Emails were only being sent if high bandwidth caps were exceeded. Fixed.

[nodemanager.git] / nm.py

diff --git a/nm.py b/nm.py
old mode 100644 (file)
new mode 100755 (executable)
index c4760fc..a2b6a4d
--- a/nm.py
+++ b/nm.py
@@ -1,5 +1,4 @@
 #!/usr/bin/python
-
 """Node Manager"""
 
 import optparse
@@ -15,34 +14,55 @@ import tools
 
 from config import Config
 from plcapi import PLCAPI 
+import random
+import net
 
-
+id="$Id$"
 savedargv = sys.argv[:]
 
+known_modules=['conf_files', 'sm', 'bwmon', 'vsys', 'codemux']
+
 parser = optparse.OptionParser()
 parser.add_option('-d', '--daemon', action='store_true', dest='daemon', default=False, help='run daemonized')
 parser.add_option('-s', '--startup', action='store_true', dest='startup', default=False, help='run all sliver startup scripts')
 parser.add_option('-f', '--config', action='store', dest='config', default='/etc/planetlab/plc_config', help='PLC configuration file')
 parser.add_option('-k', '--session', action='store', dest='session', default='/etc/planetlab/session', help='API session key (or file)')
 parser.add_option('-p', '--period', action='store', dest='period', default=600, help='Polling interval (sec)')
+parser.add_option('-r', '--random', action='store', dest='random', default=301, help='Range for additional random polling interval (sec)')
+parser.add_option('-v', '--verbose', action='store_true', dest='verbose', default=False, help='more verbose log')
+parser.add_option('-m', '--module', action='store', dest='module', default='', help='run a single module among '+' '.join(known_modules))
 (options, args) = parser.parse_args()
 
 modules = []
 
 def GetSlivers(plc):
-    data = plc.GetSlivers()
-    # net needs access to API for i2 nodes.
+    try: 
+        logger.log("Syncing w/ PLC")
+        data = plc.GetSlivers()
+    except: 
+        logger.log_exc()
+        #  XXX So some modules can at least boostrap.
+        data = {}
+    if (options.verbose):
+        logger.log_slivers(data)
+    # Set i2 ip list for nodes in I2 nodegroup.
+    try: net.GetSlivers(plc, data)
+    except: logger.log_exc()
+    #  All other callback modules
     for module in modules:
-        if module.__name__ == 'net':
-            module.GetSlivers(plc, data)
-        else:
+        try:        
             callback = getattr(module, 'GetSlivers')
             callback(data)
+        except: logger.log_exc()
 
 def run():
     try:
         if options.daemon: tools.daemon()
 
+        # set log level
+        if (options.verbose):
+            logger.set_level(logger.LOG_VERBOSE)
+
         # Load /etc/planetlab/plc_config
         config = Config(options.config)
 
@@ -55,7 +75,13 @@ def run():
             print "Warning while writing PID file:", err
 
         # Load and start modules
-        for module in ['net', 'proper', 'conf_files', 'sm', 'bwmon']:
+        if options.module:
+            assert options.module in known_modules
+            running_modules=[options.module]
+            logger.verbose('Running single module %s'%options.module)
+        else:
+            running_modules=known_modules
+        for module in running_modules:
             try:
                 m = __import__(module)
                 m.start(options, config)
@@ -70,16 +96,22 @@ def run():
             session = options.session
 
         # Initialize XML-RPC client
-        plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=options.period/2)
+        iperiod=int(options.period)
+        irandom=int(options.random)
+        plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=iperiod/2)
 
         while True:
-            try: GetSlivers(plc)
-            except: logger.log_exc()
-            time.sleep(options.period)
+        # Main NM Loop
+            logger.verbose('mainloop - nm:getSlivers - period=%d random=%d'%(iperiod,irandom))
+            GetSlivers(plc)
+            delay=iperiod + random.randrange(0,irandom)
+            logger.verbose('mainloop - sleeping for %d s'%delay)
+            time.sleep(delay)
     except: logger.log_exc()
 
 
 if __name__ == '__main__':
+    logger.log("Entering nm.py "+id)
     stacklim = 512*1024  # 0.5 MiB
     curlim = resource.getrlimit(resource.RLIMIT_STACK)[0]  # soft limit
     if curlim > stacklim: