WIP

slapos/promise/plugin/check_cpu_temperature.py

@@ -13,68 +13,60 @@ from slapos.grid.promise import interface
 class RunPromise(JSONRunPromise):
 
   def __init__(self, config):
-
     super(RunPromise, self).__init__(config)
-
     self.setPeriodicity(minute=2)
-    self.last_avg_computation_file = self.getConfig(
-      'last-avg-computation-file', 'last_avg')
-
-  def sense(self):
-
-    promise_success = True
-
-    max_spot_temp = float(self.getConfig('max-spot-temp', 90))
-    max_avg_temp = float(self.getConfig('max-avg-temp', 80))
+    self.avg_flag_file = self.getConfig('last-avg-computation-file', 'last_avg')
+    self.max_spot_temp = float(self.getConfig('max-spot-temp', 90))
+    self.max_avg_temp = float(self.getConfig('max-avg-temp', 80))
     avg_temp_duration_sec = int(self.getConfig('avg-temp-duration-sec', 0))
     if avg_temp_duration_sec:
-      avg_temp_duration = avg_temp_duration_sec
+      self.avg_temp_duration = avg_temp_duration_sec
     else:
-      avg_temp_duration = 60 * int(self.getConfig('avg-temp-duration', 5))
-    testing = self.getConfig('testing') == "True"
+      self.avg_temp_duration = 60 * int(self.getConfig('avg-temp-duration', 5))
 
-    # For theia JHGD
-    #testing = True # JHGD
+  def sense(self):
+    success = True
 
     # Get current temperature
-    if testing:
-      from random import randint
-      cpu_temp = randint(40, 75)
-    else:
-      data = psutil.sensors_temperatures()
-      cpu_temp = data['coretemp'][0][1]
-    if cpu_temp > max_spot_temp:
-      self.logger.error("Temperature reached critical threshold: %s degrees "\
-        "celsius (threshold is %s degrees celsius)" % (cpu_temp, max_spot_temp))
-      promise_success = False
+    try:
+      cpu_temp = psutil.sensors_temperatures()['coretemp'][0][1]
+    except (KeyError, IndexError) as e:
+      self.logger.error("Could not retrieve core temperature:\n%r", e, exc_info=True)
+      return
+    if cpu_temp > iself.max_spot_temp:
+      success = False
+      self.logger.error(
+        "Temperature reached critical threshold: %s °C (threshold is %s °C)",
+        cpu_temp, max_spot_temp)
 
     # Log temperature
     data = json.dumps({'cpu_temperature': cpu_temp})
     self.json_logger.info("Temperature data", extra={'data': data})
 
-		# TODO: promise should computer average only with logs between interval
+    # TODO: promise should compute average only with logs between interval
     # Computer average temperature
-    avg_computation_period = avg_temp_duration / 4
+    avg_computation_period = self.avg_temp_duration / 4
     try:
-      t = os.path.getmtime(self.last_avg_computation_file)
+      t = os.path.getmtime(self.avg_flag_file)
     except OSError:
       t = 0
     if (time.time() - t) > avg_computation_period:
-      open(self.last_avg_computation_file, 'w').close()
-      temp_list = get_data_interval_json_log(self.log_file, avg_temp_duration)
+      open(self.avg_flag_file, 'w').close()
+      temp_list = get_data_interval_json_log(self.log_file, self.avg_temp_duration)
       if temp_list:
         avg_temp = sum(map(lambda x: x['cpu_temperature'], temp_list)) / len(temp_list)
-        if avg_temp > max_avg_temp:
-          self.logger.error("Average temperature over the last %s seconds "\
-            "reached threshold: %s degrees celsius (threshold is %s degrees "\
-            "celsius)" % (avg_temp_duration, avg_temp, max_avg_temp))
-          promise_success = False
+        if avg_temp > self.max_avg_temp:
+          success = False
+          self.logger.error(
+            "Average temperature over the last %d s reached threshold: %s °C"
+            " (threshold is %s °C)",
+            avg_temp_duration, avg_temp, max_avg_temp)
       else:
+        success = False
         self.logger.error("Couldn't read temperature from log")
-        promise_success = False
 
-    if promise_success:
-      self.logger.info("Temperature OK")
+    if success:
+      self.logger.info("Temperature OK (%s °C)", cpu_temp)
 
   def test(self):
     """