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 # NOTE: for OpenStack build, we don't run:
41 # - slivermanager, since Nova takes care of creating/deleting slivers
42 # - bwmon, because it's not clear how it should interact with Nova
43 core_modules=['net', 'conf_files']
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('-f', '--config', action='store', dest='config', default='/etc/planetlab/plc_config',
55 help='PLC configuration file')
56 parser.add_option('-k', '--session', action='store', dest='session', default='/etc/planetlab/session',
57 help='API session key (or file)')
58 parser.add_option('-p', '--period', action='store', dest='period', default=NodeManager.default_period,
59 help='Polling interval (sec) - default %d'%NodeManager.default_period)
60 parser.add_option('-r', '--random', action='store', dest='random', default=NodeManager.default_random,
61 help='Range for additional random polling interval (sec) -- default %d'%NodeManager.default_random)
62 parser.add_option('-v', '--verbose', action='store_true', dest='verbose', default=False,
63 help='more verbose log')
64 parser.add_option('-P', '--path', action='store', dest='path', default=NodeManager.PLUGIN_PATH,
65 help='Path to plugins directory')
67 # NOTE: BUG the 'help' for this parser.add_option() wont list plugins from the --path argument
68 parser.add_option('-m', '--module', action='store', dest='user_module', default='', help='run a single module')
69 (self.options, args) = parser.parse_args()
75 # determine the modules to be run
76 self.modules = NodeManager.core_modules
77 # Deal with plugins directory
78 if os.path.exists(self.options.path):
79 sys.path.append(self.options.path)
80 plugins = [ os.path.split(os.path.splitext(x)[0])[1]
81 for x in glob.glob( os.path.join(self.options.path,'*.py') )
82 if not x.endswith("/__init__.py")
84 self.modules += plugins
85 if self.options.user_module:
86 assert self.options.user_module in self.modules
87 self.modules=[self.options.user_module]
88 logger.verbose('nodemanager: Running single module %s'%self.options.user_module)
91 def GetSlivers(self, config, plc):
92 """Retrieves GetSlivers at PLC and triggers callbacks defined in modules/plugins"""
94 logger.log("nodemanager: Syncing w/ PLC")
95 # retrieve GetSlivers from PLC
96 data = plc.GetSlivers()
97 # use the magic 'default' slice to retrieve system-wide defaults
98 self.getPLCDefaults(data, config)
99 # tweak the 'vref' attribute from GetSliceFamily
100 self.setSliversVref (data)
101 # dump it too, so it can be retrieved later in case of comm. failure
102 self.dumpSlivers(data)
103 # log it for debug purposes, no matter what verbose is
104 logger.log_slivers(data)
105 logger.verbose("nodemanager: Sync w/ PLC done")
108 logger.log_exc("nodemanager: failed in GetSlivers")
109 # XXX So some modules can at least boostrap.
110 logger.log("nodemanager: Can't contact PLC to GetSlivers(). Continuing.")
112 # for modules that request it though the 'persistent_data' property
113 last_data=self.loadSlivers()
114 # Invoke GetSlivers() functions from the callback modules
115 for module in self.loaded_modules:
116 logger.verbose('nodemanager: triggering %s.GetSlivers'%module.__name__)
118 callback = getattr(module, 'GetSlivers')
120 if getattr(module,'persistent_data',False):
121 module_data=last_data
122 callback(data, config, plc)
124 logger.log_exc("nodemanager: GetSlivers failed to run callback for module %r"%module)
127 def getPLCDefaults(self, data, config):
129 Get PLC wide defaults from _default system slice. Adds them to config class.
131 for slice in data.get('slivers'):
132 if slice['name'] == config.PLC_SLICE_PREFIX+"_default":
134 for attr in slice.get('attributes'): attr_dict[attr['tagname']] = attr['value']
136 logger.verbose("nodemanager: Found default slice overrides.\n %s" % attr_dict)
137 config.OVERRIDES = attr_dict
139 # NOTE: if an _default slice existed, it would have been found above and
140 # the routine would return. Thus, if we've gotten here, then no default
141 # slice is bound to this node.
142 if 'OVERRIDES' in dir(config): del config.OVERRIDES
145 def setSliversVref (self, data):
147 Tweak the 'vref' attribute in all slivers based on the 'GetSliceFamily' key
149 # GetSlivers exposes the result of GetSliceFamily() as an separate key in data
150 # It is safe to override the attributes with this, as this method has the right logic
151 for sliver in data.get('slivers'):
153 slicefamily=sliver.get('GetSliceFamily')
154 for att in sliver['attributes']:
155 if att['tagname']=='vref':
156 att['value']=slicefamily
158 sliver['attributes'].append({ 'tagname':'vref','value':slicefamily})
160 logger.log_exc("nodemanager: Could not overwrite 'vref' attribute from 'GetSliceFamily'",name=sliver['name'])
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: tools.daemon()
187 if (self.options.verbose):
188 logger.set_level(logger.LOG_VERBOSE)
190 # Load /etc/planetlab/plc_config
191 config = Config(self.options.config)
194 other_pid = tools.pid_file()
195 if other_pid != None:
196 print """There might be another instance of the node manager running as pid %d.
197 If this is not the case, please remove the pid file %s. -- exiting""" % (other_pid, tools.PID_FILE)
200 print "Warning while writing PID file:", err
203 self.loaded_modules = []
204 for module in self.modules:
206 m = __import__(module)
207 logger.verbose("nodemanager: triggering %s.start"%m.__name__)
209 self.loaded_modules.append(m)
210 except ImportError, err:
211 logger.log_exc ("ERROR while loading module %s - skipping:" % module)
212 # if we fail to load any of these, it's really no need to go on any further
213 if module in NodeManager.core_modules:
214 logger.log("FATAL : failed to load core module %s"%module)
215 except AttributeError, err:
216 # triggered when module doesn't have a 'start' method
217 logger.log_exc ("ERROR while starting module %s - skipping:" % module)
218 # if we fail to load any of these, it's really no need to go on any further
219 if module in NodeManager.core_modules:
220 logger.log("FATAL : failed to start core module %s"%module)
222 # sort on priority (lower first)
223 def sort_module_priority (m1,m2):
224 return getattr(m1,'priority',NodeManager.default_priority) - getattr(m2,'priority',NodeManager.default_priority)
225 self.loaded_modules.sort(sort_module_priority)
227 logger.log('ordered modules:')
228 for module in self.loaded_modules:
229 logger.log ('%s: %s'%(getattr(module,'priority',NodeManager.default_priority),module.__name__))
231 # Load /etc/planetlab/session
232 if os.path.exists(self.options.session):
233 session = file(self.options.session).read().strip()
239 iperiod=int(self.options.period)
240 irandom=int(self.options.random)
242 # Initialize XML-RPC client
243 plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=iperiod/2)
246 logger.log("nodemanager: Checking Auth.")
247 while plc.check_authentication() != True:
250 logger.log("nodemanager: Authentication Failure. Retrying")
252 logger.log("nodemanager: Retry Failed. (%r); Waiting.."%e)
254 logger.log("nodemanager: Authentication Succeeded!")
258 # Main nodemanager Loop
260 logger.log('nodemanager: mainloop - calling GetSlivers - period=%d random=%d'%(iperiod,irandom))
261 self.GetSlivers(config, plc)
262 delay=iperiod + random.randrange(0,irandom)
264 work_duration=int(work_end-work_beg)
265 logger.log('nodemanager: mainloop has worked for %s s - sleeping for %d s'%(work_duration,delay))
267 except: logger.log_exc("nodemanager: failed in run")
270 logger.log("======================================== Entering nodemanager.py")
273 if __name__ == '__main__':
276 # This is for debugging purposes. Open a copy of Python and import nodemanager
277 tools.as_daemon_thread(run)