various tricks and notes for smoother debugging

[nodemanager.git] / nm.py

diff --git a/nm.py b/nm.py
index 62d9ece..72fbd80 100755 (executable)
--- a/nm.py
+++ b/nm.py
@@ -1,5 +1,7 @@
 #!/usr/bin/python
-
+#
+# $Id$
+# $URL$
 #
 # Useful information can be found at https://svn.planet-lab.org/wiki/NodeManager
 #
@@ -33,13 +35,18 @@ known_modules=['net','conf_files', 'sm', 'bwmon']
 
 plugin_path = "/usr/share/NodeManager/plugins"
 
+default_period=600
+default_random=301
+
 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('-p', '--period', action='store', dest='period', default=default_period, 
+                  help='Polling interval (sec) - default %d'%default_period)
+parser.add_option('-r', '--random', action='store', dest='random', default=default_random, 
+                  help='Range for additional random polling interval (sec) -- default %d'%default_random)
 parser.add_option('-v', '--verbose', action='store_true', dest='verbose', default=False, help='more verbose log')
 parser.add_option('-P', '--path', action='store', dest='path', default=plugin_path, help='Path to plugins directory')
 
@@ -54,12 +61,14 @@ if os.path.exists(options.path):
 
 modules = []
 
-def GetSlivers(plc, config):
+def GetSlivers(config, plc):
     '''Run call backs defined in modules'''
     try: 
         logger.log("Syncing w/ PLC")
         data = plc.GetSlivers()
-        if (options.verbose): logger.log_slivers(data)
+        # used to be done only in verbose; very helpful though, and tedious to obtain,
+        # so let's dump this unconditionnally
+        logger.log_slivers(data)
         getPLCDefaults(data, config)
     except: 
         logger.log_exc()
@@ -70,7 +79,7 @@ def GetSlivers(plc, config):
     for module in modules:
         try:        
             callback = getattr(module, 'GetSlivers')
-            callback(plc, data, config)
+            callback(data, config, plc)
         except: logger.log_exc()
 
 
@@ -82,10 +91,19 @@ def getPLCDefaults(data, config):
         if slice['name'] == config.PLC_SLICE_PREFIX+"_default":
             attr_dict = {}
             for attr in slice.get('attributes'): attr_dict[attr['tagname']] = attr['value'] 
+            # GetSlivers exposes the result of GetSliceFamily() as an separate key in data
+            # It is safe to override the attributes with this, as this method has the right logic
+            try:
+                attr_dict['vref']=slice.get('GetSliceFamily')
+            except:
+                pass
             if len(attr_dict):
                 logger.verbose("Found default slice overrides.\n %s" % attr_dict)
                 config.OVERRIDES = attr_dict
-            return 
+                return
+    # NOTE: if an _default slice existed, it would have been found above and
+    #      the routine would return.  Thus, if we've gotten here, then no default
+    #      slice is bound to this node.
     if 'OVERRIDES' in dir(config): del config.OVERRIDES
 
 
@@ -134,10 +152,22 @@ def run():
         irandom=int(options.random)
         plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=iperiod/2)
 
+        #check auth
+        logger.log("Checking Auth.")
+        while plc.check_authentication() != True:
+            try:
+                plc.update_session()
+                logger.log("Authentication Failure.  Retrying")
+            except:
+                logger.log("Retry Failed.  Waiting")
+            time.sleep(iperiod)
+        logger.log("Authentication Succeeded!")
+
+
         while True:
         # Main NM Loop
             logger.verbose('mainloop - nm:getSlivers - period=%d random=%d'%(iperiod,irandom))
-            GetSlivers(plc, config)
+            GetSlivers(config, plc)
             delay=iperiod + random.randrange(0,irandom)
             logger.verbose('mainloop - sleeping for %d s'%delay)
             time.sleep(delay)
@@ -146,13 +176,6 @@ def run():
 
 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:
-        resource.setrlimit(resource.RLIMIT_STACK, (stacklim, stacklim))
-        # for some reason, doesn't take effect properly without the exec()
-        python = '/usr/bin/python'
-        os.execv(python, [python] + savedargv)
     run()
 else:
     # This is for debugging purposes.  Open a copy of Python and import nm