Changeset 876 for trunk/Monitoring/Analyze/analyze

Timestamp:

06/12/13 11:14:57 (11 years ago)

Author:

jripsl

Message:

Timeout test impl.
Add timestamp.

File:

• trunk/Monitoring/Analyze/analyze (modified) (7 diffs)

trunk/Monitoring/Analyze/analyze

@@ -21 +21 @@
 from datetime import datetime
 
-from smon import repo_io
+
+# line below is to include "smon" package in the search path
+sys.path.append("/home/jripsl/snapshot/Monitoring")
+
+import smon.repo_io as repo_io
+import smon.types as types
 
 
-
-"""
-Code list reminder
-
-0000 (la simulation démarre)
-1000 (le job d'une simulation démarre)
-2000 (PushStack)
-3000 (PopStack OK)
-9000 (PopStack NOK)
-9999 (FATAL)
-"""
 
 class CheckList():
@@ -52 +46 @@
                 diff=current_time-msg_time
 
+                # debug
+                print "diff=%s"%diff
 
                 if diff>cls.max_time_between_msg:
@@ -68 +64 @@
                   changing the simulation status 
                 """
-                running_simulations=repo_io.get_running_simulations()
-                for simulation in running_simulations:
-                        message=repo_io.retrieve_last_messages(simulation)
+
+
+                for simulation in repo_io.get_running_simulations():
+
+                        print "checking heartbeat for '%s'"%simulation.name
+
+                        try:
+
+                                message=repo_io.retrieve_last_message(simulation)
+                        except types.MessageNotFoundException, e:
+
+                                continue
+
 
                         if msg_timeout(message):
 
-                                repo_io.update_simulation_status(simulation) #from "running" to "killed"
+                                simulation.status="error"
+
+                                repo_io.update_simulation_status(simulation)
+
+                                print "heartbeat NOK (simulation status set to 'error')"%simulation.name
+
+
+                        else:
+                                print "heartbeat OK"%simulation.name
 
 class Analyzer():
@@ -80 +94 @@
         @classmethod
         def start(cls):
-                pass
+                repo_io.init() # open DB connection
+
+                Analyzer.main()
 
         @classmethod
         def stop(cls):
-                pass
+                repo_io.free() # close DB connection
 
         @classmethod
@@ -104 +120 @@
                 print ' [*] Analyzer running. To exit press CTRL+C'
 
-                CheckList.C0001()
+                while True:
+
+                        print "checking simulations heartbeats"
+
+                        CheckList.C0001()
 
 
-                sleep(10)
+                        time.sleep(1)
 
                 """
@@ -115 +135 @@
 def signal_handler(signal, frame):
                 print 'You pressed Ctrl+C!'
+
+                Analyzer.stop()
+
                 sys.exit(0)
 
@@ -122 +145 @@
 
         try:
-                Analyzer.main()
+                Analyzer.start()
 
                 sys.exit(0)