Merge pull request #651 from Alignak-monitoring/daemons-name

Daemons process title name

alignak/daemon.py

@@ -162,6 +162,8 @@ class Daemon(object):
     """
 
     properties = {
+        'daemon_type':
+            StringProp(default='unknown'),
         # workdir is relative to $(dirname "$0"/..)
         # where "$0" is the path of the file being executed,
         # in python normally known as:
@@ -435,12 +437,12 @@ def dump_memory():
         except ImportError:
             logger.warning('I do not have the module guppy for memory dump, please install it')
 
-    def load_modules_manager(self):
+    def load_modules_manager(self, daemon_name):
         """Instantiate Modulesmanager and load the SyncManager (multiprocessing)
 
         :return: None
         """
-        self.modules_manager = ModulesManager(self.name, self.sync_manager,
+        self.modules_manager = ModulesManager(daemon_name, self.sync_manager,
                                               max_queue_size=getattr(self, 'max_queue_size', 0))
 
     def change_to_workdir(self):
@@ -686,7 +688,7 @@ def _create_manager():
         manager.start()
         return manager
 
-    def do_daemon_init_and_start(self, daemon_name=None):
+    def do_daemon_init_and_start(self):
         """Main daemon function.
         Clean, allocates, initializes and starts all necessary resources to go in daemon mode.
 
@@ -695,7 +697,7 @@ def do_daemon_init_and_start(self, daemon_name=None):
         :type daemon_name: str
         :return: False if the HTTP daemon can not be initialized, else True
         """
-        self.set_proctitle(daemon_name)
+        self.set_proctitle()
         self.change_to_user_group()
         self.change_to_workdir()
         self.check_parallel_run()
@@ -999,6 +1001,7 @@ def set_exit_handler(self):
                         signal.SIGUSR2, signal.SIGHUP):
                 signal.signal(sig, func)
 
+    # pylint: disable=no-member
     def set_proctitle(self, daemon_name=None):
         """Set the proctitle of the daemon
 
@@ -1008,9 +1011,11 @@ def set_proctitle(self, daemon_name=None):
         :return: None
         """
         if daemon_name:
-            setproctitle("alignak-%s %s" % (self.name, daemon_name))
+            setproctitle("alignak-%s %s" % (self.daemon_type, daemon_name))
+            if hasattr(self, 'modules_manager'):
+                self.modules_manager.set_daemon_name(daemon_name)
         else:
-            setproctitle("alignak-%s" % self.name)
+            setproctitle("alignak-%s" % self.daemon_type)
 
     def get_header(self):
         """ Get the log file header

alignak/daemons/arbiterdaemon.py

@@ -75,7 +75,7 @@
 from alignak.stats import statsmgr
 from alignak.brok import Brok
 from alignak.external_command import ExternalCommand
-from alignak.property import BoolProp, PathProp, IntegerProp
+from alignak.property import BoolProp, PathProp, IntegerProp, StringProp
 from alignak.http.arbiter_interface import ArbiterInterface
 
 logger = logging.getLogger(__name__)  # pylint: disable=C0103
@@ -90,6 +90,8 @@ class Arbiter(Daemon):  # pylint: disable=R0902
     """
     properties = Daemon.properties.copy()
     properties.update({
+        'daemon_type':
+            StringProp(default='arbiter'),
         'pidfile':
             PathProp(default='arbiterd.pid'),
         'port':
@@ -285,8 +287,11 @@ def load_monitoring_config_file(self):  # pylint: disable=R0915
                      "with the value '%s'."
                      " Thanks." % (self.config_name, socket.gethostname()))
 
+        # Set my own process title
+        self.set_proctitle(self.myself.get_name())
+
         # Ok it's time to load the module manager now!
-        self.load_modules_manager()
+        self.load_modules_manager(self.myself.get_name())
         # we request the instances without them being *started*
         # (for those that are concerned ("external" modules):
         # we will *start* these instances after we have been daemonized (if requested)
@@ -540,6 +545,9 @@ def main(self):
             self.look_for_early_exit()
             self.do_daemon_init_and_start()
 
+            # Set my own process title
+            self.set_proctitle(self.myself.get_name())
+
             # ok we are now fully daemonized (if requested)
             # now we can start our "external" modules (if any):
             self.modules_manager.start_external_instances()

alignak/daemons/brokerdaemon.py

@@ -71,7 +71,7 @@
 from alignak.objects import *
 from alignak.misc.serialization import unserialize, AlignakClassLookupException
 from alignak.satellite import BaseSatellite
-from alignak.property import PathProp, IntegerProp
+from alignak.property import PathProp, IntegerProp, StringProp
 from alignak.util import sort_by_ids
 from alignak.stats import statsmgr
 from alignak.http.client import HTTPClient, HTTPEXCEPTIONS
@@ -88,9 +88,14 @@ class Broker(BaseSatellite):
     """
     properties = BaseSatellite.properties.copy()
     properties.update({
-        'pidfile':   PathProp(default='brokerd.pid'),
-        'port':      IntegerProp(default=7772),
-        'local_log': PathProp(default='brokerd.log'),
+        'daemon_type':
+            StringProp(default='broker'),
+        'pidfile':
+            PathProp(default='brokerd.pid'),
+        'port':
+            IntegerProp(default=7772),
+        'local_log':
+            PathProp(default='brokerd.log'),
     })
 
     def __init__(self, config_file, is_daemon, do_replace, debug, debug_file):
@@ -323,12 +328,10 @@ def manage_brok(self, brok):
         for mod in self.modules_manager.get_internal_instances():
             try:
                 mod.manage_brok(brok)
-            except Exception, exp:  # pylint: disable=W0703
-                logger.debug(str(exp.__dict__))
+            except Exception as exp:  # pylint: disable=broad-except
                 logger.warning("The mod %s raise an exception: %s, I'm tagging it to restart later",
                                mod.get_name(), str(exp))
-                logger.warning("Exception type: %s", type(exp))
-                logger.warning("Back trace of this kill: %s", traceback.format_exc())
+                logger.exception(exp)
                 self.modules_manager.set_to_restart(mod)
 
     def add_broks_to_queue(self, broks):
@@ -397,23 +400,24 @@ def get_new_broks(self, i_type='scheduler'):
                 else:  # no con? make the connection
                     self.pynag_con_init(sched_id, i_type=i_type)
             # Ok, con is not known, so we create it
-            except KeyError, exp:
+            except KeyError as exp:
                 logger.debug("Key error for get_broks : %s", str(exp))
                 self.pynag_con_init(sched_id, i_type=i_type)
-            except HTTPEXCEPTIONS, exp:
+            except HTTPEXCEPTIONS as exp:
                 logger.warning("Connection problem to the %s %s: %s",
                                i_type, links[sched_id]['name'], str(exp))
+                logger.exception(exp)
                 links[sched_id]['con'] = None
             # scheduler must not #be initialized
-            except AttributeError, exp:
+            except AttributeError as exp:
                 logger.warning("The %s %s should not be initialized: %s",
                                i_type, links[sched_id]['name'], str(exp))
+                logger.exception(exp)
             # scheduler must not have checks
             #  What the F**k? We do not know what happened,
             # so.. bye bye :)
-            except Exception, err:  # pylint: disable=W0703
-                logger.error(str(err))
-                logger.error(traceback.format_exc())
+            except Exception as exp:  # pylint: disable=broad-except
+                logger.exception(exp)
                 sys.exit(1)
 
     def get_retention_data(self):
@@ -449,6 +453,7 @@ def setup_new_conf(self):  # pylint: disable=R0915,R0912
 
         :return: None
         """
+
         with self.conf_lock:
             conf = unserialize(self.new_conf, True)
             self.new_conf = None
@@ -475,9 +480,8 @@ def setup_new_conf(self):  # pylint: disable=R0915,R0912
 
             logger.debug("[%s] Sending us configuration %s", self.name, conf)
 
-            # If we've got something in the schedulers, we do not
-            # want it anymore
-            # self.schedulers.clear()
+            # Get our Schedulers
+            logger.info("[%s] schedulers: %s", self.name, conf['schedulers'])
             for sched_id in conf['schedulers']:
                 # Must look if we already have it to do not overdie our broks
 
@@ -519,6 +523,7 @@ def setup_new_conf(self):  # pylint: disable=R0915,R0912
                 logger.info(" - %s ", daemon['name'])
 
             # Now get arbiter
+            logger.info("[%s] arbiters: %s", self.name, conf['arbiters'])
             for arb_id in conf['arbiters']:
                 # Must look if we already have it
                 already_got = arb_id in self.arbiters
@@ -553,6 +558,7 @@ def setup_new_conf(self):  # pylint: disable=R0915,R0912
                 logger.info(" - %s ", daemon['name'])
 
             # Now for pollers
+            logger.info("[%s] pollers: %s", self.name, conf['pollers'])
             for pol_id in conf['pollers']:
                 # Must look if we already have it
                 already_got = pol_id in self.pollers
@@ -588,6 +594,7 @@ def setup_new_conf(self):  # pylint: disable=R0915,R0912
                 logger.info(" - %s ", daemon['name'])
 
             # Now reactionners
+            logger.info("[%s] reactionners: %s", self.name, conf['reactionners'])
             for rea_id in conf['reactionners']:
                 # Must look if we already have it
                 already_got = rea_id in self.reactionners
@@ -623,6 +630,7 @@ def setup_new_conf(self):  # pylint: disable=R0915,R0912
                 logger.info(" - %s ", daemon['name'])
 
             # Now receivers
+            logger.debug("[%s] receivers: %s", self.name, conf['receivers'])
             for rec_id in conf['receivers']:
                 # Must look if we already have it
                 already_got = rec_id in self.receivers
@@ -673,7 +681,7 @@ def setup_new_conf(self):  # pylint: disable=R0915,R0912
                 os.environ['TZ'] = use_timezone
                 time.tzset()
 
-            # Connection init with Schedulers
+            # Initialize connection with Schedulers, Pollers and Reactionners
             for sched_id in self.schedulers:
                 self.pynag_con_init(sched_id, i_type='scheduler')
 
@@ -792,14 +800,12 @@ def do_loop_turn(self):
         for mod in ext_modules:
             try:
                 mod.to_q.put(to_send)
-            except Exception, exp:  # pylint: disable=W0703
+            except Exception as exp:  # pylint: disable=broad-except
                 # first we must find the modules
-                logger.debug(str(exp.__dict__))
                 logger.warning("The mod %s queue raise an exception: %s, "
                                "I'm tagging it to restart later",
                                mod.get_name(), str(exp))
-                logger.warning("Exception type: %s", type(exp))
-                logger.warning("Back trace of this kill: %s", traceback.format_exc())
+                logger.exception(exp)
                 self.modules_manager.set_to_restart(mod)
 
         # No more need to send them
@@ -853,8 +859,6 @@ def do_loop_turn(self):
                 self.timeout = self.timeout - (end - begin)
             self.timeout = 1.0
 
-            # print "get new broks watch new conf 1: end", len(self.broks)
-
         # Say to modules it's a new tick :)
         self.hook_point('tick')
 
@@ -873,7 +877,7 @@ def main(self):
 
             self.do_daemon_init_and_start()
 
-            self.load_modules_manager()
+            self.load_modules_manager(self.name)
 
             #  We wait for initial conf
             self.wait_for_initial_conf()

alignak/daemons/pollerdaemon.py

@@ -47,7 +47,7 @@
 This modules provides class for the Poller daemon
 """
 from alignak.satellite import Satellite
-from alignak.property import PathProp, IntegerProp
+from alignak.property import PathProp, IntegerProp, StringProp
 
 
 class Poller(Satellite):
@@ -60,9 +60,14 @@ class Poller(Satellite):
 
     properties = Satellite.properties.copy()
     properties.update({
-        'pidfile':   PathProp(default='pollerd.pid'),
-        'port':      IntegerProp(default=7771),
-        'local_log': PathProp(default='pollerd.log'),
+        'daemon_type':
+            StringProp(default='poller'),
+        'pidfile':
+            PathProp(default='pollerd.pid'),
+        'port':
+            IntegerProp(default=7771),
+        'local_log':
+            PathProp(default='pollerd.log'),
     })
 
     def __init__(self, config_file, is_daemon, do_replace, debug, debug_file):

alignak/daemons/reactionnerdaemon.py

@@ -50,7 +50,7 @@
 """
 
 from alignak.satellite import Satellite
-from alignak.property import PathProp, IntegerProp
+from alignak.property import PathProp, IntegerProp, StringProp
 
 
 class Reactionner(Satellite):
@@ -73,9 +73,14 @@ class Reactionner(Satellite):
 
     properties = Satellite.properties.copy()
     properties.update({
-        'pidfile':   PathProp(default='reactionnerd.pid'),
-        'port':      IntegerProp(default=7769),
-        'local_log': PathProp(default='reactionnerd.log'),
+        'daemon_type':
+            StringProp(default='reactionner'),
+        'pidfile':
+            PathProp(default='reactionnerd.pid'),
+        'port':
+            IntegerProp(default=7769),
+        'local_log':
+            PathProp(default='reactionnerd.log'),
     })
 
     def __init__(self, config_file, is_daemon, do_replace, debug, debug_file):

alignak/daemons/receiverdaemon.py

@@ -59,7 +59,7 @@
 
 from alignak.misc.serialization import unserialize
 from alignak.satellite import Satellite
-from alignak.property import PathProp, IntegerProp
+from alignak.property import PathProp, IntegerProp, StringProp
 from alignak.external_command import ExternalCommand, ExternalCommandManager
 from alignak.http.client import HTTPEXCEPTIONS
 from alignak.stats import statsmgr
@@ -76,6 +76,8 @@ class Receiver(Satellite):
 
     properties = Satellite.properties.copy()
     properties.update({
+        'daemon_type':
+            StringProp(default='receiver'),
         'pidfile':
             PathProp(default='receiverd.pid'),
         'port':
@@ -409,7 +411,7 @@ def main(self):
 
             self.do_daemon_init_and_start()
 
-            self.load_modules_manager()
+            self.load_modules_manager(self.name)
 
             #  We wait for initial conf
             self.wait_for_initial_conf()

alignak/daemons/schedulerdaemon.py

@@ -65,7 +65,7 @@
 from alignak.external_command import ExternalCommandManager
 from alignak.daemon import Daemon
 from alignak.http.scheduler_interface import SchedulerInterface
-from alignak.property import PathProp, IntegerProp
+from alignak.property import PathProp, IntegerProp, StringProp
 from alignak.satellite import BaseSatellite
 from alignak.stats import statsmgr
 
@@ -79,6 +79,8 @@ class Alignak(BaseSatellite):
 
     properties = BaseSatellite.properties.copy()
     properties.update({
+        'daemon_type':
+            StringProp(default='scheduler'),
         'pidfile':
             PathProp(default='schedulerd.pid'),
         'port':
@@ -252,6 +254,15 @@ def setup_new_conf(self):
             logger.debug("Conf received at %d. Un-serialized in %d secs", t00, time.time() - t00)
             self.new_conf = None
 
+            if 'scheduler_name' in new_c:
+                name = new_c['scheduler_name']
+            else:
+                name = instance_name
+            self.name = name
+
+            # Set my own process title
+            self.set_proctitle(self.name)
+
             # Tag the conf with our data
             self.conf = conf
             self.conf.push_flavor = new_c['push_flavor']
@@ -375,7 +386,7 @@ def main(self):
 
             self.do_daemon_init_and_start()
 
-            self.load_modules_manager()
+            self.load_modules_manager(self.name)
 
             self.uri = self.http_daemon.uri
             logger.info("[Scheduler] General interface is at: %s", self.uri)