log_error(f"Failed to set the admin state for {self._name}")
             return False
-        return self.dpuctl_obj.dpu_power_off(skip_pre_post=True)
+        return_value = self.dpuctl_obj.dpu_power_off(skip_pre_post=True)
+        logger.log_notice(f"Completed the admin state change for {self._name} to {up}")
+        return return_value
 
     def get_type(self):
         """