3 # Useful information can be found at https://svn.planet-lab.org/wiki/NodeManager
6 # Faiyaz Ahmed <faiyaza at cs dot princeton dot edu>
7 # Copyright (C) 2008 The Trustees of Princeton University
26 from config import Config
27 from plcapi import PLCAPI
32 PLUGIN_PATH = "/usr/share/NodeManager/plugins"
34 DB_FILE = "/var/lib/nodemanager/getslivers.pickle"
36 # the modules in this directory that need to be run
37 # NOTE: modules listed here will also be loaded in this order
38 # once loaded, they get re-ordered after their priority (lower comes first)
39 # for determining the runtime order
40 core_modules = ['net', 'conf_files', 'slivermanager', 'bwmon']
44 default_priority = 100
48 parser = optparse.OptionParser()
49 parser.add_option('-d', '--daemon', action='store_true', dest='daemon', default=False,
50 help='run daemonized')
51 parser.add_option('-f', '--config', action='store', dest='config', default='/etc/planetlab/plc_config',
52 help='PLC configuration file')
53 parser.add_option('-k', '--session', action='store', dest='session', default='/etc/planetlab/session',
54 help='API session key (or file)')
55 parser.add_option('-p', '--period', action='store', dest='period', default=NodeManager.default_period,
56 help='Polling interval (sec) - default %d'%NodeManager.default_period)
57 parser.add_option('-r', '--random', action='store', dest='random', default=NodeManager.default_random,
58 help='Range for additional random polling interval (sec) -- default %d'%NodeManager.default_random)
59 parser.add_option('-v', '--verbose', action='store_true', dest='verbose', default=False,
60 help='more verbose log')
61 parser.add_option('-P', '--path', action='store', dest='path', default=NodeManager.PLUGIN_PATH,
62 help='Path to plugins directory')
64 # NOTE: BUG the 'help' for this parser.add_option() wont list plugins from the --path argument
65 parser.add_option('-m', '--module', action='store', dest='user_module', default='', help='run a single module')
66 (self.options, args) = parser.parse_args()
72 # determine the modules to be run
73 self.modules = NodeManager.core_modules
74 # Deal with plugins directory
75 if os.path.exists(self.options.path):
76 sys.path.append(self.options.path)
77 plugins = [ os.path.split(os.path.splitext(x)[0])[1]
78 for x in glob.glob( os.path.join(self.options.path,'*.py') )
79 if not x.endswith("/__init__.py")
81 self.modules += plugins
82 if self.options.user_module:
83 assert self.options.user_module in self.modules
84 self.modules = [self.options.user_module]
85 logger.verbose('nodemanager: Running single module %s'%self.options.user_module)
88 def GetSlivers(self, config, plc):
89 """Retrieves GetSlivers at PLC and triggers callbacks defined in modules/plugins"""
91 logger.log("nodemanager: Syncing w/ PLC")
92 # retrieve GetSlivers from PLC
93 data = plc.GetSlivers()
94 # use the magic 'default' slice to retrieve system-wide defaults
95 self.getPLCDefaults(data, config)
96 # tweak the 'vref' attribute from GetSliceFamily
97 self.setSliversVref (data)
98 # dump it too, so it can be retrieved later in case of comm. failure
99 self.dumpSlivers(data)
100 # log it for debug purposes, no matter what verbose is
101 logger.log_slivers(data)
102 logger.verbose("nodemanager: Sync w/ PLC done")
105 logger.log_exc("nodemanager: failed in GetSlivers")
106 # XXX So some modules can at least boostrap.
107 logger.log("nodemanager: Can't contact PLC to GetSlivers(). Continuing.")
109 # for modules that request it though the 'persistent_data' property
110 last_data = self.loadSlivers()
111 # Invoke GetSlivers() functions from the callback modules
112 for module in self.loaded_modules:
113 logger.verbose('nodemanager: triggering %s.GetSlivers'%module.__name__)
115 callback = getattr(module, 'GetSlivers')
117 if getattr(module,'persistent_data',False):
118 module_data = last_data
119 callback(data, config, plc)
120 except SystemExit as e:
123 logger.log_exc("nodemanager: GetSlivers failed to run callback for module %r"%module)
126 def getPLCDefaults(self, data, config):
128 Get PLC wide defaults from _default system slice. Adds them to config class.
130 for slice in data.get('slivers'):
131 if slice['name'] == config.PLC_SLICE_PREFIX + "_default":
133 for attr in slice.get('attributes'): attr_dict[attr['tagname']] = attr['value']
135 logger.verbose("nodemanager: Found default slice overrides.\n %s" % attr_dict)
136 config.OVERRIDES = attr_dict
138 # NOTE: if an _default slice existed, it would have been found above and
139 # the routine would return. Thus, if we've gotten here, then no default
140 # slice is bound to this node.
141 if 'OVERRIDES' in dir(config): del config.OVERRIDES
144 def setSliversVref (self, data):
146 Tweak the 'vref' attribute in all slivers based on the 'GetSliceFamily' key
148 # GetSlivers exposes the result of GetSliceFamily() as an separate key in data
149 # It is safe to override the attributes with this, as this method has the right logic
150 for sliver in data.get('slivers'):
152 slicefamily = sliver.get('GetSliceFamily')
153 for att in sliver['attributes']:
154 if att['tagname'] == 'vref':
155 att['value'] = slicefamily
157 sliver['attributes'].append({ 'tagname':'vref','value':slicefamily})
159 logger.log_exc("nodemanager: Could not overwrite 'vref' attribute from 'GetSliceFamily'",
162 def dumpSlivers (self, slivers):
163 f = open(NodeManager.DB_FILE, "w")
164 logger.log ("nodemanager: saving successfully fetched GetSlivers in %s" % NodeManager.DB_FILE)
165 pickle.dump(slivers, f)
168 def loadSlivers (self):
170 f = open(NodeManager.DB_FILE, "r+")
171 logger.log("nodemanager: restoring latest known GetSlivers from %s" % NodeManager.DB_FILE)
172 slivers = pickle.load(f)
176 logger.log("Could not restore GetSlivers from %s" % NodeManager.DB_FILE)
180 # make sure to create /etc/planetlab/virt so others can read that
181 # used e.g. in vsys-scripts's sliceip
182 tools.get_node_virt()
184 if self.options.daemon:
188 if (self.options.verbose):
189 logger.set_level(logger.LOG_VERBOSE)
192 # Load /etc/planetlab/plc_config
193 config = Config(self.options.config)
196 other_pid = tools.pid_file()
197 if other_pid != None:
198 print """There might be another instance of the node manager running as pid %d.
199 If this is not the case, please remove the pid file %s. -- exiting""" % (other_pid, tools.PID_FILE)
202 print "Warning while writing PID file:", err
205 self.loaded_modules = []
206 for module in self.modules:
208 m = __import__(module)
209 logger.verbose("nodemanager: triggering %s.start" % m.__name__)
211 except: logger.log("WARNING: module %s did not start")
212 self.loaded_modules.append(m)
214 if module not in NodeManager.core_modules:
215 logger.log_exc ("ERROR while loading module %s - skipped" % module)
217 logger.log("FATAL : failed to start core module %s"%module)
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 work_beg = time.time()
258 logger.log('nodemanager: mainloop - calling GetSlivers - period=%d random=%d'%(iperiod,irandom))
259 self.GetSlivers(config, plc)
260 delay = iperiod + random.randrange(0,irandom)
261 work_end = time.time()
262 work_duration = int(work_end-work_beg)
263 logger.log('nodemanager: mainloop has worked for %s s - sleeping for %d s'%(work_duration,delay))
268 logger.log_exc("nodemanager: failed in run")
271 logger.log("======================================== Entering nodemanager.py")
274 if __name__ == '__main__':
277 # This is for debugging purposes. Open a copy of Python and import nodemanager
278 tools.as_daemon_thread(run)