6 # Useful information can be found at https://svn.planet-lab.org/wiki/NodeManager
9 # Faiyaz Ahmed <faiyaza at cs dot princeton dot edu>
10 # Copyright (C) 2008 The Trustees of Princeton University
28 from config import Config
29 from plcapi import PLCAPI
35 PLUGIN_PATH = "/usr/share/NodeManager/plugins"
37 DB_FILE = "/var/lib/nodemanager/getslivers.pickle"
39 # the modules in this directory that need to be run
40 # NOTE: modules listed here will also be loaded in this order
41 # once loaded, they get re-ordered after their priority (lower comes first)
42 # for determining the runtime order
43 core_modules=['net', 'conf_files', 'slivermanager', 'bwmon']
51 parser = optparse.OptionParser()
52 parser.add_option('-d', '--daemon', action='store_true', dest='daemon', default=False,
53 help='run daemonized')
54 parser.add_option('-s', '--startup', action='store_true', dest='startup', default=False,
55 help='run all sliver startup scripts')
56 parser.add_option('-f', '--config', action='store', dest='config', default='/etc/planetlab/plc_config',
57 help='PLC configuration file')
58 parser.add_option('-k', '--session', action='store', dest='session', default='/etc/planetlab/session',
59 help='API session key (or file)')
60 parser.add_option('-p', '--period', action='store', dest='period', default=NodeManager.default_period,
61 help='Polling interval (sec) - default %d'%NodeManager.default_period)
62 parser.add_option('-r', '--random', action='store', dest='random', default=NodeManager.default_random,
63 help='Range for additional random polling interval (sec) -- default %d'%NodeManager.default_random)
64 parser.add_option('-v', '--verbose', action='store_true', dest='verbose', default=False,
65 help='more verbose log')
66 parser.add_option('-P', '--path', action='store', dest='path', default=NodeManager.PLUGIN_PATH,
67 help='Path to plugins directory')
69 # NOTE: BUG the 'help' for this parser.add_option() wont list plugins from the --path argument
70 parser.add_option('-m', '--module', action='store', dest='user_module', default='', help='run a single module')
71 (self.options, args) = parser.parse_args()
77 # determine the modules to be run
78 self.modules = NodeManager.core_modules
79 # Deal with plugins directory
80 if os.path.exists(self.options.path):
81 sys.path.append(self.options.path)
82 plugins = [ os.path.split(os.path.splitext(x)[0])[1] for x in glob.glob( os.path.join(self.options.path,'*.py') ) ]
83 self.modules += plugins
84 if self.options.user_module:
85 assert self.options.user_module in self.modules
86 self.modules=[self.options.user_module]
87 logger.verbose('nodemanager: Running single module %s'%self.options.user_module)
90 def GetSlivers(self, config, plc):
91 """Retrieves GetSlivers at PLC and triggers callbacks defined in modules/plugins"""
93 logger.log("nodemanager: Syncing w/ PLC")
94 # retrieve GetSlivers from PLC
95 data = plc.GetSlivers()
96 # use the magic 'default' slice to retrieve system-wide defaults
97 self.getPLCDefaults(data, config)
98 # tweak the 'vref' attribute from GetSliceFamily
99 self.setSliversVref (data)
100 # check leases and adjust the 'alive' field in slivers
101 self.adjustReservedSlivers (data)
102 # dump it too, so it can be retrieved later in case of comm. failure
103 self.dumpSlivers(data)
104 # log it for debug purposes, no matter what verbose is
105 logger.log_slivers(data)
106 logger.verbose("nodemanager: Sync w/ PLC done")
109 logger.log_exc("nodemanager: failed in GetSlivers")
110 # XXX So some modules can at least boostrap.
111 logger.log("nodemanager: Can't contact PLC to GetSlivers(). Continuing.")
113 # for modules that request it though the 'persistent_data' property
114 last_data=self.loadSlivers()
115 # adjust again with current time
116 self.adjustReservedSlivers (data)
117 # Invoke GetSlivers() functions from the callback modules
118 for module in self.loaded_modules:
119 logger.verbose('nodemanager: triggering %s.GetSlivers'%module.__name__)
121 callback = getattr(module, 'GetSlivers')
123 if getattr(module,'persistent_data',False):
124 module_data=last_data
125 callback(data, config, plc)
127 logger.log_exc("nodemanager: GetSlivers failed to run callback for module %r"%module)
130 def getPLCDefaults(self, data, config):
132 Get PLC wide defaults from _default system slice. Adds them to config class.
134 for slice in data.get('slivers'):
135 if slice['name'] == config.PLC_SLICE_PREFIX+"_default":
137 for attr in slice.get('attributes'): attr_dict[attr['tagname']] = attr['value']
139 logger.verbose("nodemanager: Found default slice overrides.\n %s" % attr_dict)
140 config.OVERRIDES = attr_dict
142 # NOTE: if an _default slice existed, it would have been found above and
143 # the routine would return. Thus, if we've gotten here, then no default
144 # slice is bound to this node.
145 if 'OVERRIDES' in dir(config): del config.OVERRIDES
148 def setSliversVref (self, data):
150 Tweak the 'vref' attribute in all slivers based on the 'GetSliceFamily' key
152 # GetSlivers exposes the result of GetSliceFamily() as an separate key in data
153 # It is safe to override the attributes with this, as this method has the right logic
154 for sliver in data.get('slivers'):
156 slicefamily=sliver.get('GetSliceFamily')
157 for att in sliver['attributes']:
158 if att['tagname']=='vref':
159 att['value']=slicefamily
161 sliver['attributes'].append({ 'tagname':'vref','value':slicefamily})
163 logger.log_exc("nodemanager: Could not overwrite 'vref' attribute from 'GetSliceFamily'",name=sliver['name'])
165 def adjustReservedSlivers (self, data):
167 On reservable nodes, tweak the 'alive' field to instruct cyclic loop
168 about what to do with slivers.
172 def dumpSlivers (self, slivers):
173 f = open(NodeManager.DB_FILE, "w")
174 logger.log ("nodemanager: saving successfully fetched GetSlivers in %s" % NodeManager.DB_FILE)
175 pickle.dump(slivers, f)
178 def loadSlivers (self):
180 f = open(NodeManager.DB_FILE, "r+")
181 logger.log("nodemanager: restoring latest known GetSlivers from %s" % NodeManager.DB_FILE)
182 slivers = pickle.load(f)
186 logger.log("Could not restore GetSlivers from %s" % NodeManager.DB_FILE)
191 if self.options.daemon: tools.daemon()
194 if (self.options.verbose):
195 logger.set_level(logger.LOG_VERBOSE)
197 # Load /etc/planetlab/plc_config
198 config = Config(self.options.config)
201 other_pid = tools.pid_file()
202 if other_pid != None:
203 print """There might be another instance of the node manager running as pid %d.
204 If this is not the case, please remove the pid file %s. -- exiting""" % (other_pid, tools.PID_FILE)
207 print "Warning while writing PID file:", err
210 self.loaded_modules = []
211 for module in self.modules:
213 m = __import__(module)
214 logger.verbose("nodemanager: triggering %s.start"%m.__name__)
215 m.start(self.options, config)
216 self.loaded_modules.append(m)
217 except ImportError, err:
218 print "Warning while loading module %s:" % module, err
220 # sort on priority (lower first)
221 def sort_module_priority (m1,m2):
222 return getattr(m1,'priority',NodeManager.default_priority) - getattr(m2,'priority',NodeManager.default_priority)
223 self.loaded_modules.sort(sort_module_priority)
225 logger.log('ordered modules:')
226 for module in self.loaded_modules:
227 logger.log ('%s: %s'%(getattr(module,'priority',NodeManager.default_priority),module.__name__))
229 # Load /etc/planetlab/session
230 if os.path.exists(self.options.session):
231 session = file(self.options.session).read().strip()
237 iperiod=int(self.options.period)
238 irandom=int(self.options.random)
240 # Initialize XML-RPC client
241 plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=iperiod/2)
244 logger.log("nodemanager: Checking Auth.")
245 while plc.check_authentication() != True:
248 logger.log("nodemanager: Authentication Failure. Retrying")
250 logger.log("nodemanager: Retry Failed. (%r); Waiting.."%e)
252 logger.log("nodemanager: Authentication Succeeded!")
256 # Main nodemanager Loop
257 logger.log('nodemanager: mainloop - calling GetSlivers - period=%d random=%d'%(iperiod,irandom))
258 self.GetSlivers(config, plc)
259 delay=iperiod + random.randrange(0,irandom)
260 logger.log('nodemanager: mainloop - sleeping for %d s'%delay)
262 except: logger.log_exc("nodemanager: failed in run")
265 logger.log("======================================== Entering nodemanager.py")
268 if __name__ == '__main__':
271 # This is for debugging purposes. Open a copy of Python and import nodemanager
272 tools.as_daemon_thread(run)