hw-mgmt: scripts: Fix peripheral/thermal updater service crash on exit.

Sometimes, on peripheral/thermal updater service can be failed because
of race condition in logging subsystem. It happens on log emit in signal
handler.

Fix: move all log from sig handler to main loop

Bug: 4833200

Signed-off-by: Oleksandr Shamray <oleksandrs@nvidia.com>

Author: sholeksandr

usr/usr/bin/hw_management_peripheral_updater.py

@@ -87,6 +87,7 @@ class CONST(object):
 
 
 EXIT = threading.Event()
+_sig_condition_name = ""
 
 
 def _build_attrib_list():
@@ -569,10 +570,13 @@ def handle_shutdown(sig, _frame):
     @param sig: Signal number
     @param _frame: Unused frame
     """
-    EXIT.set()
-    LOGGER.notice("hw-management-peripheral-updater: received signal {}, stopping main loop".format(sig))
+    global _sig_condition_name
+    try:
+        _sig_condition_name = signal.Signals(sig).name
+    except (ValueError, AttributeError):
+        _sig_condition_name = str(sig)
 
-    return
+    EXIT.set()
 
 
 def main():
@@ -661,7 +665,7 @@ def main():
 
         EXIT.wait(timeout=1)
 
-    LOGGER.notice("hw-management-peripheral-updater: stopped main loop")
+    LOGGER.notice("hw-management-peripheral-updater: stopped main loop ({})".format(_sig_condition_name))
     return
 
 

usr/usr/bin/hw_management_thermal_updater.py

@@ -153,6 +153,7 @@ def _build_thermal_config():
 LOGGER = None
 
 EXIT = threading.Event()
+_sig_condition_name = ""
 
 # ----------------------------------------------------------------------
 
@@ -483,10 +484,12 @@ def handle_shutdown(sig, _frame):
     @param sig: Signal
     @param _frame: Unused frame
     """
+    global _sig_condition_name
+    try:
+        _sig_condition_name = signal.Signals(sig).name
+    except (ValueError, AttributeError):
+        _sig_condition_name = str(sig)
     EXIT.set()
-    LOGGER.notice("hw-management-thermal-updater: received signal {}, stopping main loop".format(sig))
-
-    return
 
 # ----------------------------------------------------------------------
 
@@ -569,7 +572,7 @@ def main():
 
         EXIT.wait(timeout=1)
 
-    LOGGER.notice("hw-management-thermal-updater: stopped main loop")
+    LOGGER.notice("hw-management-thermal-updater: stopped main loop ({})".format(_sig_condition_name))
 
 
 if __name__ == '__main__':