Changeset 859 for trunk/Monitoring

Timestamp:

05/20/13 09:13:18 (11 years ago)

Author:

jripsl

Message:

• add killed/segfault detection.
• database communication implementation (suite).
• trap BREAK signal to stop consumer before exiting.

Location:

trunk/Monitoring

Files:

• Analyze/analyze (added)
• Watch/README (added)
• Watch/watch (modified) (8 diffs)
• doc/README (modified) (2 diffs)
• doc/TODO (added)
• smon/repo_io.py (added)
• smon/test_remote_repo.py (added)

trunk/Monitoring/Watch/watch

@@ -13 +13 @@
 ##################################
 
-#from smon import dao
 import pika
 import base64
@@ -25 +24 @@
 
 
-
-
-sys.path.append("/home/jripsl/snapshot/src") 
-import elixir
-import prodiguer_repository
-
-#import prodiguer_repository.operations.mq_hooks_stub as mq_hooks
-import prodiguer_repository.operations.mq_hooks as mq_hooks
-
+from smon import repo_io
 
 
 """
-mq_hooks.retrieve_simulation(name)
-mq_hooks.delete_simulation(name)
-mq_hooks.create_simulation(activity,
-mq_hooks.update_simulation_status(name,
-mq_hooks.retrieve_messages(name)
-mq_hooks.delete_messages(name)
+Code list reminder
+
+0000 (la simulation démarre)
+1000 (le job d'une simulation démarre)
+2000 (PushStack)
+3000 (PopStack OK)
+9000 (PopStack NOK)
+9999 (FATAL)
 """
 
-"""
-from prodiguer_repository.models import (
-        ExecutionState,
-        EXECUTION_STATE_RUNNING,
-        EXECUTION_STATE_SET,
-        Message,
-        Simulation,
-        SIMULATION_SPACE_TEST
-)
-"""
-
-
-
-"""
-scenarios from ticket 136
-
-libIGCM simulent des erreurs (killed, segfault, 1% d'erreur)
-le watcher temps réel change le status des erreurs détéctés
-le watcher temps différé détécte et change le statut des killed 
-"""
-
-class Watcher():
-        # prod
-        #message_code_action_mapping = {"0000":["log","mail"],"1000":["log"],"2000":["log"],"3000":["log"],"9000":["log"],"9999":["log","mail"]}
-
-        # dev
-        message_code_action_mapping = {"0000":["log","print_stdout"],"1000":[],"2000":[],"3000":[],"9000":["log"],"9999":["log"]}
-        #message_code_action_mapping = {"0000":[],"1000":[],"2000":[],"3000":[],"9000":["log"],"9999":["log"]}
-
-        # debug
-        #message_code_action_mapping = {"0000":["print_stdout"],"1000":["log"],"2000":[],"3000":[],"9000":["log"],"9999":["log"]}
-
-        @classmethod
-        def get_fake_progress_messages(cls):
-                pass
-
-        @classmethod
-        def start(cls):
-                #dao.insert_progress_messages(cls.get_fake_progress_messages())
-                pass
-
-        @classmethod
-        def stop(cls):
-                pass
-
-        @classmethod
-        def add(cls,message):
-                pass
+class Mail():
 
         @classmethod
@@ -104 +50 @@
 
         @classmethod
-        def mail(cls):
-                cls.mail_example()
-
-        @classmethod
         def send_mail(cls,me,you,object,body):
                 msg = MIMEText(body)
@@ -119 +61 @@
                 s.quit()
 
+class Actions():
+
+        @classmethod
+        def store_msg(cls):
+                repo_io.store_messages(name)
+
+        @classmethod
+        def set_sim_status_to_error(cls):
+                repo_io.update_simulation_status()
+
+        @classmethod
+        def crea_sim(cls):
+                #repo_io.retrieve_simulation(name)
+                #repo_io.delete_simulation(name)
+                repo_io.create_simulation()
+
+        @classmethod
+        def mail(cls):
+                cls.mail_example()
+
         @classmethod
         def print_stdout(cls,message):
@@ -130 +92 @@
         @classmethod
         def log(cls,message):
-
                 with open("/home/jripsl/supervisor/log/supervisor.log", "a") as log_file:
                         log_file.write("%s %s %s\n"%(datetime.now().strftime('%Y%m%d_%H%M%S'), message["code"],message["jobid"]))
@@ -139 +100 @@
                 message_code=message["code"]
 
-                for action in cls.message_code_action_mapping[message_code]:
+                for action in MessageActionsMapping.mapping[message_code]:
                         proc_name=action
 
                         try:
-                                getattr(cls, proc_name)(message)
+                                getattr(Actions, proc_name)(message)
                         except Exception,e:
                                 traceback.print_exc()
@@ -149 +110 @@
                                 raise Exception("WATCH-ERR002","procedure error (%s,%s)"%(proc_name,str(e)))
 
+class MessageActionsMapping():
+
+        mapping = { "0000":["log", "store_msg", "crea_sim"],
+                                "1000":["log", "store_msg"],
+                                "2000":["log", "store_msg"],
+                                "3000":["log", "store_msg"],
+                                "9000":["log", "store_msg", "mail"],
+                                "9999":["log", "store_msg", "set_sim_status_to_error", "mail"] }
+
+class Watcher():
+
+        @classmethod
+        def start(cls):
+                pass
+
+        @classmethod
+        def stop(cls):
+                pass
+
         @classmethod
         def main(self):
-
-                """
-                _CONNECTION = "postgresql://postgres:Silence107!@localhost:5432/prodiguer" # Repo connection string.
-                prodiguer_repository.connect(_CONNECTION) # Connect to repo.
-                mq_hooks.create_message("test", 2, "bla")
-                elixir.session.commit()
-                #elixir.session.rollback()
-                """
-                mq_hooks.create_message()
 
                 """
@@ -217 +188 @@
 
         signal.signal(signal.SIGINT, signal_handler)
-        #print 'Press Ctrl+C'
 
         try:

trunk/Monitoring/doc/README

@@ -21 +21 @@
                                 - packages below are needed by Prodiguer library
                                         -
+                                                - pip install psycopg2
+                                                        - doesn't work (Error: pg_config executable not found)
+                                                                - solution is
+                                                                        - yum install python-psycopg2.x86_64
+                                                                        - yum install libpqxx-devel
+                                                                        - yum install python-devel
+                                                                        - pip install psycopg2
+                                                                                - works
                                                 - pip install Sqlalchemy
                                                 - pip install elixir
@@ -29 +37 @@
                                                 - SQLAlchemy-0.7.8-py2.6.egg-info
                 - Prodiguer library installation
-                        - svn co svn+ssh://<login here>@forge.ipsl.jussieu.fr/ipsl/forge/projets/prodiguer/svn/sw/src/python/trunk/prodiguer_repository/src
+                        - svn co svn+ssh://<login here>@forge.ipsl.jussieu.fr/ipsl/forge/projets/prodiguer/svn/sw/src/python/trunk/prodiguer_shared/src
                         - edit "watch" script and edit line below accordingly
                                 - sys.path.append("/home/jripsl/snapshot/src")